Luminous Landscape Forum

Site & Board Matters => About This Site => Topic started by: BJL on January 23, 2007, 11:18:24 am

Title: f-stop limits for full sensor resolution
Post by: BJL on January 23, 2007, 11:18:24 am
Nathan Myhrvold's  addition (http://luminous-landscape.com/essays/Equivalent-Lenses.shtml#another) to the discussion started by Charles Johnson is a useful additional perspective: not really digital specific, but just about what the aperture limits are if one wants to get the full resolution that one's film or sensor is capable of.

I have a disagreement on the actual numbers though, at least for the case of sensors using Bayer CFA's and interpolation, because that process lowers the resolution of the sensor output beyond "green pixel diagonal size", and thus relaxes the diffraction spot size limit a bit.

For example observations of several users of the Nikon D2X, with 5.5 micron pixel spacing, say that difffraction starts to limit resolution at somewhere between f/8 and f/11. Thom Hogan is the author of the f/11 figure, which gives it some credibility. Myhrvold's calculation instead gives about f/5.8.

So I propose a rule of thumb (dependent on how Bayer interpolation is done, details of anti-aliasing filters and such) that diffraction starts to cut into the resolution that a Bayer CFA sensor is capable of somewhere around twice the pixel spacing in microns, or as much as one stop under. That is, I would modify  "Myhrvold's formula" to

max f-stop = P x 1.4 to P x 2 (instead of P X 1.054).

That makes quite a difference in "effective useful pixel counts" at a given high f-stop, by a factor of two to four.

For example, I suggest that full sensor resolution limits one to maximum f-stops of about
- f/8-f/11 for the D2X or the 400D and similar for other new 10MP SLR's
- f/11-f/14 for the 7.2 microns pixel spacing of the 1DsMkII, and for the 6.8 microns of the E-1 (for which Olympus recommends an f/11 limit)
- f/7-f/9 for the current smallest DSLR pixels, 4.75 microns in the Olympus E-400
- f/2.8-f/4 for the current digicam sensors with pixel pitch about 2 microns or a bit under.

Has anyone have sharpness tests at various f-stops on the 1DsMkII, G7 or other cameras, to allow a test of this idea and give us a better estimate of the best constant in Myhrvold's formula?
Title: f-stop limits for full sensor resolution
Post by: madmanchan on January 23, 2007, 12:04:58 pm
I don't know the answers to your questions, but I have a couple of additional questions of my own that I hope the authors (or someone else with sufficient background) can answer:

- Are the diffraction formulas used in the articles based on the standard "single slit" diffraction model or are they based on the actual kind of diffraction that occurs in camera lenses (e.g., diffraction around the aperture blades)? Does the distinction matter? Intuitively, it seems to me that a linear row of slits is quite different from an octagonal pattern of 8 aperture blades (some blades are somewhat rounded now, too). Does this affect the formulas and/or the results in a meaningful way?

- A lot of sensors (like the ones in DSLRs) have a layer of microlenses in front of them. These are often described as focusing the incoming light more efficiently. Obviously, this must mean the direction of the optical path changes slightly. Is this purely a geometrical result, or does it affect the wave properties of light as well? Do microlenses interact with diffraction? How so?

Eric
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on January 23, 2007, 01:01:46 pm
Quote
For example, I suggest that full sensor resolution limits one to maximum f-stops of about
- f/8-f/11 for the D2X or the 400D and similar for other new 10MP SLR's
- f/11-f/14 for the 7.2 microns pixel spacing of the 1DsMkII, and for the 6.8 microns of the E-1 (for which Olympus recommends an f/11 limit)
- f/7-f/9 for the current smallest DSLR pixels, 4.75 microns in the Olympus E-400
- f/2.8-f/4 for the current digicam sensors with pixel pitch about 2 microns or a bit under.

Has anyone have sharpness tests at various f-stops on the 1DsMkII, G7 or other cameras, to allow a test of this idea and give us a better estimate of the best constant in Myhrvold's formula?
[a href=\"index.php?act=findpost&pid=97174\"][{POST_SNAPBACK}][/a]

I just did a quick test.  I filled a soda bottle cap with a layer of glossy, tiny glitter beads, and took shots at f/2.8 through f/22 (in one-stop increments) with my Tamron 90mm macro (probably the most optically perfect lens I own), using flash and 1/250s @ ISO 200 on my Canon 20D.  f/22 is noticeably duller than 16, 16 slightly duller than 11, and 8 is the sharpest of the lot, with 5.6 slightly sharper than 11.  Sweet spot is probably about 7.1 - 8.

I opened all 7 RAW in tandem in ACR, turned off all auto settings, adjusted exposure individually (varied by over a half stop with no changes in frame!), and cropped as a group.

The f/22, though a bit duller than the f/11, tightens up very quickly without significant artifacting with USM, to look like the f/11 with more DOF.
Title: f-stop limits for full sensor resolution
Post by: gkroeger on January 23, 2007, 01:39:48 pm
Quote
- Are the diffraction formulas used in the articles based on the standard "single slit" diffraction model or are they based on the actual kind of diffraction that occurs in camera lenses (e.g., diffraction around the aperture blades)? Does the distinction matter? Intuitively, it seems to me that a linear row of slits is quite different from an octagonal pattern of 8 aperture blades (some blades are somewhat rounded now, too). Does this affect the formulas and/or the results in a meaningful way?
Eric
[a href=\"index.php?act=findpost&pid=97184\"][{POST_SNAPBACK}][/a]

Eric:

Most of the calculations use an Airy disc formula for point sources... this is pretty good for a camera lens iris opening, certainly within 25% or so.

I also agree with the previous posts that Nathan's article, although very useful and well written, doesn't account for the effect of the anti-aliasing filters and demosaicing of the Bayer data.  I would agree with BJL that we get about 1 more stop than Nathan's formula predicts.  

Lloyd Chambers has done extensive DOF and diffraction testing of D2X and 1DsMkII and found about 1 stop effective difference.  One was f/11 but I don't remember which, so that's not much help, but BNLs numbers look about right.

Glenn
Title: f-stop limits for full sensor resolution
Post by: EricV on January 23, 2007, 02:06:47 pm
The tone of the discussion about pixel size and f/stop limits for full resolution seems to suggest that small pixels are somehow worse than large pixels, because they become diffraction limited at larger apertures.  This is actually an advantage of small pixels.  Rather than saying that small pixels only support large apertures, it seems less misleading to say that large pixels do not support large apertures.  

All other factors being equal (sensor size in particular), a sensor with small pixels will have better resolution than a sensor with large pixels.  At worst, this extra sensor resolution will be wasted if the optical resolution does not support it.  A sensor with small pixels requires better optical resolution to attain its full potential because its full potential exceeds that of a sensor with large pixels.

A crucial factor when considering resolution as a function of pixel size is the magnification needed to go from sensor to print.  This depends on total sensor size, not pixel size or pixel count.  This is the real reason cameras with small pixels generally have worse "resolution" than cameras with large pixels -- the total sensor size is generally smaller, requiring higher print magnification.
Title: f-stop limits for full sensor resolution
Post by: alainbriot on January 23, 2007, 02:37:03 pm
I think the essay has valuable information. Regarding landscape work,  it helps me decide the best f-stop to use when shooting at infinity or shooting relatively flat subjects.  

However, when maximum depth of field is a requirement, such as with a foreground-background wide angle composition, stopping the lens down is necessary.  

This brings an important question: what size sensor/pixel combination is required to be able to close the lens down say to f16 and not lose resolution?
Title: f-stop limits for full sensor resolution
Post by: BJL on January 23, 2007, 04:06:15 pm
Quote
The tone of the discussion about pixel size and f/stop limits for full resolution seems to suggest that small pixels are somehow worse than large pixels, because they become diffraction limited at larger apertures.  This is actually an advantage of small pixels.
[a href=\"index.php?act=findpost&pid=97197\"][{POST_SNAPBACK}][/a]
I fall right in the middle on this one: the trade-offs of diffraction and OOF effects on image sharpness are the same with different pixel sizes and formats, with the f-stop needed changing roughly in proportion to pixel size.

I prefer to think in terms of resolution rather than pixel size. The smallest aperture (and highest f-stop) that avoids diffraction limiting the resolution that the sensor is capable of is proportional to sensor resolution in lp/mm (roughly, pixel density). If for example reducing pixel size doubles resolution in lp/mm, the maximum f-stop is halved in order to half the size of the diffraction spots.
Meanwhile, half the focal length is needed to get an equally detailed image, to be combined with twice the degree of enlargement. That halved focal length and halved f-stop makes each circle of confusion on the sensor half the diameter, just as the diffraction spot diameter is halved.
With twice the degree of enlargement one gets an equally detailed image of the same size with the diffraction spots and circles of confusion on the print also the same size.

More generally, with equal aperture size (f-stop adjusted in proportion to focal length) and enlargement adjusted according to focal length to get equal print size, the diffraction effects and OOF effects on the print are the same.
What changes is that
- with larger pixels, equal shutter speed requires higher exposure index (ISO speed)
- if pixels and sensors are too small, this [equal aperture size] might require an aperture ratio so low that lens aberrations are a problem. The diffraction limited aperture ratios of SLR pixels are not low enough for that problem, but with recent digicam sensor pixel sizes, this might be an issue.
Title: f-stop limits for full sensor resolution
Post by: Ray on January 23, 2007, 07:39:19 pm
Quote
A crucial factor when considering resolution as a function of pixel size is the magnification needed to go from sensor to print.  This depends on total sensor size, not pixel size or pixel count.  This is the real reason cameras with small pixels generally have worse "resolution" than cameras with large pixels -- the total sensor size is generally smaller, requiring higher print magnification.
[a href=\"index.php?act=findpost&pid=97197\"][{POST_SNAPBACK}][/a]

I don't think this is correct. In the old days of film, the enlargement was a direct physical enlargement of a piece of film. When we make an enlargements from a digital image, the size of the sensor is not enlarged. Everything has to do with pixel count and pixel size; pixel count in relation to degree of enlargement; pixel size in relation to resolution.
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 23, 2007, 07:56:57 pm
Quote
I fall right in the middle on this one: the trade-offs of diffraction and OOF effects on image sharpness are the same with different pixel sizes and formats, with the f-stop needed changing roughly in proportion to pixel size.

I prefer to think in terms of resolution rather than pixel size. The smallest aperture (and highest f-stop) that avoids diffraction limiting the resolution that the sensor is capable of is proportional to sensor resolution in lp/mm (roughly, pixel density).
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=97217\")

Rather than theorizing about these matters, it is often better to perform tests with your own equipment. I have found Imitest a good tool for this purpose. Here are results for my Nikon D200 and the 50 mm f/1.8 expressed as MTF 50% resolution figures with and without sharpening (uncorrected and corrected).

[attachment=1617:attachment]

[a href=\"http://www.photozone.de/8Reviews/index.html]Photozone[/url] gives Imitest results for many lenses. One must be aware that you are testing the resolution of the camera system and not merely the lens. As is evident in my test, optimum resolution is around f/5.6 with significant deterioration beyond f/11.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on January 23, 2007, 09:29:12 pm
Quote
As is evident in my test, optimum resolution is around f/5.6 with significant deterioration beyond f/11.
[a href=\"index.php?act=findpost&pid=97246\"][{POST_SNAPBACK}][/a]


These results more or less confirm the general consensus that f11 is the limit for stopping down with cropped 35mm format and and f16 the limit for full frame 35mm. Standard 50mm lenses are usually very sharp at f5.6 - f8. With the average zoom lens, the differences between f5.6 and f11 would be less.

Perhaps more relevant than using Imatest is to do real world comparisons with specific lenses. I've done extensive 'real world' comparisons of my 5D with 24-105 zoom and there's no significant resolution difference between f8 and f16, at the plane of focus. I haven't however done similar comparisons using my sharpest lens, the Canon 50/1.4.
Title: f-stop limits for full sensor resolution
Post by: xtoph on January 24, 2007, 04:38:37 am
the question of diffraction limits for max rez on a digital sensor is interesting. my results do not conform to the model that myhrvold proposes. i am not an optical theorist; i am just going from empirical results. f/11, or even 16, on my 5d with the 100macro is not worse than f/9 in terms of resolution information. furthermore, i have used f/22 even on my old 20d, which has a higher pixel density, and resolution did not drop down anywhere near as far as myhrvold's model predicts. some of this may be because of the limits of lenses, but that doesn't really work; otherwise we wouldn't be able to get the kind of peak resolving power out of these lenses and sensors that we do at any aperture.

furthermore, we know that the actual photosite is not as large as a straight area density calculation would show; each photosite sits within that area, from what i've seen on a relatively small part of it. even with the latest microlenses, the actual area used for the calculations in this model ought to be significantly smaller... which would mean that diffraction limits on the dsII would kick in even sooner, if this model is correct. and this we know does not happen (innumerable tests confirm that resolution is often at a peak around f/8, which given the adjusted-for-actual-photosite-size calculations ought to be beyond the diffraction limit for this camera). does that sound incorrect to people? am i misinterpreting the model?

perhaps this consideration of actual photosite size drops out, since the model looks at the array rather than each individual photosite. but this is the part of the model i am least clear on; it is not evident to me that the bayer array should enter into the calculation in this way. photosites are not pixels; each pixel of the image, of course, contains information interpollated from many sites. i would guess that this is where the model is breaking down, around misunderstanding of how the bayer array affects the optical construction of the final image.

as i said, i have no particular expertise in optics, but what the model predicts doesn't match my real-world results, and from the other comments and tests posted here, don't match others' either. yes, diffraction becomes a consideration around f/11 in many cases, but we knew that. i get resolution at f/22 that wouldn't be representable in a 2mp image, and furthermore a very high proportion of the diffraction-related blur goes away with application of usm. as a guide to how to get the best rez from our sensors, i think that this article fails, or at most comes close for the wrong reasons (we know our lenses tend to do best at f/8 anyway).

ps i just shot a test sequence to confirm that i wasn't misremembering anything i stated above--i can say with absolute assurance that the notion that shooting at f/22 is like getting 2mp out of your sensor is rubbish. i got lines resolved clearly at barely above a pixel apart at f/22 without even trying--reducing size to 2mp doesn't come close to enough resolution to separate such lines. mr myhrvold, if that's what your model predicts, you really ought to do some empirical tests to check your model. (while you're at it, change "half of the pixel are sensitive only to green light, while the other half are split 25% to red and 25% to blue" to read something like "half of the pixels are sensitive only to green light, while the other half are divided evenly between red and blue".)
Title: f-stop limits for full sensor resolution
Post by: Ray on January 24, 2007, 08:12:08 am
Quote
furthermore, we know that the actual photosite is not as large as a straight area density calculation would show; each photosite sits within that area, from what i've seen on a relatively small part of it. even with the latest microlenses, the actual area used for the calculations in this model ought to be significantly smaller... which would mean that diffraction limits on the dsII would kick in even sooner, if this model is correct. [a href=\"index.php?act=findpost&pid=97278\"][{POST_SNAPBACK}][/a]

You might have a point there, but I doubt it. Is there a distinction to be made between pixel pitch and pixel size (or microlens size) regarding DoF? Canon have narrowed the gaps between microlenses in their new 10mp 400D so that each microlens is hardly smaller (perhaps no smaller, they don't say) than a 20D microlens. Does this mean that the same f stop (as the 20D) applies for maximum DoF consistent with good resolution? I doubt it.

Of course, the problem is that such small increases in pixel count on a sensor of the same size don't count for much. We're all engaging in rather extreme pixel peeping here.
Title: f-stop limits for full sensor resolution
Post by: Gordon Buck on January 24, 2007, 11:28:40 am
I was taught that most devices work "best" near the mid-point of the intended range of application and this generality has been very useful.
Title: f-stop limits for full sensor resolution
Post by: BJL on January 24, 2007, 11:32:54 am
Thanks for the data, and I agree about testing, but now see complications from the difference between essentially monochrome data and color data, and the extent to which sharpening can unravel the effects of Bayer interpolation.

I imagine that a sharpening algorithm that concentrates on luminosity using green information as the main measure of luminosity could more or less reproduce the resolution given by the green pixels, at least with near monochrome subjects, leading to more or less the Myhrvold threshold of "f-stop equal to pixel spacing in microns".

Next, to see diffraction effects on other colors, perhaps we need some of those test results that Foveon X3 sensor fans like: red-blue test patterns! Or at least sharpness tests on subjects with colors towards magenta (away from green).

Maybe the synthesis of observations so far is that diffraction effects start being measurable or even visibly noticeable with suitable subject matter once one passes the Myhrvold threshold (f/6 for the D200, close enough to your f/5.6 observation), with significant deterioration typically setting in beyond the "Thom Hogan" threshold of two stops beyond that (f/12 for the D200, close enough to your f/11 observation).
Title: f-stop limits for full sensor resolution
Post by: EricV on January 24, 2007, 02:58:52 pm
Quote
In the old days of film, the enlargement was a direct physical enlargement of a piece of film. When we make an enlargements from a digital image, the size of the sensor is not enlarged.
[a href=\"index.php?act=findpost&pid=97242\"][{POST_SNAPBACK}][/a]

Magnification has the same physical meaning for a digital sensor as it has for film: (print size) / (size of image on sensor).  If I make a 16" print from a 1" image, every feature on the image gets magnified by a factor of 16.  Does it really matter whether the image is recorded on a sensor composed of pixels or film grains?

Of course if you do not want to mention sensor size, you are free to substitute (pixel count) x (pixel size).  Both have real physical dimensions.

If I take a picture of a 10m tall tree from a distance of 50m, with a lens of focal length 50mm, stopped down to f/22 (diffraction blur 30um), on a sensor with 2000 pixels of size 5um, then make a print where the tree is 16" high, what is the diffraction blur size on the print?  Hint: calculate the magnification factor and ignore the irrelevant sensor parameters.
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 24, 2007, 03:13:55 pm
Quote
These results more or less confirm the general consensus that f11 is the limit for stopping down with cropped 35mm format and and f16 the limit for full frame 35mm. Standard 50mm lenses are usually very sharp at f5.6 - f8. With the average zoom lens, the differences between f5.6 and f11 would be less.

Perhaps more relevant than using Imatest is to do real world comparisons with specific lenses. I've done extensive 'real world' comparisons of my 5D with 24-105 zoom and there's no significant resolution difference between f8 and f16, at the plane of focus. I haven't however done similar comparisons using my sharpest lens, the Canon 50/1.4.
[a href=\"index.php?act=findpost&pid=97247\"][{POST_SNAPBACK}][/a]


Ray,

I agree with you entirely, but establishing the performance of your 24-105 mm would require quite a lot of testing and effort and the results are somewhat subjective. Since perceived image sharpness correlates well with the MTF at 50% contrast, one can get quite a bit of information in a couple of hours with Imitest.

On the otherhand, taking test shots of a high contrast USAF chart would be less likely to give useful information about how the lens performs in actual photographic tests.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on January 24, 2007, 06:18:25 pm
Quote
Magnification has the same physical meaning for a digital sensor as it has for film: (print size) / (size of image on sensor).  If I make a 16" print from a 1" image, every feature on the image gets magnified by a factor of 16.  Does it really matter whether the image is recorded on a sensor composed of pixels or film grains?

[a href=\"index.php?act=findpost&pid=97347\"][{POST_SNAPBACK}][/a]

Yes, it does matter. In the days of film, a particular type of fine grained film was not restricted to a particular format. You could use Provia F100 in 4x5" sheets or 35mm rolls. When people talked about the advantages of large format over small format, the enlargement factor, directly in relation to the size of the film, was the most significant factor. If we'd had a situation where Kodak Royal Gold 25 was only available for 35mm and Fuji ISO 800 film only available for 4x5" format, the resolution advantages of the larger format would have been either slight or non-existant.

Of course you are right in the sense, if you know any 2 out of the 3 factors, sensor size, pixel count and pixel pitch, you can work out the 3rd factor.
Title: f-stop limits for full sensor resolution
Post by: Ray on January 24, 2007, 07:04:41 pm
Quote
Ray,

I agree with you entirely, but establishing the performance of your 24-105 mm would require quite a lot of testing and effort and the results are somewhat subjective. Since perceived image sharpness correlates well with the MTF at 50% contrast, one can get quite a bit of information in a couple of hours with Imitest.

On the otherhand, taking test shots of a high contrast USAF chart would be less likely to give useful information about how the lens performs in actual photographic tests.

Bill
[a href=\"index.php?act=findpost&pid=97353\"][{POST_SNAPBACK}][/a]

As with many things in photography, the relevance of the technical results (whether from shooting high contrast test charts or using programs like Imatest) has to be assessed in the field and in practice.

The following images at f11 and f16 were taken in the field, partly with the conscious purpose of examining the results back home for DoF and resolution trade-off. I think I've posted similar images before. The conclusion is clear. The Canon 24-105 zoom at 85mm shows no resolution advantage at f11 in any part of the image, compared with the f16 shot, but the f16 shot is significantly sharper at the extremes of the DoF range.

The test might not be without flaws. I didn't use a tripod and the camera is tilted down very slightly in the f16 shot (compared with the f11 shot), so I can't compare the bottom centre edge, which I believe is also sharper in the f16 shot.

However, I did use the same shutter speed of 1/100th for each shot with IS turned on, a shutter speed which I think is more than sufficient to freeze camera shake at this focal length using IS. Both images were converted in RSP with exactly the same settings, including 'detail extraction' of +30 and sharpening of -10.

If you see any flaws or inconsistencies in the results, let me know   .

[attachment=1622:attachment]   [attachment=1623:attachment]   [attachment=1624:attachment]

[attachment=1625:attachment]   [attachment=1626:attachment]
Title: f-stop limits for full sensor resolution
Post by: Kirk Gittings on January 24, 2007, 07:05:48 pm
I completely agree with Ray's point on film. All film of a particular type and ASA were cut from huge "master rolls" with no difference of film quality between film formats. I.E. there was always an advantage with larger film formats to make prints of a given size. With digital it is important to compare apples with apples as with film.
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 24, 2007, 09:36:22 pm
Quote
The following images at f11 and f16 were taken in the field, partly with the conscious purpose of examining the results back home for DoF and resolution trade-off. I think I've posted similar images before. The conclusion is clear. The Canon 24-105 zoom at 85mm shows no resolution advantage at f11 in any part of the image, compared with the f16 shot, but the f16 shot is significantly sharper at the extremes of the DoF range.

The test might not be without flaws. I didn't use a tripod and the camera is tilted down very slightly in the f16 shot (compared with the f11 shot), so I can't compare the bottom centre edge, which I believe is also sharper in the f16 shot.

However, I did use the same shutter speed of 1/100th for each shot with IS turned on, a shutter speed which I think is more than sufficient to freeze camera shake at this focal length using IS. Both images were converted in RSP with exactly the same settings, including 'detail extraction' of +30 and sharpening of -10.

If you see any flaws or inconsistencies in the results, let me know   .

[a href=\"index.php?act=findpost&pid=97387\"][{POST_SNAPBACK}][/a]

Ray,

I agree with your conclusions regarding better depth of field at f/16, but with only two shots I'm not sure that the results are statistically significant. If you take a series of hand held shots with the same shooting parameters, some will be sharper than others.

At f/16 the diffraction spot is 20.7 microns for green light, well over the 2x the pixel spacing of the 5D (8.2 microns), and the system is definitely diffraction limited at f/16.

One way to test for camera shake is to photograph a point source and observe the pattern at 200% or more. Here is a test with Nikons VR 70-200 f/2.8 lens at 1/100 sec at f/5.6 with VR on the right and no VR on the left. It does work as you say. It would be best to examine the green channel of the raw file, but that is for another day. (this shot is of my neighbor's Christmas lights at about 100 meters).

[attachment=1631:attachment]

Bill
Title: f-stop limits for full sensor resolution
Post by: xtoph on January 25, 2007, 02:06:54 am
i figured i might as well supply the quick tests i did to confirm what i know from real-world photographs. here are 4 representative crops (from a longer series, on a tripod), including: full resolution shot at f/8 (f/11 is slightly better, but f/8 is closer to the theoretical limit myhrvold proposes for the 5d, so we will use that as a point of comparison); full res shot at f/22; the same f/22 shot with some sharpening applied; and finally, the f/8 shot resized to 2mp for the full frame (f/8 to give the best-case interpretation to his "2mp" claim).

you can see that at f/22 there is some loss of very fine detail (especially in the tailfeathers) just as we would expect from standard calculations for diffraction limits on film. sharpening helps a lot with much of the detail; since we're looking at 100% crops, we can conclude that sharpening will make up for most softening that you might actually see in a print, even a very, very large print. finally, we can see that myhrvold's claim that f/22 on full digital frame is equivalent to 2mp of resolution is absolutely wrong.

[from myhrvold's article "Now, I don’t think anybody would be very excited about turning their EOS 1Ds Mark II, or Canon 5D or other full frame camera into a 2 megapixel camera. It sounds pretty drastic, but that is exactly what you do when you stop down to f/22 – the diffraction limit imposes this condition. If you shoot with a full frame 24 x 36 sensor at f/22 you are throwing away a lot of resolution. There is no getting around this – it is fundamental in the physics of light."]

there is no getting around this--either the model is wrong (it's not even close), or we have in fact gotten around the "laws" of physics (which, where light is concerned, seem to be rather flexible anyway. which is how we got into this mess in the first place...)

so, conclusion: go ahead and use whatever aperture you need to obtain the depth of field or other results you want. for my 100mm macro, i will use it at f/11 whenever i can get away with it on subjects like insects, since i know that that is the peak of performance for the lens, gives adequate dof, and is occasionally obtainable with the lighting i've got. and lastly, do your own plausibility tests on authoritative info posted on the internet.

[attachment=1633:attachment]
[attachment=1634:attachment]
[attachment=1635:attachment]
[attachment=1636:attachment]
Title: f-stop limits for full sensor resolution
Post by: Herkko on January 25, 2007, 07:37:48 am
Quote
[from myhrvold's article "Now, I don’t think anybody would be very excited about turning their EOS 1Ds Mark II, or Canon 5D or other full frame camera into a 2 megapixel camera. It sounds pretty drastic, but that is exactly what you do when you stop down to f/22 – the diffraction limit imposes this condition. If you shoot with a full frame 24 x 36 sensor at f/22 you are throwing away a lot of resolution. There is no getting around this – it is fundamental in the physics of light."]

Another quote from Myhrvold's article:
["While Johnson’s treatment works for a fixed size print, the converse is to consider how to get the maximum image quality – i.e. the biggest and best print that you can possibly get. From this standpoint you don’t ask about the final print size, instead you ask what is the best quality that my camera can possibly render."]

Then we should propably define the maximum image quality? For me it's much more depending on chosen aperture for visual impact than optimal resolution on in-focus -area (which can be next to nothing at f8). If I want to deliver a visual message that requires f32, then I either:
- stop down to f32
- where possible use lens movements
- where possible stitch pictures at post processing
.. rather than settle for larger aperture and very different looking picture.

As most of timed photographers already know: the core idea of picture peaks  many times at large or very small apertures. f8 can be excellent choice sometimes, but only sometimes.

If I would be photographing flat test charts as my main hobby, the situation could be different. Resolution: for a low key still life picture with FF and f22 you can spot out dust specks that are hardly visible on bare eye. If someone requires even more resolution and is shy for diffraction, then there are of course digital backs and serious lens movements to be considered in  
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 25, 2007, 07:50:17 am
Quote
i figured i might as well supply the quick tests i did to confirm what i know from real-world photographs. here are 4 representative crops (from a longer series, on a tripod), including: full resolution shot at f/8 (f/11 is slightly better, but f/8 is closer to the theoretical limit myhrvold proposes for the 5d, so we will use that as a point of comparison); full res shot at f/22; the same f/22 shot with some sharpening applied; and finally, the f/8 shot resized to 2mp for the full frame (f/8 to give the best-case interpretation to his "2mp" claim).

you can see that at f/22 there is some loss of very fine detail (especially in the tailfeathers) just as we would expect from standard calculations for diffraction limits on film. sharpening helps a lot with much of the detail; since we're looking at 100% crops, we can conclude that sharpening will make up for most softening that you might actually see in a print, even a very, very large print. finally, we can see that myhrvold's claim that f/22 on full digital frame is equivalent to 2mp of resolution is absolutely wrong.


[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=97444\")

There is another way to look at this situation using MTF figures for various apertures as shown in the table on [a href=\"http://www.clarkvision.com/imagedetail/scandetail.html#diffraction]Roger Clark[/url]'s web site. These figures are for green light.

The 5D sensor is 23.9 mm in height and has 2912 pixels in this dimension. The Nyquist frequency is 2912/23.9/2 or 49 lp/mm. A Bayer sensor can resolve to only about 80% of Nyquist, but we will ignore this limitation for now. Frequencies above Nyquist will be removed by the low pass filter and can not be resolved by the digital sensor and result in aliasing, which is why the low pass filter is there. Film has a more gradual roll off and often resolves higher frequencies with low contrast.

When one mentions resolution without further qualification, the specification is incomplete: one must also state the MTF contrast for that resolution. MTF at 50% contrast corresponds well to perceived sharpness. The MTF at Rayleigh is about 9% and one can barely make out the image details at this contrast. If we look at Roger's table (Optical System Resolution Limits) we see that the MTF 50 for f/8 is 97 lp/mm and the MTF 50 resolutions for f/11, f/16, and f/22 are 71, 48, and 35 lp/mm respectively.

By this criterion, f/8 , f/11, and f/16 should give resolution at 50% contrast at or above Nyquist. At f/22 the resolution at 50% contrast is only 35 lp/mm, well below Nyquist. However one can still resolve 75 lp/mm at 9% MTF (Rayleigh) so you may get some detail at low contrast at f/22. At the Dawes limit the MTF is zero and you will have no detail. The Nyquist for the Nikon D200 cropped sensor is 82 lp/mm. The camera should give decent resolution at f/11, but smaller apertures would be suboptimal.

These considerations correlate well with what Ray has observed in the field with his 5D: sharpness is good at f/11 and f/16. The image will suffer at f/22.

Bill
Title: f-stop limits for full sensor resolution
Post by: xtoph on January 25, 2007, 03:18:05 pm
Quote
By this criterion, f/8 , f/11, and f/16 should give resolution at 50% contrast at or above Nyquist. At f/22 the resolution at 50% contrast is only 35 lp/mm, well below Nyquist. However one can still resolve 75 lp/mm at 9% MTF (Rayleigh) so you may get some detail at low contrast at f/22.

These considerations correlate well with what Ray has observed in the field with his 5D: sharpness is good at f/11 and f/16. The image will suffer at f/22.

Bill
[a href=\"index.php?act=findpost&pid=97463\"][{POST_SNAPBACK}][/a]

well, i basically agree with you--and we both agree that the article to which we are replying is dead wrong.

as i stated in my first post, i actually started from what i observe in the field, and just posted the dollar bill tests (green, by the way) to prove the point beyond a shadow of a doubt. frankly i find it worthwhile to just _try_ the things we are talking about. it isn't that i don't think it also worthwhile to understand the physics behind what is happening--that is important. but we are talking about a set of tools--let's try them out. if we accepted what myhrvold wrote, we'd all be worried about turning our 13-17mp sensors into "2mp" sensors. posting articles like his just gives a bad name to those who do try to gain a rigorous understanding of what is happening inside our cameras--because anyone who uses the camera can see he's wrong, and many are prone to conclude that hiphalutin' talk about frequencies and stuff is just so much fertilizer.
Title: f-stop limits for full sensor resolution
Post by: BernardLanguillier on January 26, 2007, 04:03:36 am
It is also my feeling that when discussing what realy matters to most photographers - a body AND a lens- f10 is the optimal aperture on a D2x with most lenses.

One aspect that might explain some of the difference between theory and practise  is the presence of an AA filter on the sensors. My uneducated guess is that the presence of an AA filter on the sensor does to some extend contribute to making the effect of diffraction appear later that it would on a perfectly sharp filter, since the image is always blurred to some extend - even at optimal lens/difraction aperture.

sharpening works wonders on this because of the lack of noise of modern sensors at low ISO. It is also my feeling that diffraction is overall easier to correct by sharpening with digital images than with film, which tends to compensate for the smaller size of pixels as well.

Regards,
Bernard
Title: f-stop limits for full sensor resolution
Post by: francois on January 26, 2007, 05:19:46 am
Quote
...
One aspect that might explain some of the difference between theory and practise  is the presence of an AA filter on the sensors. My uneducated guess is that the presence of an AA filter on the sensor does to some extend contribute to making the effect of diffraction appear later that it would on a perfectly sharp filter, since the image is always blurred to some extend - even at optimal lens/difraction aperture...
Bernard,
Would a Leica M8 (or MF digital backs) be closer to the theory since it has no AA filter?
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 26, 2007, 08:03:09 am
Quote
Thanks for the data, and I agree about testing, but now see complications from the difference between essentially monochrome data and color data, and the extent to which sharpening can unravel the effects of Bayer interpolation.

I imagine that a sharpening algorithm that concentrates on luminosity using green information as the main measure of luminosity could more or less reproduce the resolution given by the green pixels, at least with near monochrome subjects, leading to more or less the Myhrvold threshold of "f-stop equal to pixel spacing in microns".

Next, to see diffraction effects on other colors, perhaps we need some of those test results that Foveon X3 sensor fans like: red-blue test patterns! Or at least sharpness tests on subjects with colors towards magenta (away from green).

Maybe the synthesis of observations so far is that diffraction effects start being measurable or even visibly noticeable with suitable subject matter once one passes the Myhrvold threshold (f/6 for the D200, close enough to your f/5.6 observation), with significant deterioration typically setting in beyond the "Thom Hogan" threshold of two stops beyond that (f/12 for the D200, close enough to your f/11 observation).
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=97317\")

BJL,

Your synthesis makes sense to me. Drs. Myhrvold and Johnson are both accomplished scientists and I'm not certain that their analyses are necessarily contradictory, but rather they are approaching the problem from different perspectives. Dr Johnson states in response: "I conclude that the Canon 1Ds, Mark II, with a pixel pitch of 7.2 microns, can use all of its resolution to describe an image at f/22, and that at f/2.8 aliasing would be severe without an anti-aliasing filter". This statement appears to be true: the sensor can largely capture the degraded image presented to it at f/22, but it would do better with an image presented at a larger aperture.

Aliasing can be severe when one is taking pictures of fabrics or other subjects with regularly repeating patterns at a certain frequency, but in nature shots these patterns do not frequently occur and the aliasing may not be that apparent. The Leica M8 seems to do well in most circumstances without a low pass filter as did the full frame Kodak sensor that failed largely due to its high noise.

With regard to Bayer color array sensors, there are many complications as you mention and these are analyzed in Dr. Johnson's reference ([a href=\"http://white.stanford.edu/~brian/papers/ise/CMOSRoadmap-2005-SPIE.pdf]Stanford Paper[/url]). The mathematics of point spread functions and convolution kernels are pretty exotic to me and I would expect to most photographers. For the time being, we need some rules of thumb and your synthesis appears reasonable to me.

Bill
Title: f-stop limits for full sensor resolution
Post by: Tim Gray on January 26, 2007, 09:13:12 am
Quote
A crucial factor when considering resolution as a function of pixel size is the magnification needed to go from sensor to print.  This depends on total sensor size, not pixel size or pixel count.  This is the real reason cameras with small pixels generally have worse "resolution" than cameras with large pixels -- the total sensor size is generally smaller, requiring higher print magnification.
[a href=\"index.php?act=findpost&pid=97197\"][{POST_SNAPBACK}][/a]

wow

Just making sure I understand what you're saying:

The reason a small 10 mpx P&S has lower apparent resolution (same as "less sharp?) than a 5d full frame  is because the sensor in the P&S is so much smaller (about the size of a thumb nail) than the full 35mm in the 5d, and the size and number of pixels doesn't matter?  Did I paraphrase that correctly?

If that's correct it would seem logical to conclude that a 35mm sensor with 4 pixels, would be sharper than the P&S with 10 mpx?  Giving you the benefit of the doubt, you MIGHT be claiming that a 10 mpx P&S isn't as sharp as a 10 mpx full frame because, keeping the pixel count the same, the sensor (hmm - I guess that means the individual pixels since we've kept the count the same) is smaller?

Either way it's 100% wrong.
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 26, 2007, 10:21:28 am
Quote
wow

Just making sure I understand what you're saying:

The reason a small 10 mpx P&S has lower apparent resolution (same as "less sharp?) than a 5d full frame  is because the sensor in the P&S is so much smaller (about the size of a thumb nail) than the full 35mm in the 5d, and the size and number of pixels doesn't matter?  Did I paraphrase that correctly?

If that's correct it would seem logical to conclude that a 35mm sensor with 4 pixels, would be sharper than the P&S with 10 mpx?  Giving you the benefit of the doubt, you MIGHT be claiming that a 10 mpx P&S isn't as sharp as a 10 mpx full frame because, keeping the pixel count the same, the sensor (hmm - I guess that means the individual pixels since we've kept the count the same) is smaller?

Either way it's 100% wrong.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=97647\")

I have to agree with Tim here. A digital image consists of pixels and really has no physical size in millimeters. If the pixels are of equal quality, it makes no difference whether they were derived from a small or large sensor. However, in the real world, the effects of diffraction and noise often favor large pixels as Michael points out in his essay and as [a href=\"http://www.clarkvision.com/imagedetail//does.pixel.size.matter2/]Roger Clark[/url] demonstrates on his web site.

Bill
Title: f-stop limits for full sensor resolution
Post by: Rob C on January 26, 2007, 10:26:44 am
Quote
I completely agree with Ray's point on film. All film of a particular type and ASA were cut from huge "master rolls" with no difference of film quality between film formats. I.E. there was always an advantage with larger film formats to make prints of a given size. With digital it is important to compare apples with apples as with film.
[a href=\"index.php?act=findpost&pid=97388\"][{POST_SNAPBACK}][/a]

Ray, Kirk

I'm a little unhappy with your statement/belief that films of the same name, in different formats, are the same thing.

It's been been my experience that this is simply not the case: sometimes the film is on a thicker base, other times the results look quite different. For example, in b/w I would use TXP 120 and get great results (Hass/Zeiss) whilst on 35mm it was far too contrasty for my purposes; Ilford's FP3/4 were excellent on 35mm (Nikon/Nikkors) but did not work nicely for me on 120. Both were processed by me in D76 1+1. In colour transparency material I found 35mm Velvia 50 (Nikon/Nikkors) to be very attractive but not so in 120 (Pentax/Pentax) 6x7. I do not believe this to be anything significantly to do with the lenses, more a function of the films bearing a common name but not common properties. As far as the b/w examples go, even the ASA ratings didn't pan out the same for films of the same name - the 120 seemed to need more exposure/lower ASA rating than the similarly named 35mm.

Ciao - Rob C
Title: f-stop limits for full sensor resolution
Post by: EricV on January 26, 2007, 01:33:39 pm
Quote
A digital image consists of pixels and really has no physical size in millimeters. If the pixels are of equal quality, it makes no difference whether they were derived from a small or large sensor. However, in the real world, the effects of diffraction and noise often favor large pixels ....
Bill
[a href=\"index.php?act=findpost&pid=97652\"][{POST_SNAPBACK}][/a]

A digital image is a record of an optical image, which most certainly does have physical size, and the optical image has characteristics like diffraction blur which do have physical size.  When this information is tranferred into pixels, it is not lost.  I guess you can forget about the physical size of the pixels and the sensor, provided you keep track of something equivalent, like the size of the diffraction blur measured in pixels.  With this understanding, we proabably have no real disagreement.

(I still maintain though that there is no fundamental difference between recording an image on digital pixels or analog film, at least in the limit where the recording medium has finer resolution than the optical image.  So why maintain that film size is significant, but digital sensor size is not?  Is it just because film physically reminds you how large it is, but a digital file makes it easy to forget?)

Let me give a simple example where I would claim sensor size matters and I imagine you would claim it does not.  I think we will agree on the final conclusions, if not on the semantics :)  

A) 50mm lens at f/11 on 1/2" sensor with 4um pixels
B) 50mm lens at f/22 on 1/2" sensor with 4um pixels
C) 100mm lens at f/22 on 1" sensor with 8um pixels

All three images have the same field of view and perspective.  All three images have the same pixel count.  However, the digital images are not equivalent.  

Image A will produce a sharper print than image B (but with less depth of focus), because the optical resolution is higher (less diffraction).

Image C will also produce a sharper print than image B, even though the optical resolution is the same.  From my viewpoint, this is because less magnification is required to make the final print.  In your view, I suppose it is because the diffraction blur spreads across fewer pixels.

Images A and C will produce similar prints, with comparable sharpness and depth of focus.  I concede that in this comparison sensor size does not matter, because optical resolution was scaled to match the sensor size.
Title: f-stop limits for full sensor resolution
Post by: Tim Gray on January 26, 2007, 02:01:01 pm
So you're suggesting a 5" (diagonal) print from the 1/2" sensor @ f22 would have the same apparent sharpness as a 10" print from the 1" @ f22 sensor?

What would happen at the wider end, where diffraction isn't an issue - eg: f2.8 rather than F22 (obviously ignoring abberation)?  Given the same size prints, the 1" should produce sharper results than the 1/2"?  I don't think so.
Title: f-stop limits for full sensor resolution
Post by: Kirk Gittings on January 26, 2007, 02:03:29 pm
Rob C, That is how film is manufactured. Now, if one is on a different base (thicker say) that of course is from a different roll but may be coated with the same emulsion or perhaps not. In order to make the multi format cross referenced comparisons that you are stating, all your shutters and meters would have to be calibrated to an exact standard, and the differences of all lens coatings and film processing differences factored in. For one example Schneider lenses from the 70"s to now are a whole paper grade more contrasty and significantly more neutral in color. Plus between lens systems over time color varies significantly in addition to contrast as multicoatings advanced. I own a dozen modern view camera lenses about half and half Schneiders and Nikons. If I shoot a building in 4x5 (I am an architectural photographer by trade) mixing those lens brands, I get widely varying color which is disturbing in a magazine layout (by my standards). The Nikons are warmer and less contarsty than the Schneiders. On the same shoot I may also shoot some 35mm and 120 details. Those will have a different look, but it is the because of the variation in lenses and shutters.

I routinely shoot both 4x5 and 120 (switching backs) with the same lens, in the same camera, at the same time, with the same scene. If the film is the same ( I primarliy use Fuji Provira) the results are identical.
Title: f-stop limits for full sensor resolution
Post by: Rob C on January 26, 2007, 03:59:44 pm
Kirk

Thanks for your reply - your example of using the same name of film in two formats on the same lens/camera combo seems to make good sense as a definitive test, at least of the film you quote. That was something I couldn't do with my own equipment so your experience there certainly sounds convincing.

Come to think of it, it might be part of the same argument that led to Nikon cameras and lenses replacing the Leicas of war photographers lo those many years ago: more bite, more contrast for printing in LIFE!

Yet there we go again: I mention Leica being replaced by Nikon but Leica glass still seems to be considered as producing a more MF kind of look than do other makes of optics for 35mm, at least on colour transparency.

Funny old world.

Ciao - Rob C
Title: f-stop limits for full sensor resolution
Post by: EricV on January 26, 2007, 04:03:50 pm
Quote
So you're suggesting a 5" (diagonal) print from the 1/2" sensor @ f22 would have the same apparent sharpness as a 10" print from the 1" @ f22 sensor?

What would happen at the wider end, where diffraction isn't an issue - eg: f2.8 rather than F22 (obviously ignoring abberation)?  Given the same size prints, the 1" should produce sharper results than the 1/2"?  I don't think so.
[a href=\"index.php?act=findpost&pid=97675\"][{POST_SNAPBACK}][/a]

Comparisons should always be done for the same print size, otherwise too many other factors like viewing distance must be considered.  I am claiming that a 10" print from a 1"sensor will be sharper than a 10" print from a 1/2" sensor, provided the lens focal length is adjusted to cover the full sensor with the same field of view in both cases, provided the optical resolution is the same in both cases (which is the case if both lenses are stopped down to f/22), and provided the pixel size is much smaller than the optical resolution in both cases.  I think everyone on this forum will agree with this claim.  If you change some of these assumptions (like making the pixel size ridiculously large), then the conclusion will change.

What will happen if diffraction is not an issue?  In the extreme limit where optical resolution is much smaller than pixel size, the advantage goes to whichever camera has more pixels, independent of sensor size.  This is a pretty extreme case.  A diffraction limited lens at f/2.8 still has a resolution of 4um, which is not negligible compared to pixels on most sensors.  

In your example, do both cameras get to use diffraction limited f/2.8 lenses?  Does the larger sensor still get to use a longer focal length lens, to provide the same field of view, and does its sensor still have the same pixel count?  Then the larger sensor will still produce a sharper print, because the optical blur is not negligible and it is magnified less.

If you want to create a situation where the large sensor has no advantage over the small sensor, simply pick a wider lens aperture (better optical resolution) for the small sensor (as I noted in my previous post).
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 26, 2007, 04:32:15 pm
Quote
A digital image is a record of an optical image, which most certainly does have physical size, and the optical image has characteristics like diffraction blur which do have physical size.  When this information is tranferred into pixels, it is not lost.  I guess you can forget about the physical size of the pixels and the sensor, provided you keep track of something equivalent, like the size of the diffraction blur measured in pixels.  With this understanding, we proabably have no real disagreement.

(I still maintain though that there is no fundamental difference between recording an image on digital pixels or analog film, at least in the limit where the recording medium has finer resolution than the optical image.  So why maintain that film size is significant, but digital sensor size is not?  Is it just because film physically reminds you how large it is, but a digital file makes it easy to forget?)

[a href=\"index.php?act=findpost&pid=97667\"][{POST_SNAPBACK}][/a]

I never said that sensor size was more important with film than with digital. However, 35 mm film is the limitation for a 35 mm film camera. If we had better film and diffraction limited lenses, 35mm at f/11 would be about equal to f/45 for a 4x5 camera with current film. To keep the diffraction blur the same with respect to pixel size, you do have to use larger apertures with small pixels and you reach a limit as you approach f/1.0.

Bill
Title: f-stop limits for full sensor resolution
Post by: xtoph on January 28, 2007, 04:41:25 am
wow. this myhrvold guy just isn't listening. or looking. bad enough to post one article proposing a flawed model, but then we all make mistakes. to post yet again, and with such a condescending tone, insisting that your original mistake was right all along... well, that's really unfortunate. and reichmann isn't helping by 'sticking to the sidelines'--neutrality is fine in matters of opinion or indeterminacy, but this is a case of settled fact. the cropped images i posted above prove conclusively what most of us already knew from experience--that myhrvold's predictions were incorrect; a 5d at f/22 is not equivalent to a 2mp sensor. now myhrvold claims that at f/22 the image is equivalent to f/8 with a gaussian blur of 1.5 applied. well, here is the same crop from above at f/8, with gaussian blur. you can plainly see that it has lost the ability to distinguish fine detail, and usm will not bring it back. myhrvold is consistent at least--this image is almost indistinguishable from the one based on his original claim of 2mp.

the problem is, it bears no resemblance to the actual image you get from the 5d at f/22 (see above post). in fact, on the 'cambridge in color' website myhrvold links in his latest post to support his claim that at f/22 we will lose massive amounts of detail, we can read the following statement clearly disagreeing:
"Recall that a digital sensor utilizing a bayer array only captures one primary color at each pixel location, and then interpolates these colors to produce the final full color image.  As a result of the sensor's anti-aliasing filter (and the Rayleigh criterion above), the airy disk can have a diameter approaching about 2 pixels before diffraction begins to have a visual impact (assuming an otherwise perfect lens, when viewed at 100% onscreen)."

this author actually agrees with johnson and about the source of myhrvold's misunderstanding of the situation. their explanation sounds plausible to me. but more to the point, the empirical tests prove that myrhvold is wrong.

really, he ought to retract his posts and issue an apology at this point.
[attachment=1677:attachment]
Title: f-stop limits for full sensor resolution
Post by: Tim Gray on January 28, 2007, 09:03:54 am
I wonder if Michael has his contributors sign a waiver acknowledging the slings and arrows they are liable to be subjected to    I remember a tutorial by Mitch? on, I think, selective desaturation - or something like that.  Never heard from again...

Having said that, if you set yourself up as an authority, then you're inviting this kind of controversy.   I guess it's a little bit like academic peer review, with the gloves off.  The  advantage of the "traditional model" is that review happens before publication - in today's internet world publication and review happen simulaneously - which means the review process can be a bit spirited.  

I do think, in the end, that our overall understanding and appreciation for what's really going on does increase - but we have to be a lot more diligent in exercising our own critical analysis.   What we read here isn't from Nature Journal.
Title: f-stop limits for full sensor resolution
Post by: bjanes on January 28, 2007, 10:13:41 am
Quote
wow. this myhrvold guy just isn't listening. or looking. bad enough to post one article proposing a flawed model, but then we all make mistakes. to post yet again, and with such a condescending tone, insisting that your original mistake was right all along... well, that's really unfortunate. and reichmann isn't helping by 'sticking to the sidelines'--neutrality is fine in matters of opinion or indeterminacy, but this is a case of settled fact.

this author actually agrees with johnson and about the source of myhrvold's misunderstanding of the situation. their explanation sounds plausible to me. but more to the point, the empirical tests prove that myrhvold is wrong.

really, he ought to retract his posts and issue an apology at this point.
[attachment=1677:attachment]
[a href=\"index.php?act=findpost&pid=97893\"][{POST_SNAPBACK}][/a]

This myhrvold guy is not your typical crackpot who posts ill conceived messages. If you look at his bio on Wikipedia: "studied physics at Princeton (PhD) and held a postdoctoral fellowship at Cambridge working under Stephen Hawking for one year, but left to join a computer startup in Oakland, California. The company, Dynamical Systems Inc., sought to produce a clone of IBM's TopView graphical user interface. Microsoft purchased Dynamical Systems in 1986 and Myhrvold worked there for 13 years" and held the position of chief technology officer.

Johnson also holds a PhD in chemistry and has authored 150 scientific papers. One could say these men may not be expert in optics and digital imaging, but they do understand the scientific method and rigorous analysis and are without doubt highly intelligent.

Now, I am puzzled by Dr. Johnson's statement that the EOS 1D M2 can use all of its resolution to resolve an Airy disc at f/22. I'm sure the Airy disc would be resolved in exquisite detail, but overall image resolution would suffer IMHO. Myhrvold's analysis of resolution at f/22 makes more sense to me, but I do not understand the complex mathematics discussed in the Stanford reference they quote. I think Michael is wise to keep out of this "argument" and I am not qualified to mediate either, and I suspect that very few forum members so qualified. I hope we can learn from a few more exchanges by these two experts. However, when a complicated scientific argument contradicts common experience, we can and should make comments.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on January 28, 2007, 10:01:53 pm
Quote
However, when a complicated scientific argument contradicts common experience, we can and should make comments.

Bill
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=97914\")

That's a very reasonable argument to make, Bill. Science is based upon empirical evidence. No matter how 'impressive' and 'high faluting' the the theory appears to be, if it doesn't accord with real world conditions, or practice, then it has to be junked.

The fact that someone who has 6 PhDs (or whatever) can be simply wrong, is a difficult accusation to make. However, if the theory does not agree with the empirical evidence, then we have a problem.

In the Middle Ages in England, and later, when someone disagreed with the official (PhD sanctioned) position, they were often tortured, mutilated and burned.

I'm glad we have progressed from that situation   .

In general, I see a lack of empirical evidence in these discussions. Okay! Maybe I and one or two others post a few images, but the discussions are mostly hot air rhetoric.

Maybe I posted the following images in the wrong thread.

[a href=\"http://luminous-landscape.com/forum/index.php?act=ST&f=3&t=14317&st=0]http://luminous-landscape.com/forum/index....=3&t=14317&st=0[/url]
Title: f-stop limits for full sensor resolution
Post by: xtoph on January 29, 2007, 03:40:46 am
Quote
Now, I am puzzled by Dr. Johnson's statement that the EOS 1D M2 can use all of its resolution to resolve an Airy disc at f/22. I'm sure the Airy disc would be resolved in exquisite detail, but overall image resolution would suffer IMHO. Myhrvold's analysis of resolution at f/22 makes more sense to me, but I do not understand the complex mathematics discussed in the Stanford reference they quote. I think Michael is wise to keep out of this "argument" and I am not qualified to mediate either, and I suspect that very few forum members so qualified. I hope we can learn from a few more exchanges by these two experts. However, when a complicated scientific argument contradicts common experience, we can and should make comments.

Bill
[a href=\"index.php?act=findpost&pid=97914\"][{POST_SNAPBACK}][/a]

with all due respect, myhrvold is obviously not an expert on the question of optical diffraction effects on digital sensors. take a look at the crops i've posted here, or do them yourself (in order to test his assertions, you need to have detail visible at less than the width of two pixels). he says that while such details can be resolved at f/8 or 9, by the time you reach f/22 your full-frame sensor cannot resolve more than 2mp worth of detail. or, in his second post, he substitutes what turns out to be a similar claim; that at f/22 such an image is equivalent at best to the f/8 performance with a gaussian blur of radius 1.5 applied.

these at least have the merit of being testable assertions. i've tested them, and they are wrong. i used a lens that comes close to its peak performance at f/8 (and is darn good), so we avoid the problems with samples coming from an extreme telephoto that is struggling to perform at f/8 (ie, near wide open, including teleconverters). such a test is invalidated (especially if looking at detail much larger than 2 pixels across) because it is a poor approximation of the ideal (diffraction limited) lens. in other words, such a test would mainly reflect the (expected) performance of a less than ideal lens, ie, it gets better when stopped down.

i think it is silly to assume that just because a person is making an argument from authority and invoking some essoteric information we should stay out of the discussion. he's made clear predictions; we can test them; let's do it and move the discussion forward. sitting on the sidelines hoping to 'learn' something here is pointless--unless you're mainly interested in ego psychology.

the problem is that the one who ought to be looking to learn something here is myhrvold. but clearly he isn't. case in point: johnson pointed out that the way the sensor interpollates informatino from multiple photosites to calculate the value of single pixels may be the souce of the confusion; the cambridge in color website makes the same point (i quoted it above); and in fact i made the same suggestion in my first post, before these two got involved. so, myrhvold's prediction based on his model is demonstrably wrong; at least three people have suggested a plausible source of the error; and myhrvold is just closing his eyes and repeating that the laws of the universe are on his side.

it is very hard not to see it all as a grand political metaphor.
Title: f-stop limits for full sensor resolution
Post by: BernardLanguillier on January 29, 2007, 04:29:36 am
Quote
Bernard,
Would a Leica M8 (or MF digital backs) be closer to the theory since it has no AA filter?
[a href=\"index.php?act=findpost&pid=97631\"][{POST_SNAPBACK}][/a]

That would indeed be my guess. At least it should be more useful as an experiment subject than a 1s2 or D2x.

It would of course still not account for the influence of the lens.

Regards,
Bernard
Title: f-stop limits for full sensor resolution
Post by: francois on January 29, 2007, 08:16:51 am
Quote
...It would of course still not account for the influence of the lens.
...
Of course, we all know that Leica lenses are perfect! At least, that's what the local camera store owner told to a potential M8 customer.
Title: f-stop limits for full sensor resolution
Post by: marcmccalmont on January 29, 2007, 10:24:54 am
Quote
That would indeed be my guess. At least it should be more useful as an experiment subject than a 1s2 or D2x.

It would of course still not account for the influence of the lens.

Regards,
Bernard
[a href=\"index.php?act=findpost&pid=98056\"][{POST_SNAPBACK}][/a]
I did a quick test with a 5D that has no AA filter, F8 was a bit sharper than F22 what one would expect from the lens in its sweet spot not drastic like 13 mp to 2 mp?
Marc
Title: f-stop limits for full sensor resolution
Post by: jani on January 31, 2007, 05:52:24 pm
Even though people have PhDs and whatnot, they can still make mistakes. That's what friends, colleagues and editors are for. Unfortunately, the peer review process apparently is a bit less stringent when posting to a website, so it's entirely possible that mistakes don't get caught.

In Myhrvold's calculations, it could be something as simple as a missing square root somewhere, which also seems to match the difference he notes between his usage and that of Cambridge in Colour.

Now for the 1.5 radius gaussian blur statement; that was for the 1Ds MkII, not the 5D, which xtoph (Christoph(e)?) posted images for.

Let's pretend that there was a square root error, and that in addition, we need to divide by the factor of minimum f-stops between the 1Ds MkII and the 5D (1.125), to find the correct gaussian blur radius to use. That would then be a radius of 0.943 instead of 1.500.

Here's xtoph's f/22 picture:

[attachment=1710:attachment]

And here's the f/8 with a 0.9 radius blur:

[attachment=1707:attachment]

Okay, that's still a bit off.

But what if Myhrvold confused radius and diameter?

The f/8 with a 0.7 radius blur:

[attachment=1708:attachment]

And what if we assume that Cambridge in Colour is right (1.414 factor reduction), as well as that Myhrvold confused radius and diameter?

The f/8 with a 0.5 radius blur:

[attachment=1709:attachment]

Hmm ...

Edit: A consistent typo, AKA "brainfart", of 1.5 instead of 0.5 would also explain the observations.
Title: f-stop limits for full sensor resolution
Post by: xtoph on February 09, 2007, 04:01:50 am
well jani, i appreciate the effort to track down where myhrvold went wrong. i agree that even the titled are 'entitled' to make mistakes (though if someone politely points out that you're wrong, and you rudely insist you haven't made any mistake, well, that's a different case). i realize that the 1.5 gaussian blur radius factor was probably based on the 1dsII, but i don't have one handy to do that test (my old 20d though, with even greater pixel density than the 1dsII, gives remarkably similar results to the 5d, better than the example on the cambridge site shows--not sure what is going on with that and can't do further tests, as i gave the 20d to my brother. but we don't know what lens the cambridge site used, and we don't really know much else--raw or jpg, noise reduction or not, etc, which may have affected that image. i don't deny that the 20d captures less detail at f/22 than the 5d--i just didn't see the very dramatic kinds of effects illustrated on their textured fabric). what i found interesting was that the results of a 1.5 radius blur on the f/8 5d image look exactly the same as the image which myhrvold first suggested, ie that at f/22 you had only 2mp worth of resolution (a statement he applied to both the 1dsII and the 5d--see his original post).

so here's the problem: myhrvold actually was completely consistent (and equally wrong) in both of his posts. this makes it harder to believe that the way he got there was through misplacing decimals on the radius calculation, or confusing radius and diameter (though perhaps that was a factor).

i still suspect, as i stated in my first post, that the problem is related to misunderstanding how the interpolated data from multiple photosites relates to 'pixels'. that would be an interesting subject to take up, in my opinion. it would be helpful to know how interpollation affects the diffraction limits (and i am unconvinced by myhrvold's dismissal of it given my tests). i could read the papers johnson references; i bet they would help. but it just isn't that high on my list of things to do. in the meantime i will continue to get results from my camera at f/16-22 which, according to myrhvold, violate the laws of physics.
Title: f-stop limits for full sensor resolution
Post by: jani on February 15, 2007, 05:57:06 pm
Quote
well jani, i appreciate the effort to track down where myhrvold went wrong. i agree that even the titled are 'entitled' to make mistakes (though if someone politely points out that you're wrong, and you rudely insist you haven't made any mistake, well, that's a different case).
To put it this way: even a title in computer science doesn't mean that you've done anything worth s... in programming, or know your way around complex algorithms.

That being said, even if you had the relevant experience, you could still mess up and don't realize it.

It's a core human problem that when we've written something of our own, it's very hard to find fault with it. Most people -- even among the experts -- will need a second set of eyes, a peer reviewer if you like, who can find these faults.

Neither Johnson nor Myhrvold appear to have had anyone review their articles/musings before they were published, which the general quality of what they wrote speaks volumes about.

If they'd been submitting a scientific paper, you can bet that they'd both been more thorough.

Instead, the tone was informal.

That's why I think it's entirely likely that a basic mistake can sneak into the text and remain there, in spite of any number of doctorates the person holds.

But enough said about that; Myhrvold was clearly wrong in the factual details of what he wrote, whatever the reason.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 15, 2007, 06:26:52 pm
Quote
Neither Johnson nor Myhrvold appear to have had anyone review their articles/musings before they were published, which the general quality of what they wrote speaks volumes about.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101123\")

One could consider the posting of their articles on LL a form of seeking second opinion. I notice that professor Charles Johnson has now invited comments on a book he's preparing on the science and technology of photography. No-one (except me) seems to have shown much interest on this site so far. The thread is [a href=\"http://luminous-landscape.com/forum/index.php?showtopic=14767&hl=]http://luminous-landscape.com/forum/index....topic=14767&hl=[/url] and Charles Johnson's first 3 chapters of his book are at http://photophys.com/photophys/ (http://photophys.com/photophys/)
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 15, 2007, 06:32:44 pm
I've done macro shots at f/36 with a Canon 35-350 L zoom, which is far from the sharpest lens Canon makes, and the resulting images still have significantly more than 2MP of detail. Myhrvold is clearly off in his calculations somewhere.
Title: f-stop limits for full sensor resolution
Post by: BJL on February 15, 2007, 06:51:26 pm
Quote
I've done macro shots at f/36 with a Canon 35-350 L zoom, which is far from the sharpest lens Canon makes, and the resulting images still have significantly more than 2MP of detail. Myhrvold is clearly off in his calculations somewhere.
[a href=\"index.php?act=findpost&pid=101135\"][{POST_SNAPBACK}][/a]
I think he is off in his interpretation: the "brick wall" fallacy that the resolution of an optical system is equal to the resolution of the lowest resolution component. Some reading about the multiplicative nature of MTF might be in order.

Maybe at f/22, diffraction spots are about as large as the photo-sites of a 2MP 35mm format sensor, but reducing the pixel size (increasing pixel count well beyond 2MP) will still increase overall resolution. Given the soft edges of diffraction spots (so that they do not have a single clear cut diameter), I am fairly sure that one could reduce pixel size to half that much and still be gaining in overall resolution, so maybe about 8MP in 35mm format is more like the resolution limit of f/22.

That seems to fit the data better anyway: film matching about 8MP to 16MP in 35mm format and f/16 to f/22 being where diffraction srarts to significantly reduce resolution with film.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 15, 2007, 08:57:19 pm
Quote
I've done macro shots at f/36 with a Canon 35-350 L zoom, which is far from the sharpest lens Canon makes, and the resulting images still have significantly more than 2MP of detail.
[a href=\"index.php?act=findpost&pid=101135\"][{POST_SNAPBACK}][/a]

More significant detail than which 2MP, with which lens at which f stop?

This entire issue is confused by 2 factors. What the heck do we mean by 'diffraction limitation' and what the heck is a pixel?

We know that a Foveon pixel (consisting of red, green and blue element) carries far more weight than a Bayer type pixel which is in fact monochrome.

True diffraction limitation is something one perhaps never sees demonstrated. My understanding of diffraction limitation at a particular f stop is that an image will show no trace of aberration at that f stop. A diffraction limited lens is, in a sense, bearing in mind the limitation of Physics, a perfect lens.

That is, at f16 diffraction is more predominant than at f8, but there may still be traces of other aberrations at that f stop. At f8 aberrations such as spherical aberration , coma and chromatic aberration may predominate, with aperture diffraction taking a back seat but still contributing to the degradation of the image.

I think we kid ourselves if we think that even the best 35mm lenses are truly diffraction limited at f8 or f11 or even f16. All we can say is that they are more diffraction limited as we stop down, and more aberration limited as we stop up.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 15, 2007, 09:58:37 pm
Looking at it another way, if one defines 'diffraction limitation' as the f stop at which image degradation due to diffraction is greater than 50% of all the factors, in aggregate, that conspire to reduce image sharpness, then the f stop at which diffraction limitation applies will vary enormously with the lens used.

Am I right or am I right?
Title: f-stop limits for full sensor resolution
Post by: BJL on February 16, 2007, 10:50:23 am
Quote
Looking at it another way, if one defines 'diffraction limitation' as the f stop at which image degradation due to diffraction is greater than 50% of all the factors ...
[a href=\"index.php?act=findpost&pid=101158\"][{POST_SNAPBACK}][/a]
I prefer to avoid attaching a single number to what is in fact a gradual transition: a range of values is a better indication of the situation. (What scientists and engineers might call specifying an "error bar" for a result.)  The physical reality is that at _any_ aperture, diffraction has some effect, though very slight at large apertures, and the effect increases is a continuous way as aperture is increased, with no sudden onset of diffraction effects at some particular aperture.

I am happy with the following two stop range, supported by observation and theory:
- diffraction effects are negligible at f-stops less than the the pixel spacing of standard Bayer CFA sensors.
- diffraction effects are clearly significant at f-stops greater than twice this length scale.

E.g. the transition zone for current 10MP 6 micron SLR sensors is about f/6 to f/12.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 16, 2007, 12:05:59 pm
Quote
E.g. the transition zone for current 10MP 6 micron SLR sensors is about f/6 to f/12.
[a href=\"index.php?act=findpost&pid=101216\"][{POST_SNAPBACK}][/a]

There are some lenses that are sharper at f12 than at f6. If you are very lucky, you might have a lens that is marginally sharper at f4 than at f8.
Title: f-stop limits for full sensor resolution
Post by: BJL on February 16, 2007, 01:41:14 pm
Quote
There are some lenses that are sharper at f12 than at f6.
[a href=\"index.php?act=findpost&pid=101231\"][{POST_SNAPBACK}][/a]
It might be better to say that "some lenses are less sharp at f/6 than at f/12", which is almost certainly due to aberration problems at f/6: they in no sense have an advantage over other lenses at f/12, but only a disadvantage relative to other lenses at f/6. The discussion of diffraction versus sensor resolution ignores such poor lens performance as a factor.

Quote
If you are very lucky, you might have a lens that is marginally sharper at f4 than at f8.
[a href=\"index.php?act=findpost&pid=101231\"][{POST_SNAPBACK}][/a]
Perhaps I am lucky then with my Olympus 50-200/2.8-3.5, for which this is reportedly the case. However I am not sure if it is a matter of luck so much as a matter of choice of lenses and formats. The f-stop giving best resolution tends in general to increase as focal length and image circle size increases, due to natural scaling laws applied to lens designs, so absolute statements about particular f-stops like f/4 and f/8 are at best applicable only to one format or some specific range of formats.  Resolution tests I have seen show that many new "digital specific" lenses for formats smaller than 35mm (DX, EF-S, FourThirds, etc.) have their best resolution at f/4 or lower, and certainly at below f/8.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 16, 2007, 07:20:17 pm
Quote
.... they in no sense have an advantage over other lenses at f/12, but only a disadvantage relative to other lenses at f/6. The discussion of diffraction versus sensor resolution ignores such poor lens performance as a factor.
[a href=\"index.php?act=findpost&pid=101246\"][{POST_SNAPBACK}][/a]

Except in the case where a lens is sharper at f16 than it is at f12. I believe there are such lenses. They tend to be budget telephoto zooms. Michael has reviewed such a lens using the DXO analyzer.

Ignoring such factors of course is what Myhrvold does and consequently his calculations appear to be at odds with real world experience.

Without getting into the complication of trying to equate Airy disc size with pixel size, the following statements appear to be true.

1. A lens that is fully and completely diffraction limited at f8 will also be fully and completely diffraction limited at larger f stops of f11 and f16 and so on.

2. A lens that is fully diffraction limited at f8 will have twice the resolution, at f8, as it has at f16.

3. An 8mp sensor has twice the resolution of a 2mp sensor, all else being equal.

4. This same lens, fully diffraction limited at f8, will have 1.4x the resolution at f16 as it has at f22.

5. A 16mp sensor has 1.4x the resolution of an 8mp sensor.

Put briefly, this ideal lens, which probably doesn't exist, has 2.8x the resolution at f8 as it has at f22. A 16mp sensor has 2.8x the resolution of a 2mp sensor. Quadruple the pixel count and you double the resolution. Double the pixel count and you increase resolution by a factor of 1.4x.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 16, 2007, 07:54:27 pm
Quote
Perhaps I am lucky then with my Olympus 50-200/2.8-3.5, for which this is reportedly the case. However I am not sure if it is a matter of luck so much as a matter of choice of lenses and formats. [a href=\"index.php?act=findpost&pid=101246\"][{POST_SNAPBACK}][/a]

I was of course referring to 35mm lenses. There are very few that would be sharper at f4 than at f8. One should also bear in mind that such lenses, whether they are rare examples of 35mm lenses or more common examples of Zuiko lenses, are nowhere near diffraction limited at f4.
Title: f-stop limits for full sensor resolution
Post by: BJL on February 17, 2007, 06:21:49 pm
Quote
whether they are rare examples of 35mm lenses or more common examples of Zuiko lenses, are nowhere near diffraction limited at f4.
[a href=\"index.php?act=findpost&pid=101297\"][{POST_SNAPBACK}][/a]
I am not sure that I care about whether it is diffraction or something else that primarily limits resolution; I mainly care about what the overall resolution is, either in "lines per picture height", or in "lines per mm" (the latter particularly when my longest focal length is not enough to fill the frame, so I need to use some combination of tele-convertors and cropping). But to what extend that resolution limits come from diffraction, aberration, OOF effects, etc. is not of much significance to the end product.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 17, 2007, 07:51:47 pm
Quote
But to what extend that resolution limits come from diffraction, aberration, OOF effects, etc. is not of much significance to the end product.
[a href=\"index.php?act=findpost&pid=101439\"][{POST_SNAPBACK}][/a]

It has great significance in the context of Myhrvold's claim that using a lens at f22 with a 16mp camera such as the 1Ds2, is equivalent to using a 2mp camera of the same pixel pitch at f8, a claim that is widely disputed in this thread.

There might be a number of reasons why this does not appear to be true in practice, but I think a major reason is that a lens at f22 is much closer to being diffraction limited than a lens at f8, generally, in practice.

An interesting experiment would be to shift the range of apertures down so we can be more confident that the lens at those apertures really is diffraction limited.

For example, my 100-400 IS, at 400mm, stops down to f40. I think we can be confident that at this aperture the lens is diffraction limited. 3 stops up from f40 we get (approx.) f14 which, with this lens, might not be a fully diffraction limited f stop, but is much closer to being fully diffraction limited than possibly any 35mm lens at f8.

To give Myhrvold's assertion a fairer test, I would propose the following experiment.

1. Shoot a detailed subject at f40 with the 100-400 IS and 1Ds2.

2. Step back the appropriate distance so that a 2mp crop of the same scene will have the same FoV. (Requires a bit of mathematical calculation.)

3. Shoot the same scene at f14. Crop that shot to the same FoV as the first shot.

4. Compare detail.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 18, 2007, 04:08:34 am
Quote
To give Myhrvold's assertion a fairer test, I would propose the following experiment.

1. Shoot a detailed subject at f40 with the 100-400 IS and 1Ds2.

2. Step back the appropriate distance so that a 2mp crop of the same scene will have the same FoV. (Requires a bit of mathematical calculation.)

3. Shoot the same scene at f14. Crop that shot to the same FoV as the first shot.

4. Compare detail.

There's a much easier and faster option:

1. Shoot a detailed subject at  f/22 or smaller.
[attachment=1865:attachment]

2. Downsize the the shot to 2MP.

3. Upsize the shot back to the original pixel dimensions.
[attachment=1866:attachment]

4. Compare the original and the downsized-upsized images. If Myhrvold's assertions are correct, there won't be much difference between the two. But if the original is significantly better, he misplaced some decimal points or something.

The attached images are crops of a 1Ds macro shot with the 35-350L at f/32, which I sharpened and processed normally, saved the first crop, downsized to 2MP (1154x1734), upsized back to original size, and saved the second crop. I'll let you all compare for yourselves and come to your own conclusions.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 18, 2007, 08:45:44 am
Quote
There's a much easier and faster option:

The attached images are crops of a 1Ds macro shot with the 35-350L at f/32, which I sharpened and processed normally, saved the first crop, downsized to 2MP (1154x1734), upsized back to original size, and saved the second crop. I'll let you all compare for yourselves and come to your own conclusions.
[a href=\"index.php?act=findpost&pid=101490\"][{POST_SNAPBACK}][/a]

Jonathan,
That's quite interesting that you should interpret Myhrvolds's article in this way. I simply don't read it this way. There's no doubt in my mind that an 11mp shot at f22, downsized to 2mp then upsized back to 11mp, will have lower resolution than the original 11mp image. I wouldn't even bother to do the experiment to confirm this, although I recognise that the scientific method requires that one test the obvious. Sometimes there are surprises.

My interpretation of Myhrvold's  assertions are that the higher MTF response of a lens which is diffraction limited at f8 will give the same resolution with a 2mp camera as f22 will produce with a 12 or 16mp camera; same sensor size.

Below are the relevant quotes from his article. I agree that clarity could be improved.

Quote
Camera Max f/stop Max practical f/stop

Canon 5D f/8.6 f/9
Canon 1Ds f/9.3 f/9
Canon 1Ds Mark II f/7.6 f/8
Nikon D2X f/5.8 f/5.6
Canon 20D f/6.76 f/7.1
Canon G7 f/2.06 f/2

Note that in each case this is the maximum f-stop to get the full resolution of the camera. It is perfectly OK to stop the lens down further than this, but know that when you do, you will be getting less than the full resolution. This may or may not matter – lots of people obsess about resolution pointlessly (pun intended).

So, for example, if you take a Canon EOS 1Ds Mark II and stop it down to f/9, you are going to get greater depth of field, but you will not get the full 16 million pixel resolution – instead you’ll get resolution more like a 5D or 1Ds. That is still plenty good for many purposes. In fact, if you want the greater depth of field, then it may well be worth it. So, I still stop down to f/16 or even f/22 on occasion, but only when I decide that depth of field is more important that resolution – everywhere. When I know that I want to make a very large print, I stay at or below the maximum f-stop for diffraction limited resolution.
-----------------------------------------------------------------------------------------------------
Now, I don’t think anybody would be very excited about turning their EOS 1Ds Mark II, or Canon 5D or other full frame camera into a 2 megapixel camera. It sounds pretty drastic, but that is exactly what you do when you stop down to f/22 – the diffraction limit imposes this condition. If you shoot with a full frame 24 x 36 sensor at f/22 you are throwing away a lot of resolution. There is no getting around this – it is fundamental in the physics of light.

This comes as a shock to many photographers I have talked to, who assume that f/22 is the way to get the sharpest possible prints. Well, it just ain’t so. Diffraction limits your resolution at high f-numbers.
---------------------------------------------------------
Note that this is not quite the same as saying that it is equal to a 2 megapixel camera – that would only be true if the 2 megapixel camera was exposed with an f-stop that was within its diffraction limit.

I would add in reference to his comment,
Quote
This comes as a shock to many photographers I have talked to, who assume that f/22 is the way to get the sharpest possible prints
, that I think there would be few photographers who think that f22 will get them the sharpest print. I don't know who Myhrvold has been talking to, but everyone who knows anything about 35mm photography surely knows that f8 is more often than not the f stop that gives them the sharpest results, but not of course the greatest DoF.

However, there's something in the mathematics that might lead one to suppose that a lens which is truly diffraction limited at f8 is capabale of rendering the same detail with a 2mp camera as the same lens at f22 will render with a 16mp camera.

The problem is, I don't believe there are any 35mm lenses available which are diffraction limited at f8. Or to put it another way, I don't believe there are any 35mm lenses which have double the resolution at f8 that they have at f16.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 18, 2007, 09:24:21 am
I don't think anyone is arguing that diffraction does not have a degrading effect on image quality at f/22, but Myhrvold's statement that diffraction's effect when shooting at f/22 turns a "full frame camera into a 2 megapixel camera" is clearly wrong, as my example shows. 7-10MP, maybe, but certainly not 2. Even with the lens aberrations (which the 35-350 has plenty), it's definitely more than 2MP. Even given your interpretation of his statement, he's still way off. He's correct that diffraction decreases resolution as aperture diameter decreases, but way off with regard to how much.
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on February 18, 2007, 12:17:32 pm
Quote
There's a much easier and faster option:

1. Shoot a detailed subject at  f/22 or smaller.

2. Downsize the the shot to 2MP.

3. Upsize the shot back to the original pixel dimensions.

4. Compare the original and the downsized-upsized images. If Myhrvold's assertions are correct, there won't be much difference between the two. But if the original is significantly better, he misplaced some decimal points or something.
[a href=\"index.php?act=findpost&pid=101490\"][{POST_SNAPBACK}][/a]

One caveat here; downsampling and sampling back to original resolution is not the same thing as having an image at the lower resolution.  WHen you downsample, you wind up with a better image MTF and color resolution than you ever could have had with a lower-MP original version from the camera.  Of course, this helps your case even more, since a real 2MP image would have had poorer MTF, and less pixel-meaningful color.  IOW, your worst image would have looked even worse with a real 2MP camera.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 18, 2007, 12:41:29 pm
One could also argue that the double resampling process introduces some artifacts of its own, but in general I'd say your point is valid.

If theory and reality correlate poorly, it's generally not reality's fault.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 18, 2007, 04:25:08 pm
Quote
There's a much easier and faster option:
I'll let you all compare for yourselves and come to your own conclusions.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101490\")

I examined the images that Jonathan posted, and IMO the first has greater detail. However, I think that further analysis is needed before one can draw definite conclusions. Conclusions without any numeric quantification and analysis are always subjective, and Lord Kelvin (Sir William Thomson, 19th century British scientist, whose temperature scale we use in white balance) has summed up the matter:

"I often say that when you can measure what you are speaking about, and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meagre and unsatisfactory kind; it may be the beginning of knowledge, but you have scarcely in your thoughts advanced to the state of Science, whatever the matter may be." [PLA, vol. 1, "Electrical Units of Measurement", 1883-05-03]


Drs. Johnson and Myhrvold have published brilliant essays that explain their basic assumptions very well, but they have caused controversy when they extend their conclusions. Johnson states that "I conclude that the Canon 1Ds, Mark II, with a pixel pitch of 7.2 microns, can use all of its resolution to describe an image at f/22", which is somewhat cryptic. Myhrvold states that stopping down to f/22 causes his 16 mp camera to have 2 MP effective resolution.

The Brian Wadell et al paper from Stanford relates resolving power to the Nyquist limits of digital sensors for both monochrome and color work with a Bayer array imager. For monochrome, maximal resolution is achieved when the diffraction spot radius is equal to the pixel pitch, but for a Bayer array the effective pixel size is the 2 by 2 pixel kernel. The math becomes quite complex and an observational approach is more understandable for most of us. Myhrvold derives his 2 MP figure from the size of the Airy disc at f/22 using an assumed relationship with resolution.

[a href=\"http://www.photozone.de/8Reviews/lenses/canon_60_28/index.htm]Photozone[/url] has published resolution data for the Canon EF-S 60 mm f/2.8 macro and EOS 350D using Imatest. The results are expressed in line pairs/picture height, and this can be converted to lp/mm. This camera has a 22.4 x 14.8 mm sensor with pixel dimensions of 3458 x 2304 pixels, giving a pixel pitch of 6.4 microns and MP value of 8. The Nyquist limit for this sensor is 78 lp/mm. The data are summarized in here in tabular form:

[attachment=1879:attachment]

In modern optical theory using MTF, one must always state the contrast for a given resolution. Resolution without reference to contrast is ambiguous. Roger Clark (http://www.clarkvision.com/imagedetail/scandetail.html#diffraction) has published a table relating resolution to contrast and lens aperture for green light and the data are summarized in this graph. When one stops down to f/22, the Rayleigh resolution (about 9% MTF) is 75 lp/mm, only slightly below Nyquist as Dr. Johnson states, but this level of contrast does not yield a good image.

[attachment=1869:attachment]

Perceived sharpness in an image is related to the resolution at 50% contrast (MTF 50), and these values have been determined observationally by Klaus at Photozone. In the following table, I have interpolated the results for f/11 and f/22 and also calculate an effective megapixel resolution using the MTF 50 resolution for the 22.4 by 14.8 mm sensor.

[attachment=1881:attachment]

It is apparent that maximal MTF 50 is achieved at about f/4-f/5.6 when the Airy disc size is about equal to the pixel spacing. As one stops down further, the Airy disc expands and the MTF 50 decreases. At f/22, the effective resolution falls to 2.9 MP for this 8 MP camera.

Now perhaps Jonathan can repeat his experiment for 2.9 MP  . And it would be useful to have additional comments by the original expert authors on these matters.

Bill
Title: f-stop limits for full sensor resolution
Post by: BJL on February 18, 2007, 04:32:42 pm
Quote
It has great significance in the context of Myhrvold's claim that using a lens at f22 with a 16mp camera such as the 1Ds2, is equivalent to using a 2mp camera of the same pixel pitch at f8, a claim that is widely disputed in this thread.
[a href=\"index.php?act=findpost&pid=101446\"][{POST_SNAPBACK}][/a]
But Ray, you were talking about lenses not being diffraction limited at f/4, and comparing lens sharpness at f/4 to f/8, and resolutions limits at those relatively large apertures are not relevant to Myhrvold's claim that when you stop down to f/22 you turn [a] full frame camera into a 2 megapixel camera (and he uses the word "exactly".) Performance at f/4 is only relevant to this sort of claim if one moves to pixel size of about 4 microns or less, or 54MP territory for 35mm format.

I would hope that all my 35mm format lenses are largely diffraction limited by f/22, and my FourThirds lenses by f/11.  (Because otherwise they are "Coke bottles".)
Title: f-stop limits for full sensor resolution
Post by: Ray on February 18, 2007, 08:01:19 pm
Quote
But Ray, you were talking about lenses not being diffraction limited at f/4, and comparing lens sharpness at f/4 to f/8....

BJL,
Just as an aside in case some readers confuse the aperture at which a lens is sharpest as the aperture at which it is diffraction limited.

Quote
I would hope that all my 35mm format lenses are largely diffraction limited by f/22, and my FourThirds lenses by f/11.  (Because otherwise they are "Coke bottles".)

Perhaps you could do some experiments with your Olympus E1 and finest prime Zuiko lens at f22 and f8 and tell us how many pixels at f8 are equivalent to 5mp at f22.

These could be your first images posted on this site   .
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 18, 2007, 08:25:49 pm
Quote
It is apparent that maximal MTF 50 is achieved at about f/4-f/5.6 when the Airy disc size is about equal to the pixel spacing. As one stops down further, the Airy disc expands and the MTF 50 decreases. At f/22, the effective resolution falls to 2.9 MP for this 8 MP camera.

I disagree. My images prove that at f/32, the Airy disc size is not significantly larger than 1 pixel on a 1Ds. I'm seeing resolution approaching single-pixel level with a reasonably good level of contrast. The formulas are simply not correct; they do not accurately predict real-world results. As I said before, when reality and theory diverge, it is not reality's problem. I may not be able to exactly quantify what I'm seeing, but it is obvious that the real-world results do not match what the formulas (at least those used by Myhrvold) predict.

Another thing. Bayer sensors are capable of resolving much better than a 2x2 pixel kernel. The AA filter prevents the sensor from resolving all the way to single-pixel detail, but resolution much better than 4:1 is certainly possible.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 18, 2007, 08:27:58 pm
Quote
I don't think anyone is arguing that diffraction does not have a degrading effect on image quality at f/22, but Myhrvold's statement that diffraction's effect when shooting at f/22 turns a "full frame camera into a 2 megapixel camera" is clearly wrong, as my example shows. 7-10MP, maybe, but certainly not 2. Even with the lens aberrations (which the 35-350 has plenty), it's definitely more than 2MP. Even given your interpretation of his statement, he's still way off. He's correct that diffraction decreases resolution as aperture diameter decreases, but way off with regard to how much.
[a href=\"index.php?act=findpost&pid=101504\"][{POST_SNAPBACK}][/a]

He's certainly way off by your interpretation and he's also way off by my interpretation, although clearly not by as much.

I'm simply making the point that the major reason he is way off (by my interpretation) is due to some confusion as to what is meant by 'diffraction limitation'. As BJL has pointed out, there is a state of transition over a range of f/stops from the first f/stop (when stopping down) that begins to show the first hint of diffraction, to the f /stop that shows not the slightest hint of any of the other aberrations, just the effects of diffraction.

Myhrvold has made the following statements that I repeat from his article, which you appear to have missed. The implication here, I would suggest, is that Myhrvold is not saying what you attribute to him.

Quote
Camera Max f/stop Max practical f/stop
Canon 1Ds Mark II f/7.6 f/8

In other words, if you want the sharpest result s with the 1Ds2, do not stop down below f8 (although the theoretical limit is f7.6).

Quote
Note that this is not quite the same as saying that it is equal to a 2 megapixel camera – that would only be true if the 2 megapixel camera was exposed with an f-stop that was within its diffraction limit.

I understand by 'an f-stop within its diffraction limit' to mean an f-stop at which the lens is fully diffraction limited.
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on February 18, 2007, 08:39:59 pm
Quote
I understand by 'an f-stop within its diffraction limit' to mean an f-stop at which the lens is fully diffraction limited.
[a href=\"index.php?act=findpost&pid=101579\"][{POST_SNAPBACK}][/a]

What exactly would that mean, though?

In my mind, unless optical effects of diffraction cause the entire sensor to recieve the same light, there will always be *some* contrast between neighboring pixels, concerning a real-world high-contrast edge or point.  Just like the AA filter knocks down the possible contrast between adjacent pixels, so does diffraction.  And like the AA filter, the effects of diffraction are easily dealt with in digital, where math is at your command.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 18, 2007, 08:42:04 pm
Quote
Myhrvold has made the following statements that I repeat from his article, which you appear to have missed. The implication here, I would suggest, is that Myhrvold is not saying what you attribute to him.

Bull***t. The statement about a full-frame DSLR turning into a 2MP camera at f/22 is a direct quote from Mhyrvold. And it is clearly incorrect.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 18, 2007, 09:00:43 pm
Quote
What exactly would that mean, though?
[a href=\"index.php?act=findpost&pid=101580\"][{POST_SNAPBACK}][/a]

The degree to which sharpening can compensate for the softening of 35mm images at f16 and f22 is another issue. As you know, whole books have been devoted to nothing but sharpening routines.

Generally, my impression is that at f22, some low contrast detail is completely lost and cannot be recovered by any amount of skillful sharpening. At f8, that same low contrast detail will probably be captured. Furthermore, such detail can then be enhanced with appropriate sharpening.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 18, 2007, 10:29:14 pm
Quote
Bull***t. The statement about a full-frame DSLR turning into a 2MP camera at f/22 is a direct quote from Mhyrvold. And it is clearly incorrect.
[a href=\"index.php?act=findpost&pid=101582\"][{POST_SNAPBACK}][/a]

Jonathan,
I'm surprised at your sense of logic. Someone with a name so difficult to spell (and remember) as Myhrvold, must know something.  
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 19, 2007, 05:57:32 am
Quote
Generally, my impression is that at f22, some low contrast detail is completely lost and cannot be recovered by any amount of skillful sharpening. At f8, that same low contrast detail will probably be captured. Furthermore, such detail can then be enhanced with appropriate sharpening.

Nobody is disputing that diffraction has a negative effect on resolution at small apertures like f/22. The image I posted required more local contrast enhancement and more aggressive sharpening than a typical f/8 image, no question. I think a more realistic estimate of full-frame diffraction-limited resolution is something like 8MP at f/32, 32MP at f/16, and 128MP at f/8. It's certainly better than 2MP @ f/22.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 09:16:18 am
Quote
I disagree. My images prove that at f/32, the Airy disc size is not significantly larger than 1 pixel on a 1Ds. I'm seeing resolution approaching single-pixel level with a reasonably good level of contrast. The formulas are simply not correct; they do not accurately predict real-world results. As I said before, when reality and theory diverge, it is not reality's problem. I may not be able to exactly quantify what I'm seeing, but it is obvious that the real-world results do not match what the formulas (at least those used by Myhrvold) predict.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101578\")

Your disagreement is preposterous. The pixel spacing of the 1Ds MII is 7.21 microns and the Nyquist frequency is 69 lp/mm. For green light (500 nm), the Airy disc is 36 microns in diameter. At this aperture, the MTF 50 is 24 lp/mm, the resolution at Rayleigh (about 9% contrast) is 51 lp/mm and the resolution at Dawes (0% contrast) is 63 lp/mm. At 9% contrast, you will be able to resolve very high contrast objects such as star pairs in the night sky (the raison d'etre of the Rayleigh criterion), but this level of contrast is not very useful for practical photography of average scenes. You can recover some contrast with proper sharpening, but aliasing artifacts will also be accentuated. The mathematical derivation of the size of the Airy disc is given [a href=\"http://cnx.org/content/m13097/latest/]here[/url], along with nice 3 dimensional graphs of the discs at Rayleigh. I don't think your pictures at f/32 prove anything in a scientific way.

Myhrvold is a very smart guy, but I agree that the 2MP figure is off and he should clarify his derivation of that figure.

Quote
Another thing. Bayer sensors are capable of resolving much better than a 2x2 pixel kernel. The AA filter prevents the sensor from resolving all the way to single-pixel detail, but resolution much better than 4:1 is certainly possible.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101578\")

The 2x2 kernel has more to do with aliasing than actual resolution, according to the paper by [a href=\"http://white.stanford.edu/~brian/papers/ise/CMOSRoadmap-2005-SPIE.pdf]Wandell[/url] from the electrical engineering department at Stanford University. The anti-alaising filter is not completely effective, and if you look at the resolution tests at Dpreview.com near Nyquist, you will see plenty of detail, but much of it is false detail: aliasing. The MTF 50 of your system (in terms of lp/mm) would be slightly inferior to that of the D350 as shown on Photozone. However, the MTF 50 in terms of picture height would be better.

The mathematical modeling of Wandell et al is quite sophisticated. Just as nuclear scientists can test the design of an atom bomb with super computers, obviating the need for an actual detonation, Wandell could enter the parameters into his computer and end this discussion.

Bill
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 19, 2007, 09:48:59 am
Quote
Your disagreement is preposterous. The pixel spacing of the 1Ds MII is 7.21 microns and the Nyquist frequency is 69 lp/mm. For green light (500 nm), the Airy disc is 36 microns in diameter. At this aperture, the MTF 50 is 24 lp/mm, the resolution at Rayleigh (about 9% contrast) is 51 lp/mm and the resolution at Dawes (0% contrast) is 63 lp/mm. At 9% contrast, you will be able to resolve very high contrast objects such as star pairs in the night sky (the raison d'etre of the Rayleigh criterion), but this level of contrast is not very useful for practical photography of average scenes. You can recover some contrast with proper sharpening, but aliasing artifacts will also be accentuated. The mathematical derivation of the size of the Airy disc is given here (http://cnx.org/content/m13097/latest/), along with nice 3 dimensional graphs of the discs at Rayleigh. I don't think your pictures at f/32 prove anything in a scientific way.

They certainly prove that the 1Ds with a decent, but not excellent, lens at f/32 can still capture 11MP of image data with an MTF somewhere on the sunny side of 50%. Your MTF 50 figures are what is preposterous; if they were correct, the smallest details in my spider image would be 3-4 pixels in size, and the hairs in the first image would be indistinguishable form those in the second image. Since single-pixel-wide details are obviously present in the first image (the body/leg hairs), and are equally obviously NOT aliasing artifacts, there is an obvious error in the formulas being used to predict MTF-50 resolution for a given aperture and a diffraction-limited lens. I don't know exactly what the error is, or where it is, but there certainly is an error somewhere.
Title: f-stop limits for full sensor resolution
Post by: 01af on February 19, 2007, 10:49:06 am
Quote
Myhrvold is obviously not an expert on the question of optical diffraction effects on digital sensors.
Obviously not. He seems to believe the resolution resulting from a chain of two imaging devices (lens and sensor) where each has its own resolution limit was equal to the lower resolution of the two. I'd call this the "weakest-link theory." And this theory is wrong.

Actually the resulting resolution R[span style=\'font-size:8pt;line-height:100%\']res[/span] depends on the sequence of two input resolutions R[span style=\'font-size:8pt;line-height:100%\']1[/span] and R[span style=\'font-size:8pt;line-height:100%\']2[/span] like this:

1/R[span style=\'font-size:8pt;line-height:100%\']res[/span] = 1/R[span style=\'font-size:8pt;line-height:100%\']1[/span] + 1/R[span style=\'font-size:8pt;line-height:100%\']2[/span]

If this formula looks familiar to you---yes, it's the same that also describes the resulting resistance of two parallel resistors.

What does this mean? Let's say we have a sensor that due to pixel pitch can resolve up to 40 lp/mm. And we have three lenses that, for a given subject contrast, can resolve 40 lp/mm, 60 lp/mm, and 80 lp/mm respectively. Now, when using the 40 lp/mm lens on the 40 lp/mm sensor this seems like a good match, doesn't it? And using the better lenses on that sensor seems like a waste of resolving power as the poor sensor cannot exploit it, right? Wrong! Actually on the 40 lp/mm sensor, the 60 lp/mm lens will yield a sharper image than the 40 lp/mm lens, and the 80 lp/mm lens a sharper image still (albeit not twice as sharp as the 40 lp/mm lens).

Of course this works out the same the other way around. When using a lens that can resolve, say, 40 lp/mm, then a 60 lp/mm sensor will yield a sharper image than a 40 lp/mm sensor, and an 80 lp/mm sensor will yield a still sharper image (albeit not twice as sharp as the 40 lp/mm sensor).

So, the Myhrvold threshold of "f-stop equal to pixel spacing in microns" is a completely wrong concept---even when augmented by a corrective factor or two. Generally, the optimal f-stop roughly correlates to image size (among other things, the most important being lens quality) ... but not to pixel count or pixel pitch. With all other things equal, a higher pixel count will establish a higher overall resolution level and will make the degradation more obvious---but it will occur at the same aperture.


Quote
... observations of several users of the Nikon D2X, with 5.5 micron pixel spacing, say that diffraction starts to limit resolution at somewhere between f/8 and f/11.
With good lenses, this matches my own observations exactly. And my D-SLR camera has only half the D2X's pixel count (that is, 6 MP) and consequently, a pixel pitch of 7.8 microns which is 1.4× the pixel pitch of the D2X. So according to Myhrvold I should see diffraction setting in at apertures one stop smaller than D2X owners. But as a matter of fact I am seeing it at the same aperture as D2X owners do. And the only thing my camera has in common with the Nikon D2X is the image size which is APS-C (form factor 1.5×, relative to 35-mm format).

With good or very good (but not exceptional) lenses on APS-C format (form factor 1.5× or 1.6×), diffraction starts to become visible---upon very close inspection!---at apertures between f/8 and f/11 typically ... no matter what the pixel count is. With 35-mm-format cameras, the limiting f-stop is somewhere between f/11 and f/16. With medium-format cameras, it's around f/22. And so on.

With exceptionally good lenses, the limit is reached at apertures one or maybe even two stops larger (i. e. smaller f-stop numbers).

However, don't let these facts keep you from stopping down beyond these limits whenever you need the depth-of-field! Overall image quality does not so much depend on highest resolution at the plane of focus. If a composition depends on DOF then give it all the DOF it needs (but not more)!

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: BJL on February 19, 2007, 10:57:33 am
Diffraction does not smear light into a clearly defined disk with an unambiguous diameter. Instead, intensity falls of gradually from the center out to zero at a certain radius (the edge of the Airy disk), with further weak rings of light continuing beyond that. In principle, the rings continue out to infinity, but no one infers from that that diffraction reduces resolution to zero; instead, one has to ask what fraction of light is smeared by a certain distance, and look at questions like what MTF one achieves at a given length scale of number on line pairs per mm.

According to Norm Koren at http://www.normankoren.com/Tutorials/MTF6.html#Diffraction (http://www.normankoren.com/Tutorials/MTF6.html#Diffraction) diffraction alone give 50% MTF at 0.38/( N *W ) lp/mm where N is aperture ratio, W is the wavelength of light, and this corresponds to line pairs of width 1.078 times the Airy disk diameter of 2.44*N *W. That is, close enough to line pair width equal to the Airy disk diameter. By the usual estimate that it takes between two and three Bayer interpolated pixels to resolve a line pair, this suggest that for a sensor to extract all the "50% MTF detail" allowed by diffraction would need pixel spacing P between 1/2 and 1/3 of the Airy disk diameter, a factor of two or three smaller than the pixel size beyond which Myhrvold claims that results will be _exactly_ the same, due to resolution being determined by diffraction effects.

This suggests that for any given aperture ratio N, pixel size reduction can continue to improve resolution at least down to somewhere in the range 1.3*N*W to 0.88*N*W. Using wavelength W=555nm as Johnson does, the range is about P=0.7*N to P=0.5*N.

Turning this around, at a given pixel size P, one can probably increase aperture ratio at up to the range from N=1.4*P to N=2*P, and go to even higher aperture ratios if one can settle for significantly less than 50% MTF in the finest details of interest.

Honestly, I did not pre-plan the calculations to fit so nicely to my previous observation-based rule of thumb of aperture ratio one or two stops higher than pixel spacing in microns.
Title: f-stop limits for full sensor resolution
Post by: 01af on February 19, 2007, 11:11:10 am
Quote
Honestly, I did not pre-plan the calculations to fit so nicely to my previous observation-based rule of thumb of aperture ratio one or two stops higher than pixel spacing in microns.
And still this seemingly perfect fit is just pure coincidence ... as I explained in my previous post. Actually the aperture where diffraction will set in visibly does not I repeat: NOT depend on the pixel pitch. In several cameras with different pixel counts but equal image sizes (and with the same lenses) you will always observe the same limiting f-stop.

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 11:15:14 am
Quote
Actually the resulting resolution R[span style=\'font-size:8pt;line-height:100%\']res[/span] depends on the sequence of two input resolutions R[span style=\'font-size:8pt;line-height:100%\']1[/span] and R[span style=\'font-size:8pt;line-height:100%\']2[/span] like this:

1/R[span style=\'font-size:8pt;line-height:100%\']res[/span] = 1/R[span style=\'font-size:8pt;line-height:100%\']1[/span] + 1/R[span style=\'font-size:8pt;line-height:100%\']2[/span]
If this formula looks familiar to you---yes, it's the same that also describes the resulting resistance of two parallel resistors.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101659\")

Your formula applies only for MTF around the Rayleigh resolution criterion for contrast of 9%. For a more realistic MTF of 50% one must convert the MTFs into the frequency domain via a Fourier transform, multiply the frequency components, and then perform an inverse transform back to the spatial domain via a complicated process called convolution. This is explained by [a href=\"http://www.normankoren.com/Tutorials/MTF.html]Norman Koren[/url] on his web site.

Bill
Title: f-stop limits for full sensor resolution
Post by: 01af on February 19, 2007, 11:44:19 am
Quote
Your formula applies only for MTF around the Rayleigh resolution criterion for contrast of 9 %. For a more realistic MTF of 50 % one must convert [blah blah]. This is explained by Norman Koren (http://www.normankoren.com/Tutorials/MTF.html) on his web site.
You're right. But exactly that is also where Norman explains my formula is a good first-order approximation. And that's all we need here in our context.

The point is: Of course does pixel count have an effect on absolute resolution (as I said). But it does NOT have an effect as to where the limiting f-stop for a given lens is. The lens has an optimal f-stop where it has the highest resolution. And this resolution peak will "show through" the sensor, no matter what its own resolution is. If the sensor's resolution is high enough to make the difference between the optimal f-stop and one f-stop away from the optimum visible, then all higher-resolving sensors will show the optimum at the very same f-stop.

The only possible difference between two sensors might be the accuracy they nail down the lens' optimum. With a lower-resolving sensor we might be able to find the lens' optimum somewhere between, say, f/8 and f/11. With a higher-resolving sensor we might be able to find it between f/9 and f/10. That's all. If you're willing to call the difference between f/11 and f/10 the advantage of the higher-resolution sensor ... fine. But still the f-stop limit definitely is by far not linear to the pixel pitch, as Myhrvold has suggested.

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 11:44:40 am
Quote
Your MTF 50 figures are what is preposterous;

 I don't know exactly what the error is, or where it is, but there certainly is an error somewhere.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101651\")

Hardly preposterious, why don't you look at these references?


[a href=\"http://www.clarkvision.com/imagedetail/scandetail.html#diffraction]Roger Clark[/url]

Norman Koren (http://www.normankoren.com/Tutorials/MTF6.html#Diffraction)

Furthermore, these MTF calculations agree with what is observed in actual testing. For example, the above referenced test of the Canon D350, as summarized in the following table:

[attachment=1884:attachment]

The Nyquist frequency of this sensor is 78 lp/mm and at f/4.0 the system resolves at 91.5% of Nyquist with a contrast of 50%. As one stops down, the resolution of the system becomes diffraction limited and the system MTF50 tracks the theoretical MTF50 of the lens within reasonable limits. At f/32, the observed system resolution is 32 lp/mm at 50%, whereas the MTF 50 of the lens is 24 lp/mm. Resolution beyond the theoretical limit of the lens is due to sharpening artifact and limitations of the slanted edge method used by Imatest, as Norman Koren (http://forums.dpreview.com/forums/read.asp?forum=1021&message=16693479) explains. The MTF 50 best correlates with perceived image sharpness, but  if you want to use MTF 0 at the Dawes limit, that is OK with me.

It is very difficult to debate with a person who ignores established facts and does not follow a scientific approach, instead injecting subjective philosophical arguments into the discussion. Unless you come to your senses, I see no point in further discussion. It reminds me of the ancient Greek philosophers trying determine the number of teeth in a horse's mouth by deduction rather than actually looking into the animal's mouth.  

Bill
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 11:49:26 am
Quote
You're right. But exactly that is also where Norman explains my formula is a good first-order approximation. And that's all we need here in our context.

-- Olaf
[a href=\"index.php?act=findpost&pid=101669\"][{POST_SNAPBACK}][/a]

Your formula is a good first order approximation for MTF of 9% as with resolution tests with the USAF target, but it fails for MTF 50, where the Fourier transform and multiplication is necessary.

Really, I do not see why all this argument is necessary. All you have to do is look at the measured MTF 50 responses for various f/stops as shown in tests.

Bill
Title: f-stop limits for full sensor resolution
Post by: 01af on February 19, 2007, 12:09:01 pm
Quote
Your formula is a good [...], but it fails for MTF 50, where the Fourier transform and multiplication is necessary.
It doesn't fail; it just becomes less accurate ... but still a good-enough approximation.


Quote
Really, I do not see why all this argument is necessary. All you have to do is look at the measured MTF 50 responses for various f-stops as shown in tests.
Now this is really funny! Why don't you take a look at real images made through real lenses on real sensors? You remind me of the ancient Greek philosophers trying determine the number of teeth in a horse's mouth by deduction rather than actually looking into the animal's mouth.  

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 19, 2007, 12:22:47 pm
Quote
It reminds me of the ancient Greek philosophers trying determine the number of teeth in a horse's mouth by deduction rather than actually looking into the animal's mouth. 

And that's exactly what you are doing; citing the same formulas as the individual that sparked this thread, while ignoring a real-world example that clearly shows somebody is miscounting the horse's teeth. The hair detail in the spider shots I posted clearly show that 1-pixel-wide image features can be successfully resolved with a 1Ds at f/32. Those hairs are NOT aliasing artifacts. I've gone to the horse's mouth and actually counted the teeth. You're the one citing formulae without evaluating whether those formulae accurately predict or reflect real-world results.

What is your explanation for the results I show? You haven't even attempted to explain them, other than a quasi-ad hominem attack. If you are right, there shouldn't be any significant difference between the before and after crops, other than the sizing artifacts. The "before" shot clearly shows more detail than the one reduced to 2MP and back. If you want to criticize my conclusions, at least have the decency to point out the flaws in my methodology and explain why my examples (and the conclusions I draw from them) aren't meaningful or valid.
Title: f-stop limits for full sensor resolution
Post by: 01af on February 19, 2007, 01:05:28 pm
Quote
If you want to criticize my conclusions, at least have the decency to point out the flaws in my methodology and explain why my examples (and the conclusions I draw from them) aren't meaningful or valid.
Jonathan, don't hold your breath! He'll never do that. Because he can't. Because your example and your conclusions are perfectly valid (albeit your methodology---down- and upsizing to mimic a 2-MP shot---is not).

The resolution you got from the lens stopped down to f/32 sure is a bit lower than what you'd gotten from the same lens at f/11 or f/16. But it's by far not as low as a 2 MP image would be. Myhrvold's statement simply was an exaggeration.

Assuming an APS-C-sized 8 MP sensor and a lens that has its optimum aperture at f/16 (super-zoom lens at very close distance). Then stopping down beyond the optimum to f/32 will yield a resolution that comes close to a 4 MP sensor's, not a 2 MP one. You'll get the 2 MP figure when applying the "weakest-link theory" ... which however is wrong, as I explained in my first post.

And all this does not mean your photograph of the spider (Black Widow?) would have come out better at f/11 or f/16. Maximum sharpness would have been slightly better but depth-of-field not---and that would have a detrimental effect on that macro shot.

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: Paul Kay on February 19, 2007, 01:11:55 pm
Quote
Now this is really funny! Why don't you take a look at real images made through real lenses on real sensors?
[a href=\"index.php?act=findpost&pid=101676\"][{POST_SNAPBACK}][/a]

Forgive me for saying so, but generally this is what we did in those olden days of film, and it seemed to work ok! Unless there is an exceedingly good reason for mathematical analysis of resolution (ie for purely scientific or technical results which require this precise data) then I'm not sure what all the fuss is about. Although I trained as a scientific photographer/Photoscientist (many years ago I should add) I now use my own eyes and experience to determine whether the results that a particular lens/camera system produce are acceptable to me for my applications.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on February 19, 2007, 02:22:49 pm
Quote
The resolution you got from the lens stopped down to f/32 sure is a bit lower than what you'd gotten from the same lens at f/11 or f/16. But it's by far not as low as a 2 MP image would be. Myhrvold's statement simply was an exaggeration.

I've never claimed otherwise, and previously noted that I had to boost contrast and sharpen more agressively with this image than what is typical for an image shot at f/8 or so. It's the magnitude of the effect that I'm disputing, not the existence of the effect. We're in agreement here.

And yes, it is a black widow spider. There was a bumper crop of them one year where I lived, so I took advantage of the opportunity.
Title: f-stop limits for full sensor resolution
Post by: jani on February 19, 2007, 03:10:34 pm
Quote
I've never claimed otherwise, and previously noted that I had to boost contrast and sharpen more agressively with this image than what is typical for an image shot at f/8 or so. It's the magnitude of the effect that I'm disputing, not the existence of the effect. We're in agreement here.
It seems that there still is a misunderstanding regarding the use of MTF, though.

It also seems to me that you're confusing two different cameras; you're talking about a 1Ds (pixel pitch 8.8 microns), bjanes about the 1Ds MkII (pixel pitch 7.2 microns, different AA filter, too).

If you don't boost contrast and sharpen before you do the crop and resize, and if you do so according to the theoretical background that bjanes has put forward, what are your results then?

I'd also appreciate it if you would test with images with different contrasts, not just high-contrast ones. Then you can claim to be looking in the horse's mouth and not just theoretizing about what a horse's mouth looks like, from looking at a frog.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 03:30:34 pm
Quote
And that's exactly what you are doing; citing the same formulas as the individual that sparked this thread, while ignoring a real-world example that clearly shows somebody is miscounting the horse's teeth. The hair detail in the spider shots I posted clearly show that 1-pixel-wide image features can be successfully resolved with a 1Ds at f/32. Those hairs are NOT aliasing artifacts. I've gone to the horse's mouth and actually counted the teeth. You're the one citing formulae without evaluating whether those formulae accurately predict or reflect real-world results.

What is your explanation for the results I show? You haven't even attempted to explain them, other than a quasi-ad hominem attack. If you are right, there shouldn't be any significant difference between the before and after crops, other than the sizing artifacts. The "before" shot clearly shows more detail than the one reduced to 2MP and back. If you want to criticize my conclusions, at least have the decency to point out the flaws in my methodology and explain why my examples (and the conclusions I draw from them) aren't meaningful or valid.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=101677\")

I note that you have responded selectively, and omit anything about my MTF 50 figures being way off and affecting my analysis. What was your point?

The hairs on the spider's back can be seen in the image even though they are beyond the resolving power of the camera, just as a distant star (essentially a point source) can appear in the image. However, if two hairs on the spider's back were placed more closely than the camera could resolve, they would not be distinguishable in the image, just as a binary star beyond the resolving power of a telescope would not be resolved.

I have already agreed that the first of your images shows more detail than the second and that the 2 MP figure is wrong. My own calculations based on my MTF 50 calculations would be closer to 5.6 MP for your 16 MP camera at f/22 and 2.7 MP at f/32 (36% and 17% of base resolution respectively). Your statement that a decent lens at f/32 can still capture 11MP of image data with an MTF somewhere on the sunny side of 50% is completely without basis, and you are counting teeth by deduction rather than observation.  

Furthermore, you may have seen the tests reported on by [a href=\"http://query.nytimes.com/gst/fullpage.html?res=9507E1DD113FF93BA35751C0A9619C8B63]David Pogue[/url] in the NY Times. He used downsampling from 13 MP to 8 and 5 MP followed by printing with a Durst Lambda, which up-reses the image to 400 ppi for printing. Observers could see no differences in the print and he concluded that MP counts do not correlate well with perceived image quality. A pro photographer repeated his test with a 16.7-megapixel Canon EOS-1Ds Mark II and got the same results.

How can this be? The answer is that MTF at relatively low resolution determines the perceived sharpness of prints.

Bill
Title: f-stop limits for full sensor resolution
Post by: BJL on February 19, 2007, 03:30:45 pm
Quote
Generally, the optimal f-stop roughly correlates to image size (among other things, the most important being lens quality) ... but not to pixel count or pixel pitch.
[a href=\"index.php?act=findpost&pid=101659\"][{POST_SNAPBACK}][/a]

I think you are right that the resolution of the sensor (including film as a type of sensor) does not affect the f-stop at which resolution is optimal. That still leaves the question of how much resolution one gets at that optimal f-stop, or at any specified f-stop, where we agree that sensor resolution is relevant.

When DOF needs come into play, the optimal choice of f-stop can be considerably smaller than the values you talk about, which are based on optimizing the sharpness of images of subjects lying at the focal distance, ignoring OOF effects for subjects at other distances.

However, when one stops down for greater DOF, sensor resolution becomes even less significant in overall resolution, with diffraction and OOF effects becoming the main considerations at small enough apertures.

The main connection to sensor resolution is that, if a given aperture is chosen on the basis of best handling aberrations, diffraction and OOF effects, there is a resolution limit, no better than that due to diffraction alone, which means that increasing sensor resolution too much beyond diffraction based resolution has little benefit for overall resolution. My 1.4*N to 2*N estimate is thus mostly a guideline for that "maximum useful sensor resolution at a given aperture ratio."

For example, landscapes are one of the dominant examples of the most extreme resolution needs, and typically have DOF needs leading to use of apertures f/8, f/11 or smaller with 35mm. At such apertures, there is little benefit to reducing photo-site size much below about  4 to 5.6 microns, or about 27 to 54MP.

But it gets worse: when one wants to make use of new higher resolution of sensors 16MP and up, seeing that resolution requires a combination of larger prints and closer viewing than with lower resolution images, which makes OOF effects more noticeable, so that adequate sharpness across the desired DOF will tend to push the needed f-stop higher. Once one needs about f/14 or higher, for adequate DOF in the desired "big, closely viewed prints", diffraction and other effects limit useful pixel counts to about that 16MP.

Since landscapes are almost the dominant example of photography that can benefit from high pixel counts, this makes me wonder how much use there will be for pixel counts significantly higher that 16MP. (Larger sensors do not change these DOF/diffraction based limits on useful pixel counts at all, since higher f-stops are needed for equal DOF with the larger focal lengths of a larger format increasing the diffraction spot size in proportion to focal length).

I wonder how the artistic effect of landscapes and such is changed by the step from 16 or 22 up to 31 or 39 MP. I do not mean "print peeping" comparisons showing greater sharpness at the focal plane when using the same aperture with a higher pixel count, but the overall artistic impression, including more noticeable OOF effects when one looks closely enough to see that extra sharpness. Maybe it still works because
1. "Overall viewing", from far enough away to see the whole image, looks at least as sharp with fewer pixels, and probably with as much DOF.
2. Closer scrutiny shows some parts of the scene at least with more sharpness and detail, even though this benefit is limited to subjects only within a part of the "overall in-focus range".


And I suppose that other specialties like architectural images often benefit from high resolution without very much DOF.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 19, 2007, 03:42:07 pm
Quote
Jonathan, don't hold your breath! He'll never do that. Because he can't. Because your example and your conclusions are perfectly valid (albeit your methodology---down- and upsizing to mimic a 2-MP shot---is not).

The resolution you got from the lens stopped down to f/32 sure is a bit lower than what you'd gotten from the same lens at f/11 or f/16. But it's by far not as low as a 2 MP image would be. Myhrvold's statement simply was an exaggeration.

Assuming an APS-C-sized 8 MP sensor and a lens that has its optimum aperture at f/16 (super-zoom lens at very close distance). Then stopping down beyond the optimum to f/32 will yield a resolution that comes close to a 4 MP sensor's, not a 2 MP one. You'll get the 2 MP figure when applying the "weakest-link theory" ... which however is wrong, as I explained in my first post.


-- Olaf
[a href=\"index.php?act=findpost&pid=101684\"][{POST_SNAPBACK}][/a]

I have already posted my own photographic resolution tests with a Nikon D200 and the 50 mm f/1.8 lens (post #9 in this thread). The optimum system resolution (MTF 50) peaks at f/5.6 to f/8 and falls off appreciably at f/22. The lens does not go to f/32, but Photozone has published similar results for an APS sized Canon with a high grade macro lens.

No one is applying the "weakest link" theory here. You multiply MTFs and system resolution can and does exceed the weakest link. According to Norman Koren, your reciprocal method is not adequate for MTFs in the range of 50% and you have presented no evidence to the contrary, rather than stating that the approximation is "good enough". What a joke.

Bill
Title: f-stop limits for full sensor resolution
Post by: 01af on February 20, 2007, 01:48:25 pm
Quote
I have already posted my own photographic resolution tests with a Nikon D200 and the 50 mm f/1.8 lens (post #9 in this thread). The optimum system resolution (MTF 50) peaks at f/5.6 to f/8 and falls off appreciably at f/22.
So we can now rest reassured that your 50 mm standard lens basically behaves just like any other 50 mm standard lens.

And your point was?


Quote
No one is applying the "weakest link" theory here.
Myhrvold did.


Quote
... and system resolution can and does exceed the weakest link.
That's what I am saying.


Quote
According to Norman Koren, your reciprocal method is not adequate for MTFs in the range of 50 % and you have presented no evidence to the contrary ...
I am under the impression that you are trying to make a point which is not the point of this thread, is it?

Myhrvold stated that in order to fully exploit a sensor's resolving power, you must not stop down beyond the pixel pitch's diffraction limit. At larger apertures, you'll get the full resolution. At smaller apertures, you'll get a limited resolution that is not up to the pixel count's potential.

And I say he's wrong.

Assume a pixel size that, after Myhrvold, implies a diffraction limit at f/11. Also assume a fine lens that is diffraction-limited at f/5.6. According to Myhrvold, you can stop down to f/11 without losing image quality. But in fact the lens-sensor system will yield best performance at f/5.6 (the lens' limit); at f/8 you'll notice a slightly degraded image quality, and at f/11 degraded even more. The decreasing lens performance will show, even though the sensor has not yet reached his own "Myhrvold limit."

So Myhrvold's formula does not help to get the best-possible image quality out of a given camera. My advice is: Stop down no further than to the lens' diffraction limit, period ... no matter what the pixel size is.

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: Ray on February 20, 2007, 06:54:49 pm
I'm surprised that I'm the only one who sees the complete irrelevance to Myhrvold's article of Jonathan's Black Widow spider experiment. Jonathan has clearly misunderstood the point Myhrvold is making with his '2 megapixel camera' analogy.

I agree that Myhrvold is not as clear as one would like and the fact that Jonathan has misunderstood Myhrvold's point is not entirely Jonathan's fault.

However, if you read the whole article it should be clear that Myhrvold precedes his '2 megapixel camera' analogy with statements to the effect that f8 to f9 gets you the sharpest results with cameras such as the 1ds2 and 5D and that as you stop down you are effectively throwing away resolution to the point where at f22 you could have got the same result with a 2 megapixel camera but not necessarily with a 2 megapixel camera used at f22. Where has Myhrvold stated that?

When there's some doubt as to what is meant precisely, you might as well cut the guy some slack and consider the more sensible interpretation of what was meant rather than the less sensible interpretation.

Myhrvold is really trying to say that a 2 megapixel camera used at f8 will give you the same sharpness at the plane of focus as a 1Ds2 at f22. However, the fact that the DoF at f8 will be different seems to have escaped Myhrvold. This is no doubt due to the fact that Photography is a new field to him. He seems to be under the impression that stopping down beyond f8 reduces resolution in all parts of the image, as implied in the following quote from his article.

Quote
The bottom line is that the Canon EOS 1Ds Mark II will suffer lower resolution from diffraction at any f-stop above f/8, at which point both red and green light are at their diffraction limited resolution.

Another way to say this is that if you stop down below f/8 with this camera, you will reduce the resolution everywhere – including in the plane of perfect focus.

What Jonathan should do is take another couple of shots at f11 and f32, with his 1Ds, of scenes where DoF considerations are not significant (in order not to confuse the issue). Downsize the f11 shot to 2mp then upsize and compare to the f32 shot.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 21, 2007, 06:23:46 pm
Quote
I'm surprised that I'm the only one who sees the complete irrelevance to Myhrvold's article of Jonathan's Black Widow spider experiment. Jonathan has clearly misunderstood the point Myhrvold is making with his '2 megapixel camera' analogy.

I agree that Myhrvold is not as clear as one would like and the fact that Jonathan has misunderstood Myhrvold's point is not entirely Jonathan's fault.

However, if you read the whole article it should be clear that Myhrvold precedes his '2 megapixel camera' analogy with statements to the effect that f8 to f9 gets you the sharpest results with cameras such as the 1ds2 and 5D and that as you stop down you are effectively throwing away resolution to the point where at f22 you could have got the same result with a 2 megapixel camera but not necessarily with a 2 megapixel camera used at f22. Where has Myhrvold stated that?

When there's some doubt as to what is meant precisely, you might as well cut the guy some slack and consider the more sensible interpretation of what was meant rather than the less sensible interpretation.

Myhrvold is really trying to say that a 2 megapixel camera used at f8 will give you the same sharpness at the plane of focus as a 1Ds2 at f22. However, the fact that the DoF at f8 will be different seems to have escaped Myhrvold. This is no doubt due to the fact that Photography is a new field to him. He seems to be under the impression that stopping down beyond f8 reduces resolution in all parts of the image, as implied in the following quote from his article.
What Jonathan should do is take another couple of shots at f11 and f32, with his 1Ds, of scenes where DoF considerations are not significant (in order not to confuse the issue). Downsize the f11 shot to 2mp then upsize and compare to the f32 shot.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=102023\")

Ray,

I agree that we should cut Myhrvold some slack, since he probably knows more physics than the rest of us combined. I also agree that his 2MP figure was not well explained and is on the low side, but not entirely unreasonable.

In any case, people are deluding themselves when they think that their camera has an effective resolution of anywhere near that indicated by the nominal pixel count. Consider for example, the Canon 1DsM2, which is the current resolution champ in 35 mm style cameras with 16.6 MP.

Lets look at an Imatest MTF 50 plot of that camera (taken from the test shot published on DpReview):

[attachment=1903:attachment]

An ideal digital camera would resolve 0.5 cycles per pixel at Nyquist, but in the real world one gets about 0.33 cycles per pixel with sharpening. Without sharpening, the figure is worse. The 1DsM2 gets 0.344 cycles/pixel with standardized sharpening for MTF 50, which is a very good figure. In terms of idealized MTF 50 pixels, the effective megapixel count is 6.93 MP. This illustration shows how important sharpening is for MTF, and any analysis which does not take sharpening into account is incomplete.

Now lets look at the Nikon D200 with the 50 mm f/1.8 lens at f/5.6 (which is the optimum aperture in my tests):

[attachment=1904:attachment]

This camera resolves 0.324 cycles/pixel with standardized sharpening at MTF 50, for an idealized MTF megapixel count of 4.21.

Now, let's consider this combo at f/22:

[attachment=1905:attachment]

The resolution is now 0.193 cycles/pixel for an effective idealized MP count of 1.5 MP.  Stopping down to f/22 costs 64% of my resolution. Instead of MTF 50 idealized pixels one could use the Rayleigh criterion of 9% MTF, but the results would again show a dramatic loss of resolution. Jonathan cound take another picture of the spider, if the specimen is still available. However, in keeping with Lord Kelvin's quote, I think that quantitative analysis is appropriate.

Quote
So we can now rest reassured that your 50 mm standard lens basically behaves just like any other 50 mm standard lens.

[a href=\"index.php?act=findpost&pid=101950\"][{POST_SNAPBACK}][/a]

No, I did not really determine the behavior of my standard 50 mm lens, but rather the behavior of the camera and lens combination and quantititated the loss of resolution incurred by stopping down from f/5.6 to f/22. So far, all you have contributed is a lot of conjecture without any data.

For further explanation of the Imatest terms, please refer to Norman's [a href=\"http://www.imatest.com/]web site[/url]. The latest version of Imatest now contains an SQF (subjective quality factor) module, which has been shown in extensive tests at Kodak to correlate well with perceived image quality and when I upgrade I might show these results for 16 by 20 inch images at the two f/stops.

Bill
Title: f-stop limits for full sensor resolution
Post by: 01af on February 22, 2007, 10:20:19 am
Quote
I agree that we should cut Myhrvold some slack ...
"Some" slack!? You're kidding! Make that "a mountain of slack," then we'd get closer to the truth.

Myhrvold said, "Now, I don’t think anybody would be very excited about turning their EOS 1Ds Mark II, or Canon EOS 5D, or other full frame camera into a 2 megapixel camera." And that's pure nonsense.

It may (repeat: MAY) come closer to reality if he'd said, "At f/22, a full-frame DSLR has an effective resolution that corresponds to an ideal MTF-50 2 MP image." And that would have been a completely different statement (although one might still argue about the 2 MP figure). After all, a 2 MP camera (as opposed to a MTF-50 2 MP ideal) is subject to the same ideal-vs-real-world limitations as everything else is.

So, all your calculations of MTF-50 megapixel equivalents are very nice but pointless. It still remains a fact that a real-world 16 MP camera has a higher resolution than a real-world 10 MP camera has a higher resolution than a real-world 6 MP camera has a higher resolution than a real-world 2 MP camera. And a low-performing lens (be it due to poor quality or due to diffraction at too-small an aperture) tends to level these resolution differences but won't cancel them out completely ... provided the lens still offers at least some reasonable resolution.

Myhrvold also said, "we need to consider the pixel size. The formula is max f-stop = P x 1.054, where P = pixel size in microns". And that's nonsense, too.

This formula computes the f-stop where the lens' resolution limit starts to limit the output resolution more than the sensor's resolution limit does. But "more" does not mean the sensor's limit suddenly was irrelevant after that f-stop ... nor does it mean the lens' limit wasn't relevant before. So it does not (repeat: NOT) tell you at what f-stop you'll get the best-possible output resolution. The output resolution is always affected by both input resolutions ... only to varying degrees.


Quote
In any case, people are deluding themselves when they think that their camera has an effective resolution of anywhere near that indicated by the nominal pixel count. Consider for example the Canon EOS 1Ds Mk II ...
This is an absolutely pointless consideration. It doesn't matter what some freaks consider "the effective pixel count" by some obscure definition. The point is: with a good-enough lens the effective pixel count of  a 16 MP camera is twice the effective pixel count of an 8 MP camera, period. With a lesser lens, 16 MP effectively is not twice as much but still more than 8 MP effectively. The "Myhrvold f-stop" approximately indicates where the higher pixel count's advantage starts to become less than the quotient of the two pixel counts. Still the higher pixel count will always keep an advantage over the lower pixel count, even at f/22 or f/32.

Don't get me wrong: I do agree that the effective pixel count (whatever that may be exactly) always is somewhat less than the nominal pixel count. But please don't let scholastic arguments fool you into thinking higher pixel counts were no better than lower ones. Better take a good look into the horse's mouth  

-- Olaf
Title: f-stop limits for full sensor resolution
Post by: Ray on February 22, 2007, 10:36:46 am
Bill,
The test charts and imatest results are useful for predicting what we might expect from certain lenses and cameras, but we still need to check out the significance of these numbers in real world scenes.

For example, it's implied in a Roger Clark graph of diffraction limitation at various f stops, you posted a couple of pages back, that a lens diffraction limited at f8 should resolve 40 lp/mm (it looks about 40) at 80% MTF.

The best Canon lenses can manage only 70% MTF at 40 lp/mm. Medium quality lenses like the 100-400 IS at 400mm are at best 60% in the centre, falling significantly, away from the centre.

I would expect all good lenses to be equal at f22, but clearly not equal at f8.

It would be interesting to see some real world results comparing images at f8, downsampled to 2mp then upsampled back to the original size, with the same scene at f22. Just how close would such images be (at the focal plane)?

I'm disappointed that BJL has not taken up this challenge. He's written so much about the superiority of the Zuiko lenses. If any lens is likely to be diffraction limited at f8, then surely it's a Zuiko lens   .

ps. My excuse: I'm not at my studio at present.
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 22, 2007, 11:51:48 am
Quote
So, all your calculations of MTF-50 megapixel equivalents are very nice but pointless. It still remains a fact that a real-world 16 MP camera has a higher resolution than a real-world 10 MP camera has a higher resolution than a real-world 6 MP camera has a higher resolution than a real-world 2 MP camera. And a low-performing lens (be it due to poor quality or due to diffraction at too-small an aperture) tends to level these resolution differences but won't cancel them out completely ... provided the lens still offers at least some reasonable resolution.

Myhrvold also said, "we need to consider the pixel size. The formula is max f-stop = P x 1.054, where P = pixel size in microns". And that's nonsense, too.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=102370\")

I do not know your level of photographic sophistication, but I do note that you have made only a few posts on this forum. Many unsophisticated or beginning photographers fall into the [a href=\"http://query.nytimes.com/gst/fullpage.html?res=9507E1DD113FF93BA35751C0A9619C8B63]Megapixel Myth[/url]. Extra MP are nice since they allow bigger prints and some freedom in cropping. However, for prints that most of us make, the extra MP may not mean much. Furthermore, resolution varies with the square root of MP: a 16MP sensor has 1.4 times the resolution of an 8MP sensor, not double the resolution.

As an illustration I post the following SQF (http://www.bobatkins.com/photography/technical/mtf/mtf1.html) analyses for the D200 50 mm f/1.8 combo at f 5.6 and f/22 along with the Canon EOS 1DsM2. For details concerning the test method click here (http://www.imatest.com/docs/sqf.html).

D200, f/5.6
[attachment=1907:attachment]

D200, f/22
[attachment=1908:attachment]

Canon 1DsM2
[attachment=1909:attachment]

Interpretation criteria
[attachment=1910:attachment]

As is apparent from the graph, f/5.6 on the Nikon enables excellent print quality up to  a picture height of 34 cm or 13.23 inches. At f/22, one can get equivalent quality only by decreasing the picture height to 13 cm or about 5 inches. The Canon enables excellent quality up to 40 cm (16 inches). For smaller prints, the prints from the Canon might not show much advantage, since it takes a SQF difference of about 5 to be significant.

Quote
This formula computes the f-stop where the lens' resolution limit starts to limit the output resolution more than the sensor's resolution limit does. But "more" does not mean the sensor's limit suddenly was irrelevant after that f-stop ... nor does it mean the lens' limit wasn't relevant before. So it does not (repeat: NOT) tell you at what f-stop you'll get the best-possible output resolution. The output resolution is always affected by both input resolutions ... only to varying degrees.
This is an absolutely pointless consideration. It doesn't matter what some freaks consider "the effective pixel count" by some obscure definition. The point is: with a good-enough lens the effective pixel count of  a 16 MP camera is twice the effective pixel count of an 8 MP camera, period. With a lesser lens, 16 MP effectively is not twice as much but still more than 8 MP effectively. The "Myhrvold f-stop" approximately indicates where the higher pixel count's advantage starts to become less than the quotient of the two pixel counts. Still the higher pixel count will always keep an advantage over the lower pixel count, even at f/22 or f/32.

Don't get me wrong: I do agree that the effective pixel count (whatever that may be exactly) always is somewhat less than the nominal pixel count. But please don't let scholastic arguments fool you into thinking higher pixel counts were no better than lower ones. Better take a good look into the horse's mouth  
[a href=\"index.php?act=findpost&pid=102370\"][{POST_SNAPBACK}][/a]

I wouldn't consider Norman Koren to be a freak, but rather a very knowledgeable scientist and photographer. Let Myhrvold defend his statements, but the fact is that it is useful to consider the camera as sensor or lens limited. In practice, as shown by the Photozone tests on the Canon, stopping down the lens does not have a major effect on system resolution until the diffraction limit of the lens drops below the resolution of the sensor. Above that point, increased lens resolution makes a relatively minor change in system resolution.

It is always better to look into the horses mouth than make predictions from a scientific model, but that model has been thoroughly tested. Perhaps you can publish some images to show your point rather than employing intuition.

Bill
Title: f-stop limits for full sensor resolution
Post by: bjanes on February 22, 2007, 11:57:46 am
Quote
The test charts and imatest results are useful for predicting what we might expect from certain lenses and cameras, but we still need to check out the significance of these numbers in real world scenes.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=102023\")

Ray,

Something like Michael's megapixel shootout would be informative, but that would require everyone to get together, shoot the same scene, and use similar processing. Optimum sharpening is critical and would be hard to standardize since it depends on many factors, including the anti-aliasing filter that most 35 mm style cameras use. Even with all the efforts Micheal and his expert colleagues took, many still criticized the results. IMHO, the methodology was outstanding.

Quote
What Jonathan should do is take another couple of shots at f11 and f32, with his 1Ds, of scenes where DoF considerations are not significant (in order not to confuse the issue). Downsize the f11 shot to 2mp then upsize and compare to the f32 shot.
[a href=\"index.php?act=findpost&pid=102023\"][{POST_SNAPBACK}][/a]

I agree with your post that it would be useful for Jonathan to make the tests you suggest. The suggested method is similar to that used by [a href=\"http://query.nytimes.com/gst/fullpage.html?res=9507E1DD113FF93BA35751C0A9619C8B63]David Pogue[/url] with somewhat surprising results. Can you comment on that analysis?

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on February 22, 2007, 07:00:06 pm
Quote
The suggested method is similar to that used by David Pogue (http://query.nytimes.com/gst/fullpage.html?res=9507E1DD113FF93BA35751C0A9619C8B63) with somewhat surprising results. Can you comment on that analysis?
[a href=\"index.php?act=findpost&pid=102389\"][{POST_SNAPBACK}][/a]

Bill,
I did comment on that analysis when the article was first mentioned on this site in another thread.

The explanation for these results is that the general public at large is not particularly interested in matters of resolution and image sharpness. It's not something they give much thought to. I wasn't there, in the library where the 3 different shots from the 1Ds2 were shown. However, 3 observers were able to see the resolution differences between the 3 enlargements. To put it another way, 6% of the participants knew what to look for. The rest, quite possibly, didn't have much of a clue.

These results are consistent with the general public's attitude to the introduction of high definition television. When digital broadcasting was first implemented in Europe, there was no provision for HD broadcasts. It was thought (by people who had done some market research) that the general public was not sufficiently interested to spend the extra money on a new display that would (initially only partially) support HD broadcasts. Furthermore, it was thought that the smaller, more affordable HDTV sets, say 32" diagonal, would be too small to have any impact with the increased resolution.

Bearing in mind that an HDTV image has at least double the resolution of  standard definition (and much more than double if one is comparing the full spec of 1920x1080p), it is not surprising that some people, when asked to compare images that differ by considerably less than double the resolution, might have some difficulty.
Title: f-stop limits for full sensor resolution
Post by: Ray on February 23, 2007, 03:56:25 am
Another way of looking at this; 3 people out of 50 were able to tell the difference between the 7, 10 and 16.7mp images. But that presumably does not include David Pogue and Ellis Vener. Making the fairly safe assumption that the person who took the shots and the person who organised the tests were also able to discern the difference, that makes 5 out of 52 or 10%, since we can't have a fraction of a person.

10% of a group who are particularly discerning is fairly typical. For example, 10% of all computer users are Mac users, and just to make the point that I am not being elitist, I belong to the set of 90% non-discerning rabble who use the PC   .
Title: f-stop limits for full sensor resolution
Post by: xtoph on April 21, 2007, 05:36:19 pm
dude--i already _did_ what you suggest here. look up higher in the thread. the sample pics are posted; they prove conclusively that myrhvold's (sp?) statements (both the 2mp one and the gaussian blur one) are absolutely false. what more do you want? i've already given him the most charitable interpretation conceivable--his claims are complete nonsense (beyond the platitude that most lens-dslr combinations will perform at their peak resolution somewhere close to f/8--which we already know, and not for the reasons myrhvold says).

i am confused as to why this discussion kept on proceeding as though no one had bothered to do real-world tests.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 21, 2007, 10:06:29 pm
Quote
dude--i already _did_ what you suggest here. look up higher in the thread. the sample pics are posted; they prove conclusively that myrhvold's (sp?) statements (both the 2mp one and the gaussian blur one) are absolutely false. what more do you want? i've already given him the most charitable interpretation conceivable--his claims are complete nonsense (beyond the platitude that most lens-dslr combinations will perform at their peak resolution somewhere close to f/8--which we already know, and not for the reasons myrhvold says).

i am confused as to why this discussion kept on proceeding as though no one had bothered to do real-world tests.
[a href=\"index.php?act=findpost&pid=113576\"][{POST_SNAPBACK}][/a]

Are you talking to me, dude? If so, what point have I made that you are disputing?

I'm well aware that f16 is a perfectly usable f stop on my 5D, only marginally less sharp than f8 in real world scenes. F22, however, is a noticeable compromise. I rarely use it.

It's clear that Myhrvold is a theoretician. The reason he appears to be wrong is that he hasn't taken into consideration the failings, inadequacies and characteristics of the equipment we use. He's assuming that lenses and sensors are as perfect as his mathematical formulae and optical theories.

Jonathan made an easy experiment by downsampling a 1Ds image at f22 to 2mp and then upsampling back to its original size. I made the point it would have been more relevant if the image which was downsampled had been taken at f8 or f11, and compared with the same scene taken at f22 or f32. But even then, I suspect that the f22 shot would be marginally sharper and more detailed.

The problem as I see it is that no 35mm lens is diffraction limited at f8, whereas many 35mm lenses really are diffraction limited at f22. Additionally, if one did have a lens that is diffraction limited at f8, there is no 35mm sensor capable of capturing all the detail that such a lens would provide above say a 30% MTF cut-off point.

From a theoretical point of view, a lens which is diffraction limited at f8 will have 2.8x the resolution of that same lens at f22. But to capture such resolution at f8 you need 2.8^ x the numer of pixels, ie. 7.8x that you would need at f22.

So yes, Myhrvold is wrong in practice, but it's still useful to work out what he was trying to get at. The reasons why f16 with the 5D seems hardly less sharp than at f8, are (1) lenses are not as good as we think at f8. They are not diffraction limited. (2) the 5D does not have sufficient pixel density to reveal all the differences in lens resolution between these two f stops.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 23, 2007, 12:49:26 am
Quote
Are you talking to me, dude? If so, what point have I made that you are disputing?

The problem as I see it is that no 35mm lens is diffraction limited at f8, whereas many 35mm lenses really are diffraction limited at f22. Additionally, if one did have a lens that is diffraction limited at f8, there is no 35mm sensor capable of capturing all the detail that such a lens would provide above say a 30% MTF cut-off point.

[a href=\"index.php?act=findpost&pid=113594\"][{POST_SNAPBACK}][/a]

Ray,

Just a comment on the semantics of diffraction limited resolution.

Every optical lens now in existence is diffraction limited at all apertures--that is diffraction sets a limit on the maximum resolution possible at that aperture. In practice, the resolution limit is often imposed by aberrations rather than diffraction. When we say a lens is diffraction limited at f/8, we mean that aberrations are no longer the limiting factor at this aperture, and that the lens resolves at the limit set by diffraction for that aperture and for smaller apertures. Many current high quality lenses exhibit better resolution at f/5.6 than f/8. In this case, if the resolution at f/5.6 is limited only by diffraction, we say that the lens is diffraction limited at f/5.6.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on April 23, 2007, 02:05:01 am
Quote
Ray,

Just a comment on the semantics of diffraction limited resolution.

Every optical lens now in existence is diffraction limited at all apertures--that is diffraction sets a limit on the maximum resolution possible at that aperture. [a href=\"index.php?act=findpost&pid=113732\"][{POST_SNAPBACK}][/a]

Bill, if we are getting into semantics, I would say that definition serves no purpose other than to point out that there is a concept called diffraction which may be a limit to resolution at a particular f stop.

There are many technological factors which limit the resolution of lenses. Diffraction is one of them. But 'diffraction limitation', in my view, is a situation that arises only when all other abberrations have been reduced to a level that is less significant than the effects of diffraction. In other words, a situation where the only option to improve resolution is to reduce the effects of diffraction; a situation where further reduction in other lens aberrations would serve no purpose.

At the same time, it might well be the case that the finest 35mm prime lenses are fairly close to being diffraction limited at f8, at least at the centre. The Photodo ratings for lenses at f8 seem to vary from about 0.77 to 0.87.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 23, 2007, 10:28:21 am
Quote
Bill, if we are getting into semantics, I would say that definition serves no purpose other than to point out that there is a concept called diffraction which may be a limit to resolution at a particular f stop.

There are many technological factors which limit the resolution of lenses. Diffraction is one of them. But 'diffraction limitation', in my view, is a situation that arises only when all other abberrations have been reduced to a level that is less significant than the effects of diffraction. In other words, a situation where the only option to improve resolution is to reduce the effects of diffraction; a situation where further reduction in other lens aberrations would serve no purpose.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=113738\")

Ray,

I agree with your explanation of diffraction limited and it coincides with with what I was trying to get across.

Quote
At the same time, it might well be the case that the finest 35mm prime lenses are fairly close to being diffraction limited at f8, at least at the centre. The Photodo ratings for lenses at f8 seem to vary from about 0.77 to 0.87.
[a href=\"index.php?act=findpost&pid=113738\"][{POST_SNAPBACK}][/a]

Here is where we begin to differ. The new PhotoDo ratings use Imatest to derive their resolution figures, and measure the system resolution, not lens resolution. An idealized system would yield 0.5 cycles per pixel, but a Bayer array sensor can resolve only about 75% of this at 50% contrast, and I assume that the ratios of 0.77 to 0.87 that you mention are derived from this type of analysis.

For example in a recent test of the new [a href=\"http://www.photodo.com/topic_156.html]Nikon AFS 105 f/2.8 MicroNikkor[/url], maximum resolution in the center of the field occurred at f/4 (f/2.8 was not tested), and degrades with stopping down beyond f/4. In this test the lens with the D200 camera resolved about 0.475 cycles/pixel at f/4 (which is dubious). These data would imply that the lens is diffraction limited at f/4, and not the f/8 figure you mention.

Another way to look at resolution is in terms of line pairs/picture height as shown in the test of the same lens and camera at PhotoZone (http://www.photozone.de/8Reviews/lenses/nikkor_105_28vr/index.htm). These data show that maximum resolution in the center is reached at f/4 and that resolution is lost by stopping down further, in agreement with the PhotoDo results.

In both of these test cases, the relative contributions of lens versus sensor and demosiacing algorithms on the test results is not clearly determined.

One can also use the diffraction calculator devised by Sean McHugh (http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm) with stringent parameters for observer visual acuity and print size and determine whether the limits of diffraction are reached. For example,  the Nikon D200 camera used in the above Imatest reports is 10 MP with a CF of 1.5. Using 36 inches maximal print dimension and vision of 20/20, we see that the system is diffraction limited at apertures above f/1.4. Due to optical aberrations and limitations in the sensor and demosaicing process, degradation of the image is not observed until f/4 and significant degradation does not occur until considerably smaller apertures are used. This is true only for the center of the image; you may have to stop down more to improve resolution at the edges. The Nikon guru Thom Hogan states that diffraction begins to limit the D200 performance beyond f/13, and this is consistent with observations made by others.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on April 23, 2007, 11:00:40 pm
Quote
Here is where we begin to differ. The new PhotoDo ratings use Imatest to derive their resolution figures, and measure the system resolution, not lens resolution. An idealized system would yield 0.5 cycles per pixel, but a Bayer array sensor can resolve only about 75% of this at 50% contrast, and I assume that the ratios of 0.77 to 0.87 that you mention are derived from this type of analysis.
[a href=\"index.php?act=findpost&pid=113779\"][{POST_SNAPBACK}][/a]

I don't believe so, Bill. I usually refer to the old Photodo tests which I understand test the MTF of the lens only, in relation to 10, 20 and 40 lp/mm.

These tests are a few years old and it's a pity that Photodo has not continued with the same testing procedures. Their finest result is for a lens which is now discontinued, the Canon 200/1.8 which has an even higher score at f4 (0.90) than at f8 (0.87).

The difficulty I have here is related to a view that a lens is very unlikely to have less aberration at f4 than at f8 or any other larger f stop number. I don't know if this is always true, but it seems a reasonable assumption.

We can therefore deduce that the slightly higher performance at f4 (of this lens) is due to the fact that any increase in aberration at f4 is less than the reduction in diffraction effects at f4.

But we cannot deduce that the 200/1.8 is therefore diffraction limited at f8.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 24, 2007, 12:56:51 am
Quote
I don't believe so, Bill. I usually refer to the old Photodo tests which I understand test the MTF of the lens only, in relation to 10, 20 and 40 lp/mm.

These tests are a few years old and it's a pity that Photodo has not continued with the same testing procedures. Their finest result is for a lens which is now discontinued, the Canon 200/1.8 which has an even higher score at f4 (0.90) than at f8 (0.87).

The difficulty I have here is related to a view that a lens is very unlikely to have less aberration at f4 than at f8 or any other larger f stop number. I don't know if this is always true, but it seems a reasonable assumption.

We can therefore deduce that the slightly higher performance at f4 (of this lens) is due to the fact that any increase in aberration at f4 is less than the reduction in diffraction effects at f4.

But we cannot deduce that the 200/1.8 is therefore diffraction limited at f8.
[a href=\"index.php?act=findpost&pid=113917\"][{POST_SNAPBACK}][/a]

Ray,

Your use of the old PhotoDo figures clarifies things, but I think you are misinterpreting the data. The weighted MTF figures merely refer to the percent contrast at a given frequency (resolution), not resolution expressed as a fraction of the diffraction limited resolution. For example the theoretical diffraction limited resolution of a lens at f/2.0 is 160 lp/mm at 80% contrast (0.8 in the PhotoDo charts), 390 lp/mm at 50% contrast, and 820 lp/mm at Rayleigh about 9% contrast). At f/8 the resolutions are 40, 97, and 200 lp/mm respectively. Since the PhotoDo tests max out at 40 lp/mm, there is no way to determine the resolution of the 200/1.8 as compared to the theoretical values with the lens wide open.

At f/8 it the MTF at 80% contrast is about 30 lp/mm by eyeball estimation from the MTF chart, whereas the theoretical diffraction limit is 40 lp/mm, so the resolution is about 75% of theoretical, more or less in keeping with your figures. So, the lens is operating near the diffraction limit at 80% contrast. However, at MTF 50 (the value that correlates best with perceived sharpness) the PhotoDo analysis is not helpful, since testing at anywhere near the theoretical limit of 97 lp/mm was not done.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on April 24, 2007, 03:58:49 am
Quote
For example the theoretical diffraction limited resolution of a lens at f/2.0 is 160 lp/mm at 80% contrast (0.8 in the PhotoDo charts), 390 lp/mm at 50% contrast, and 820 lp/mm at Rayleigh about 9% contrast). At f/8 the resolutions are 40, 97, and 200 lp/mm respectively. Since the PhotoDo tests max out at 40 lp/mm, there is no way to determine the resolution of the 200/1.8 as compared to the theoretical values with the lens wide open.

Bill,
I agree it's not possible to precisely determine from the Photodo figures if the lens is diffraction limited at f8 unless we have the diffraction limited MTF responses at the same resolutions that Photodo have tested, which are 10, 20 and 40 lp/mm, but you have already provided a good indication of what to expect, from your usual sources (Roger Clark, Norman Koren??).

For example, you say a lens which is diffraction limited at f8 should be capable of 80% MTF at 40 lp/mm. Checking the Photodo MTF graphs (one doesn't have to rely upon weighted figures) we can examine the MTF response of what might be the finest lens that Canon ever made, the now discontinued 200/1.8 which at full aperture is as sharp as the Canon 28-135 at its sharpest focal length of 50mm and its sharpest aperture of f8 (weighted rating of 0.82 in both cases).

What do we get? Well, most lenses fall off in MTF response towards the edges, but if we consider just the central part of the image with a diameter of say 18mm, and if we average the sagital and meridional lines, we get an MTF response for 40 lp/mm of approximately 73%. Beyond 9mm from the centre, the MTF falls to about 65% at 18mm, and further beyond that

The conclusion is, the very expensive and probably uneconomical-to-produce 200/1.8 is very close to being diffraction limited at f8 in the central part of the image. Perhaps as close as matters.

Another fine lens, the Canon 50/1.4 doesn't quite make it in any part of the image, having an MTF response at 40lp/mm of just 70%. So I think we can say, on the basis of the accuracy of your figures, that the 50/1.4 is definitely not diffraction limited at f8, but is still reasonably close. The difference between 70% and 80% is probably not particularly significant. Also, the 50/1.4 is very consistent in this response right out to 18mm from the centre.

Lesser quality lenses and zooms rarely seem to rise as far as 70% at f8 in any part of the image.
Title: f-stop limits for full sensor resolution
Post by: BJL on April 24, 2007, 11:54:42 am
Ray,

   firstly, the idea that a lens, at any given aperture ratio, is either "diffraction limited" or "not diffraction limited" is a false dichotomy, and the attempt to declare than one particular f-stop (f/8 for almost all good lenses according to you it seems!) is the limit and is therefore the lowest completely "worthy" f-stop is trying to make things far more rigid than they really are. It is a cousin of Myrvold's mistake in apparently declaring that as soon as the pixel size is smaller than the diffraction spot size at a given aperture ratio, resolution is determined mostly or entirely by diffraction, so that decreasing pixel size at that aperture ratio will not increase resolution; contrary to clear experimental evidence.


At any aperture, the resolution of a lens is affected to some extent by both diffraction and by aberrations, with diffraction effects increasing in proportion to aperture ratio, while aberrations almost certainly decrease with increasing aperture ratio. At very high aperture ratios, resolution is determined almost entirely by diffraction, while at very low aperture ratios, the resolution of most or all lenses is probably determined almost entirely by aberrations.

But at the intermediate apertures where the vast majority of photographs are taken, like f/4, f/8 and even f/2.8 and f/11, the resolution of good lenses seems to be significantly affected by both diffraction and aberrations. The simple fact that even good lenses vary noticeably in resolution at f/8 and f/11 (as seen in MTF graphs) shows that aberrations are a significant factor in resolution. Only if the MTF graph of a lens at some aperture ratio was identical to the theoretical MTF graphs calculated from diffraction alone would that lens achieve your holy grail of being "purely diffraction limited" at that aperture ratio.

Rather than debate if and when one limit on lens resolution is dominant, why not just look at the practical bottom line: the angular resolution [see note] performance of the lens as a whole at various choices of aperture. Perhaps in relation to speed and DOF possible at that aperture. If a lens has significantly higher resolution at f/4 than it does at f/8, what is the significance of worrying about it being "not diffraction limited at f/4"?


Note on angular resolution
I prefer, like many astronomers and lens optics professionals, to think of resolution and MTF in terms of angular resolution, as in "cycles per radian" or "line per degree", and MTF at various levels of cycles per radian or such. This a fairly direct measure of the ability of the lens to resolve subject features of a certain size when at a certain distance from the subject, which is my bottom line for resolution in most situations.
It is roughly equivalent to "lines per picture height" when comparing lenses covering the same angular FOV but possibly of different focal lengths and producing different sized images at the focal plane.


P. S. Given that the vast majority of SLR's (and even a clear majority of professional SLRs) are now in formats smaller than 35mm, it puzzles me why you persist in looking only at 35mm format lenses in arguing for your favored f/8. Even with the 18.7x28.7mm format 1D series, the use of only part of the 35mm image circle somewhat invalidates the old PhotoDo measurements, because they to some extent weigh MTF measurements towards the edge of the 35mm format frame. In the now dominant SLR formats, it seems that even some good amateur level zoom lenses achieve peak resolution at f/4 or below, not just a few very expensive prime lenses like the old Canon 200/1.8.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 24, 2007, 07:52:32 pm
Quote
Note on angular resolution
I prefer, like many astronomers and lens optics professionals, to think of resolution and MTF in terms of angular resolution, as in "cycles per radian" or "line per degree", and MTF at various levels of cycles per radian or such. This a fairly direct measure of the ability of the lens to resolve subject features of a certain size when at a certain distance from the subject, which is my bottom line for resolution in most situations.
It is roughly equivalent to "lines per picture height" when comparing lenses covering the same angular FOV but possibly of different focal lengths and producing different sized images at the focal plane.
[a href=\"index.php?act=findpost&pid=114016\"][{POST_SNAPBACK}][/a]

BJL,
I get the impression you've just responded to my last post and haven't followed the developing arguments of the last few posts.

I'll recap for you.

(1) I've made the point that the differences in resolution as seen in real world shots at f8 and f16, with a reasonably good 35mm lens, do not seem nearly as great as one might think they would be, using a FF DSLR such as the 5D.

(3) I've suggested the reasons for this are that, whilst the lens at f16 is probably truly diffraction limited and therefore giving as good a resolution as is technically possible within the laws of physics, that same lens at f8 is probably quite far from being diffraction limited. Furthermore, a camera such as the 5D does not have even nearly sufficient pixel density to reveal the full extent of such differences.

(4) Bill has responded with some semantic issues regarding the definition of 'diffraction limitation' but agrees that the term should apply in situations (at f stops) where it is not possible to improve resolution by further reducing lens abberations in the lens design.

(5) Bill has also questioned the accuracy of my assertion that most 35mm lenses at f8 are not in fact diffraction limited. He seems to think they might be, but at the same time has provided the information that a lens should have an MTF response of 80% at 40 lp/mm at f8 if it is diffraction limited at that f stop.

(6) On the basis of the information that Bill has provided, and by examining the MTF curves at the Photodo site, I have made my case and proved my point, I believe   .

Now to your point. Comparing angular resolution or 'lines per picture height' is a valid alternative method and it's a method which would reduce the differences between f16 and f8 even further because f16 tends to produce a more even performance from edge to edge. With such a method, the humble Canon 50/1.4 would have at least as much angular resolution at f8 (from the appropriate closer distance) as that very expensive and discontinued 200/1.8.

The disadvantage to your method is, I simply don't have the information regarding angular resolution to work on. Where are the published results for angular resolution of the lens only? There are data at dpreview regarding lines per picture height, but these are lens/sensor combinations.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 25, 2007, 02:36:58 am
Quote
   firstly, the idea that a lens, at any given aperture ratio, is either "diffraction limited" or "not diffraction limited" is a false dichotomy, and the attempt to declare than one particular f-stop (f/8 for almost all good lenses according to you it seems!) is the limit and is therefore the lowest completely "worthy" f-stop is trying to make things far more rigid than they really are.

BJL,
Perhaps I should have answered this point first. We have to define terms clearly or we simply go round in circles. I realise there is a transition stage from the first hint of the effects of diffraction to the last hint of other aberrations which are eventually completely obscured by the effects of diffraction.

If you think it is necessary to break up the degree of diffraction limitation into separate descriptions such as 'very slightly diffraction limited', 'moderately diffraction limited', 'significantly diffraction limited', 'completley diffraction limited', then I have a problem with the semantics.

We all know what a 'limit' is. It's a 'stop'; a point beyond which one cannot go. It would be very silly to describe speed limits on the highway as, 'slightly limited to 60km/hour', 'significantly limited to 60km/hour', 'completley limited to 60km/hour etc. etc. Complete confusion would result.

My understanding is, rightly or wrongly, whilst diffraction may have an effect over a range of f stops, there will likely be one particular f stop at which (and beyond which) the effects of diffraction will limit the possibility of further resolution improvements in the design of the lens in question. At such a point, the lens is said to be diffraction limited.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 25, 2007, 09:47:27 pm
Quote
(4) Bill has responded with some semantic issues regarding the definition of 'diffraction limitation' but agrees that the term should apply in situations (at f stops) where it is not possible to improve resolution by further reducing lens abberations in the lens design.

(5) Bill has also questioned the accuracy of my assertion that most 35mm lenses at f8 are not in fact diffraction limited. He seems to think they might be, but at the same time has provided the information that a lens should have an MTF response of 80% at 40 lp/mm at f8 if it is diffraction limited at that f stop.

(6) On the basis of the information that Bill has provided, and by examining the MTF curves at the Photodo site, I have made my case and proved my point, I believe   
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=114108\")

Ray,

The PhotoDo original MTF charts are not conducive to  determining how the lens performs with respect to the diffraction limit. I made my best attempt previously. I saw a better method of plotting on the [a href=\"http://bobatkins.com/photography/technical/mtf/mtf2.html]Bob Atkins[/url] site. He seems pretty knowledgeable (even though he is somehow associated with Pop Photo). Note the charts from the old German paper.

Quote
The disadvantage to your method is, I simply don't have the information regarding angular resolution to work on. Where are the published results for angular resolution of the lens only? There are data at dpreview regarding lines per picture height, but these are lens/sensor combinations.
[a href=\"index.php?act=findpost&pid=114108\"][{POST_SNAPBACK}][/a]

With Imatest one can only determine lens/sensor resolution but here is a test with the EOS 1Ds MII with an 85 mm lens (I don't know which one since I downloaded the test shot from the DPReview web site) with a graphical comparison to Atkins' data for a diffraction limited f/8 lens. The test is without any sharpening.

For real world work with a given camera and lens, one is more interested in determining the optimum aperture for maximum resolution, and one can do this with Imatest. When the image starts to degrade with stopping down, I assume it is due to diffraction.

[attachment=2405:attachment]

Here is the original Imatest plot

[attachment=2374:attachment]
Title: f-stop limits for full sensor resolution
Post by: Ray on April 26, 2007, 12:56:31 am
Quote
The PhotoDo original MTF charts are not conducive to  determining how the lens performs with respect to the diffraction limit. I made my best attempt previously. I saw a better method of plotting on the Bob Atkins (http://bobatkins.com/photography/technical/mtf/mtf2.html) site.

Bill, what part of your explanation do you think I didn't understand?

I well appreciate the fact that those sagital and meridional lines can wander all over the place, but the fact remains, if it is true that a diffraction limited lens at f8 should theoretically have an MTF response of 80% at 40 lp/mm, but the measured MTF response of a particular lens at F8 and 40 lp/mm never reaches beyond say 70% and is well below 70% over most of the image circle, then we can confidently declare that such a lens is not diffraction limited at f8.

Why mince words? Why make things more complicated than they need be? Why introduce other factors such as SQF? You might as well introduce the effects of different sharpening algorithms and the different contrast requirements for big prints and small prints, all of which are other issues.

I note also in the Bob Atkins article you referred to that his definition of Diffraction Limitation seems to coincide with mine. I quote.

Quote
Real lenses rarely, if ever, come close to the theoretical maximum MTF at apertures below about f8 as stated above, though a few (expensive) lenses may be an exception to this rule. Performance at the maximum theoretical MTF is called "diffraction limited" performance, since diffraction is the reason why MTF falls with increasing spatial frequency, even for a "perfect" lens - thus diffraction ultimately limits the lens' performance.

I conclude from the above quote that Bob Atkins does not know of any lenses that are diffraction limited at apertures wider than f8, at the same time he allows for the possibility that some may exist.

Bob could have been more helpful if he had mentioned a few lenses that actually are diffraction limited at f8, since his statement appears to imply that such lenses do exist. A perusal of the old Photodo MTF charts will reveal there is at least one lens that reaches 80% MTF at f8 and 40 lp/mm, but only dead in the centre. From there on it's all down hill, so I would agree with BJL here that this does not constitute diffraction limitation.
Title: f-stop limits for full sensor resolution
Post by: BJL on April 26, 2007, 12:57:50 pm
Quote
My understanding is, rightly or wrongly, whilst diffraction may have an effect over a range of f stops, there will likely be one particular f stop at which (and beyond which) the effects of diffraction will limit the possibility of further resolution improvements in the design of the lens in question. At such a point, the lens is said to be diffraction limited.
[a href=\"index.php?act=findpost&pid=114150\"][{POST_SNAPBACK}][/a]
If that is the way that you want to use the words, then essentially you are saying that a lens is diffraction limited when its MTF curves are very close to what you would get from a theoretical calculation based on diffraction effects alone, assuming perfection as far as the "geometrical optics" part of lens performance (no aberrations at all). In that case, none of the MTF curves I have seen fit that, even at f/8 or f/11.

But I repeat my question: if the overall overall resolution keeps rising as one stops downs from f/11 to f/8 to f/5.6 and even to f/4, what is the possible practical significance of whether or not it is truly diffraction limited at f/8, or f/11? The only one I can see is telling me that it is futile to seek resolution improvement at that combination of focal length and aperture by upgrading to a better lens.

What is absolutely does not tell me is that this diffraction limited threshold dictates the maximum possible resolution. To think so would be in the spirit of Mhyrvold's error.

It is a common characteristic of design optimization that the best solution is one in which the several greatest imperfections are roughly in balance (like diffraction and aberration effects roughly equal), so that further reducing one requires increasing the other by a greater amount, making the overall result worse, not better.

It is instead a characteristic of dilettantish technical discussions to insist on an ideal of reducing one particular imperfection like aberrations, shadow noise, or camera weight as much as possible, or to completely insignificant levels, not matter how much other imperfections must be increased in the process.

(Lately, in this misguided over-simplification of seeking a single measure of technical excellence, one debate is whether resolution limits or shadow noise at high ISO is "the" single enemy that should be focussed on exclusively.)
Title: f-stop limits for full sensor resolution
Post by: Ray on April 26, 2007, 11:45:50 pm
Quote
If that is the way that you want to use the words, then essentially you are saying that a lens is diffraction limited when its MTF curves are very close to what you would get from a theoretical calculation based on diffraction effects alone, assuming perfection as far as the "geometrical optics" part of lens performance (no aberrations at all). In that case, none of the MTF curves I have seen fit that, even at f/8 or f/11.

BJL,
What other way is there to use the term, apart from the semantically absurd, "slightly diffraction limited', moderately 'diffraction limited' etc.?

Somewhere on Photodo's site there is (or was) an explanation as to why they did not test lenses at f11. The reason given was, the Photodo team considered all lenses equally bad at f11. Perhaps not to be taken literally, but I think we could interpret that as meaning the differences at f11 amongst all good lenses are so slight that the time, trouble and expense of carrying out MTF tests could not be justified.

My understanding of the term 'diffraction limitation' is not dependent on the condition that the lens should be absolutely free of all other aberrations, but that such aberrations are insignificant compared with the magnitude of the Airy discs.

A good analogy might be 'sensor resolution limitation'. When a sensor's resolution is limited by its pixel density, we do not assume that finer detail does not exist, ie. cannot be transmitted by the lens. We know that detail that is smaller than the pixel simply cannot be recorded by such a sensor. Likewise, aberrations that are smaller than the Airy discs, or that are swamped and effectively completely obscured by the effects of the Airy discs, may still exist, but are not relevant with regard to the recorded image. Those are the conditions for diffraction limitation.

Quote
But I repeat my question: if the overall overall resolution keeps rising as one stops downs from f/11 to f/8 to f/5.6 and even to f/4, what is the possible practical significance of whether or not it is truly diffraction limited at f/8, or f/11? The only one I can see is telling me that it is futile to seek resolution improvement at that combination of focal length and aperture by upgrading to a better lens.

It matters in proportion to how closely the resolution at f8 reaches that diffraction limited ideal. I get the impression that the lenses that are closest to being diffraction limited at f8 are the ones that deliver even greater resolution as one stops up to f5.6 and f4, but it's not always the case. You should know better than I do that lenses can be optimised for sharpest results at a particular aperture.

An example would be the Canon 400/2.8 II USM. This lens (or at least the copy that Photodo tested) is sharpest at full aperture of f2.8. Wow! And this is not just sharp in the centre but extends almost to the very edges. The lens has remarkably similar performance at f2.8 to the 50/1.4 at f8.

Unfortunately, this 400/2.8 does not have IS. The IS version does not come near to matching this performance at full aperture, but both lenses are about the same at f8. What does it matter? If you owned both of these lenses would you not experience a degree of indecision   .

Let's suppose the 400/2.8 IS version was actually diffraction limited at f8. It would be noticeably sharper at f8 than the non-IS lens at full aperture. The IS would largely compensate for the slower shutter speed at f8, and if not, an extra stop of ISO would. The images at f8 would be noticeably sharper at the point of focus. The focus point would not be quite as critical because of the greater DoF, yet the DoF would still be fairly shallow if that quality was desired.

Quote
What is absolutely does not tell me is that this diffraction limited threshold dictates the maximum possible resolution. To think so would be in the spirit of Mhyrvold's error.

Of course not. Where have I implied that? The diffraction limited threshold dictates the maximum possible resolution at a specified f stop.

Quote
It is a common characteristic of design optimization that the best solution is one in which the several greatest imperfections are roughly in balance (like diffraction and aberration effects roughly equal), so that further reducing one requires increasing the other by a greater amount, making the overall result worse, not better.

It is instead a characteristic of dilettantish technical discussions to insist on an ideal of reducing one particular imperfection like aberrations, shadow noise, or camera weight as much as possible, or to completely insignificant levels, not matter how much other imperfections must be increased in the process.

Get off your soap box, BJL   . I merely make the point that the best 35mm lenses cannot quite manage diffraction limitation at f8. Lesser quality 35mm lenses are not even close.

Quote
(Lately, in this misguided over-simplification of seeking a single measure of technical excellence, one debate is whether resolution limits or shadow noise at high ISO is "the" single enemy that should be focussed on exclusively.)

Since my views are supported by the hundreds of MTF measurements that are shown on Photodo's MTF charts, I don't see how you can accuse me of seeking a single measure. But the single (weighted) measure can be a useful, quick summary.

The development of high resolution sensors with low shadow noise is definitely the way forward, but lenses have to keep up. Ever since the introduction of the fairly low pixel density 1Ds, photographers have been grumbling about the inadequacy of lenses, so much so that some people seem to have arrived at the erroneous conclusion that further pixel density will serve no purpose.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 27, 2007, 07:34:02 am
Quote
Bill, what part of your explanation do you think I didn't understand?
[a href=\"index.php?act=findpost&pid=114297\"][{POST_SNAPBACK}][/a]

After a bit of discussion, I think we are pretty much on the same wave length, and I merely thought you might be interested in the information.
Quote
Why mince words? Why make things more complicated than they need be? Why introduce other factors such as SQF? You might as well introduce the effects of different sharpening algorithms and the different contrast requirements for big prints and small prints, all of which are other issues.

I note also in the Bob Atkins article you referred to that his definition of Diffraction Limitation seems to coincide with mine. I quote.

[a href=\"index.php?act=findpost&pid=114297\"][{POST_SNAPBACK}][/a]

SQF takes human perception into account. The eye is most sensitive to contrast in the frequency range of 3-12 cycles per degree (BJL would be pleased) rather than at its maximum resolution of about 1 minute of arc, and this sweet spot translates to 0.5-2.0 cycles per mm on a print viewed at 34 cm. This correlates to 4-16 cycles/mm on a 35 mm negative or full frame 35mm sensor with an 8x10 inch print (8x magnification). A high MTF in this range is more important than that at 40 cycles/mm. Since you like big prints, the important frequency range for a 16x20 inch print would be 8-32 cycles/mm. SQF is merely the area under the MTF curve between the critical frequencies.

Yes, I think that we are all using Diffraction Limited in the same way.
Title: f-stop limits for full sensor resolution
Post by: BJL on April 27, 2007, 03:00:28 pm
Ray, there are too many point for me to reply to them all now, so let me be selective. And do not misunderstand; I am happy to work with your definition of diffraction limited.

Quote
1. Somewhere on Photodo's site there is (or was) an explanation as to why they did not test lenses at f11. The reason given was, the Photodo team considered all lenses equally bad at f11.

2. It matters in proportion to how closely the resolution at f8 reaches that diffraction limited ideal.

3. Since my views are supported by the hundreds of MTF measurements that are shown on Photodo's MTF charts, I don't see how you can accuse me of seeking a single measure. But the single (weighted) measure can be a useful, quick summary.
[a href=\"index.php?act=findpost&pid=114482\"][{POST_SNAPBACK}][/a]

1. That suggest that the (mostly 35mm format) lenses they tested are indeed more or less diffraction limited. The f/11 curve I have seen are for larger format lenses, and definitely showed aberration effects.

2. But what is the practical importance of that? If the goal is to compare to other lenses the MTF curves at various apertures will do the same job more directly. Intermediate questions about being diffraction limited do not seem to add to the final judgement.

3. Are you playing dumb Ray? I said single measure, not single measurement. The "single measure" I refer to is something like overall effect of aberration on resolution, and a one dimensional ideal of "absence of significant aberration so that resolution is detemined mostly by diffraction". Not to any one measurement of MTF.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 27, 2007, 08:55:19 pm
Quote
This correlates to 4-16 cycles/mm on a 35 mm negative or full frame 35mm sensor with an 8x10 inch print (8x magnification). A high MTF in this range is more important than that at 40 cycles/mm. Since you like big prints, the important frequency range for a 16x20 inch print would be 6-24 cycles/mm. SQF is merely the area under the MTF curve between the critical frequencies.

[a href=\"index.php?act=findpost&pid=114519\"][{POST_SNAPBACK}][/a]

Bill,
How do we get from 4-16 lp/mm at 8x enlargement, to 6-24 lp/mm at 16x enlargement? Shouldn't that be 8-32 lp/mm for a 16x20 print?

Since I have a 24" wide printer I occasionally make 22x33" prints from a single, uncropped 5D image, in which case the critical resolutions for SQF should be 11-44 lp/mm, resolutions (approx) which the Photodo MTF charts address.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 27, 2007, 09:46:01 pm
Quote
1. That suggest that the (mostly 35mm format) lenses they tested are indeed more or less diffraction limited.

BJL.
I presume you mean more or less diffraction limited at f11. The conclusion I reach is that Photodo does not test lenses at their diffraction limit, but one stop below it so we can see how closely the lenses reach that diffraction limited ideal.

Quote
3. Are you playing dumb Ray? I said single measure, not single measurement. The "single measure" I refer to is something like overall effect of aberration on resolution, and a one dimensional ideal of "absence of significant aberration so that resolution is detemined mostly by diffraction". Not to any one measurement of MTF.

You are falling into obscurantism, BJL. The overall effect of aberrations on resolution at any given f stop is the only variable that determines resolution. Diffraction is an invariable property of the aperture chosen. It's in the very fabric of the optics. It is non-negotiable.

The Photodo MTF charts are all about the issue of how successfully such aberrations have been minimised at apertures of f8 and wider. Presumably, if that exceptionally fine Canon 200/1.8 had had a fairly flat MTF response at f8, close to 80% at 40 lp/mm right out to the edges, it would have got a rating of 5.5 out of 5 and Photodo would have had to revise its scales.
Title: f-stop limits for full sensor resolution
Post by: BJL on April 28, 2007, 04:04:10 pm
Ray, indeed I meant at f/11.

Of course at equal f-stop, diffraction is equal, but in practice choices often need to be made between different apertures, for the sake of best overall resolution/sharpness depending on
- diffraction (better at larger apertures)
- aberration effects (better at smaller apertures almost certainly, but in a way that is less predictable)
- out of focus effects on sharpness (better at smaller apertures: I am talking specifically about image resolution only, not artistic blurring of backgrounds and such.)
As soon as such balancing is needed, it is pointless to focus on reducing one source of imperfection to negligible levels.

Since you have still offered no useful answer to my question, let me restate it again.

What is the practical relevance of knowing "at which f-stops a lens's resolution is determined mostly by diffraction", beyond what is revealed more directly and in more detail by overall measurements of resolution such as MTF charts or l/mm at which MTF is 50%?

- It it not useful when choosing between different f-stops for the sake of optimal resolution: just looking at which f-stop gives the best resolution answers that question, even if at that optimal f-stop, aberration is significant.

- It is not needed and not particularly useful when comparing the resolution of two lenses: again directly examining their resolution at various f-stops is a more direct and informative way to do that.

- But maybe is it useful in predicting how far it is worth increasing sensor resolution (in 35mmFF say), but even then overall MTF data is probably more useful.

For example, if few or no 35mm lenses can offer resolution better than can be attained at f/8 with a diffraction limited lens, then lenses limit resolution more than 35mmFF sensors once pixel spacing is below about 8 microns, suggesting that resolution gains in 35mm format will be rather limited if pixel spacing moves much below the current minimum of 7.2 microns in the 1DSMkII.

(Smaller format sensors using the same lenses are exempted, since the PhotoDo MTF measurements you cite average over various parts of the 35mm frame, so average MTF could be significantly higher over only the smaller frames of most DSLR's that use 35mm format lenses.)


This makes me wonder how you expect your "not diffraction limited at less than f/8" lenses will look when used on an imagined 35mm format DSLR with the 5.5 micron pixel spacing of the Nikon D2X sensor.

I wildly predict that we will never see substantially more than about "20MP worth" of resolution in 35mm format, due to you bad news about 35mm format lens resolution , except perhaps in very expensive, low volume bodies only worth using with a collection of mostly very expensive and as yet non-existent super-lenses. (Higher pixel counts might arise for other purposes, like cropping to the sufficiently sharp part of the frame for extra telephoto/macro reach, or to help reduce moire by "oversampling".)
Title: f-stop limits for full sensor resolution
Post by: Ray on April 28, 2007, 09:08:58 pm
Quote
Since you have still offered no useful answer to my question, let me restate it again.

What is the practical relevance of knowing "at which f-stops a lens's resolution is determined mostly by diffraction", beyond what is revealed more directly and in more detail by overall measurements of resolution such as MTF charts or l/mm at which MTF is 50%?

BJL,
Knowing that a lens is diffraction limited at a particular f stop gives one the confidence that no lens can be better, at that f stop. Photodo's statement that all (35mm) lenses are equally bad at f11 is not actually true, I'm sure you'll agree.

There are some zooms that at certain focal lengths are actually sharpest at f16, at which aperture such a lens probably is diffraction limited. If I had such a zoom, I'd definitely like to know that the lens at that particular FL is not diffraction limited at f11, wouldn't you?

Of course, if you already have possession of all the measurements at all the f stops you are likely to use, either in terms of Photodo type MTF charts, or lp/mm at 50% MTF, then knowing which f stop is diffraction limited is of more academic interest, but perhaps with some very pratical economic implications.

For example, if I know that a particular lens which I use frequently at f11 is actually diffraction limited at f11 pretty much out to the edges of the frame, then I can stop looking around for a better lens, and possibly save some money   .

Quote
For example, if few or no 35mm lenses can offer resolution better than can be attained at f/8 with a diffraction limited lens, then lenses limit resolution more than 35mmFF sensors once pixel spacing is below about 8 microns, suggesting that resolution gains in 35mm format will be rather limited if pixel spacing moves much below the current minimum of 7.2 microns in the 1DSMkII.

You've lost me here, BJL. If it's true that a lens, diffraction limited at f8, can deliver 97 lp/mm at 50% MTF, then such a lens would probably (but not necessarily) improve upon that figure at wider apertures. Canon's current highest resolving DSLR, the 400D with 5.5 micron spacing only resolves around 61 lp/mm (at presumably 10% MTF), according to dpreviews 'lines per picture height tests'. That's a far cry from 97 lp/mm at 50% MTF and an even further cry from the 120 lp/mm that we could expect at f4 from such a lens which is diffraction limited at f8.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 29, 2007, 05:30:27 am
I always like to back up my theoretical pronouncements with some hard visual facts, if possible, so I took the trouble to test my 50/1.4 at f8, f11, f16 and f22 on my own test board consisting of Norman Koren line charts plus various textured surfaces stuck on a piece of plywood.

I was already aware that the resolution differences between f8 and f16 with a zoom lens like the 24-105 on my 5D were too subtle to bother with, for landscapes.

What surprised me was just how similar the differences were with the 20D, which is in effect a cropped 22mp FF camera.

Here is the comparison. All shots were taken on tripod, with remote release and mirror lock-up (would I do anything less?)

[attachment=2394:attachment]

To my eyes, there's no difference between f8 and f11. At f16 there's a very marginal fall in clarity but still not a major concern. The softness at f22 is pretty obvious. I wouldn't use this f stop unless maximum DoF was paramount.

So what conclusions can we derive from these results? Let's examine the unknowns as well as the knowns.

(1) The 50/1.4 is not diffraction limited at its sharpest f stop of f8. There's room for improvement.

(2) The 50/1.4 might be diffraction limited at f11, but if it is, a 22mp FF sensor is not able to tell us.

Conclusion: A 22mp FF sensor has not even sufficient pixel density to capture the resolution potential of existing 35mm lenses, never mind future lenses.
Title: f-stop limits for full sensor resolution
Post by: bjanes on April 30, 2007, 03:10:19 pm
Quote
I always like to back up my theoretical pronouncements with some hard visual facts, if possible, so I took the trouble to test my 50/1.4 at f8, f11, f16 and f22 on my own test board consisting of Norman Koren line charts plus various textured surfaces stuck on a piece of plywood.

What surprised me was just how similar the differences were with the 20D, which is in effect a cropped 22mp FF camera.

Here is the comparison. All shots were taken on tripod, with remote release and mirror lock-up (would I do anything less?)

To my eyes, there's no difference between f8 and f11. At f16 there's a very marginal fall in clarity but still not a major concern. The softness at f22 is pretty obvious. I wouldn't use this f stop unless maximum DoF was paramount.

So what conclusions can we derive from these results? Let's examine the unknowns as well as the knowns.

(1) The 50/1.4 is not diffraction limited at its sharpest f stop of f8. There's room for improvement.

(2) The 50/1.4 might be diffraction limited at f11, but if it is, a 22mp FF sensor is not able to tell us.

Conclusion: A 22mp FF sensor has not even sufficient pixel density to capture the resolution potential of existing 35mm lenses, never mind future lenses.
[a href=\"index.php?act=findpost&pid=114830\"][{POST_SNAPBACK}][/a]

Ray,

I'm happy that people were not reading my posts carefully, since in my previous post of the Imatest results with the Canon EOS 1Ds MII I inadvertently forgot to convert from lines/mm to line pairs/mm and the results looked much closer to ideal than they should. The Nyquist limit of the 1Ds MII is 69.3 lp/mm. I reposted the correct results in the original post and am repeating the chart here. As you correctly observe, the system resolution does not approach the diffraction limited resolution of the lens, but nonetheless, the results are quite good and Michael frequently talks about his sensor out resolving the lens.  How than this be?

[attachment=2407:attachment]

I think the answer if found in consideration of the SQF data and concept. According to SQF, the eye is most sensitive to contrast in the frequency range of 3-12 cycles per degree as seen on the print as viewed from the observation distance. For an 8 by 10 inch print, the critical resolutions are from 4 to 16 lp/mm on a 35 mm negative or with a full frame sensor (8 times enlargement) and 8 to 32 lp/mm for a 16 by 20 inch enlargement. With proper sharpening, the MTF of the camera is quite good in this range as demonstrated by the Imatest SQF plot. Higher frequencies are not resolved by the camera, but these are not critical for perceived image sharpness.

[attachment=2406:attachment]

In your tests with Norman Koran's resolution chart, the MTF in these critical frequencies is not readily determined by examination with the naked eye. I downloaded your images and used PixelProfile as suggested by Mr. Koren to estimate the MTF in the region of 40 to 100 lp/mm in the test images.

[attachment=2408:attachment]

[attachment=2409:attachment]

As is apparent from inspection of the test shots, there is only aliasing beyond the Nyquist limit of 78 lp/mm, but MTF resolution at f/22 in the important lower frequencies is limited.
Title: f-stop limits for full sensor resolution
Post by: Ray on April 30, 2007, 10:23:31 pm
Quote
As you correctly observe, the system resolution does not approach the diffraction limited resolution of the lens, but nonetheless, the results are quite good and Michael frequently talks about his sensor out resolving the lens.  How than this be?

Bill,
As you know, we can't really expect the system resolution to approach the diffraction limited resolution of the lens because system resolution is always some sort of product of both sensor resolution and lens resolution, the result always being less than either sensor or lens resolution taken separately. This is why I prefer MTF charts of lens-only performance. I buy my lenses separately to the camera body. I use the same lenses on different formats with significantly different pixel densities and I hope to continue using such lenses with future DSLR models. I therefore want to know the performance of the actual lens itself, not just a diluted performance in conjunction with a particular DSLR.

This concept of sensor outresolving lens should really be defined, as all terms should be. The only sensible definition I can think of is when the sensor (or film), when tested separate from the lens, delivers better specs than the lens.

For example, if a lens can resolve 40 lp/mm at 70% MTF but the sensor can resolve 40 lp/mm at 80% MTF, then there's a reasonable case to be made that the sensor is outresolving the lens.

I always remember the specs of T-Max 100 B&W film because the MTF response was so amazing compared with color film, being 100% up to 50 lp/mm and 60% at 100 lp/mm (according to Kodak). No lens can match this, so I think it would be fair to say that T-Max 100 really does (did) outresolve the lens (except for that bloody grain   ) .

As far as I can see from the available evidence, there are no DSLRs which can outresolve 'good' prime lenses in the central area of the image circle, which is roughly covered by the cropped formats.

The grey area is between the edges of the cropped format and the edges of the FF sensor. In this area of the image circle, one could say that all of Canon's FF DSLRs are outresolving all but the best lenses.

Quote
In your tests with Norman Koran's resolution chart, the MTF in these critical frequencies is not readily determined by examination with the naked eye. I downloaded your images and used PixelProfile as suggested by Mr. Koren to estimate the MTF in the region of 40 to 100 lp/mm in the test images.

Bill, I have a problem with this approach. In one breath you are talking about SQF and in another your are expounding on the opposite. What cannot be determined with the naked eye is surely only of academic interest. If the naked eye cannot determine relevant differences in a 200% enlargement on screen, which represents a huge print, probably about 6ft x 9ft (I haven't calculated), then why should we bother?

Quote
As is apparent from inspection of the test shots, there is only aliasing beyond the Nyquist limit of 78 lp/mm, but MTF resolution at f/22 in the important lower frequencies is limited.

Only aliasing beyond the Nyquist limit? How can that be? I thought it was only the Foveon type sensor which could reach resolution to the Nyquist limit? This is the explanation for a 3.3MP Foveon sensor equalling the resolution of a 6mp Bayer type sensor.
Title: f-stop limits for full sensor resolution
Post by: BJL on May 01, 2007, 12:15:25 am
Quote
Knowing that a lens is diffraction limited at a particular f stop gives one the confidence that no lens can be better, at that f stop.
[a href=\"index.php?act=findpost&pid=114792\"][{POST_SNAPBACK}][/a]
If you consider having such confidence to be of practical importance, then fine.

To refine my comments about lens resolution limits on the pixel sizes that will be worthwhile in 35mm format:

Firstly, I was talking about what I have seen (and heard from you) about actual lenses, not hypothetical ones that are purely diffraction limited at f/8.

Mainly, my rough reading of MTF curves is that for lenses other than very good and very long telephotos, the fall-off from center to edge is greater than the improvement on-axis in going from f/8 to any smaller f-stop. If so, then off-center performance at any f-stop is less than on axis performance at f/8 and so in turn worse than the limits set by diffraction alone at f/8. So once one cares about sharpness across a significant fraction of the 35mm frame (not, say, just with an "APS-C" crop), the lens resolution limit is probably not a lot less than around what 8 micron pixel spacing gives. At a very rough estimate, I doubt that the 5.5 micron pixel spacing of the D2Xs will ever have much use in 35mm format, and even less so the 4.7 micron spacing of the new Panasonic nMOS 10MP 4/3" sensors.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 01, 2007, 07:44:04 am
Quote
This concept of sensor outresolving lens should really be defined, as all terms should be. The only sensible definition I can think of is when the sensor (or film), when tested separate from the lens, delivers better specs than the lens.

For example, if a lens can resolve 40 lp/mm at 70% MTF but the sensor can resolve 40 lp/mm at 80% MTF, then there's a reasonable case to be made that the sensor is outresolving the lens.
[a href=\"index.php?act=findpost&pid=115114\"][{POST_SNAPBACK}][/a]
Ray,

It's not that simple. In your example you mention 40 lp/mm as a reference point, but other frequencies also come into play. As you know, MTF curves for lenses start out at 100% and fall to zero. What happens between these extremes is important. MTF at 10 lp/mm (good contrast) is often considered more important in 35 mm photography than at 40 lp/mm (resolution).

By the process of Reductio ad absurdum , with your line of reasoning, one could say that lens choice is not important for digital cameras since even a cheap lens will out resolve the camera. However, a good lens will often give better contrast at the important lower frequencies.

Quote
Bill, I have a problem with this approach. In one breath you are talking about SQF and in another your are expounding on the opposite. What cannot be determined with the naked eye is surely only of academic interest. If the naked eye cannot determine relevant differences in a 200% enlargement on screen, which represents a huge print, probably about 6ft x 9ft (I haven't calculated), then why should we bother?
[a href=\"index.php?act=findpost&pid=115114\"][{POST_SNAPBACK}][/a]
MTF and SQF complement each other. Whenever you reduce a complicated measurement to one number, you are losing information in the process of simplifying things. However, both are relatively objective and reproducible. In contrast, the naked eye is highly non-reproducible and subjective. Remember David Pogue's essay in the  New York Times where observers could not tell a 6 from a 22 MP image?

Your pixel peeping approach determines detail at high frequency, but not at the important lower frequencies. To evaluate the lower frequencies, one must stand back and inspect the overall image.

Quote
Only aliasing beyond the Nyquist limit? How can that be? I thought it was only the Foveon type sensor which could reach resolution to the Nyquist limit? This is the explanation for a 3.3MP Foveon sensor equalling the resolution of a 6mp Bayer type sensor.
[a href=\"index.php?act=findpost&pid=115114\"][{POST_SNAPBACK}][/a]

Bayer array sensors resolve about 70-80 percent of the Nyquist limit at low contrast, as can be seen by looking at the resolution tests on DPReview. However, as Norman Koren states, it does not make sense to measure resolution at the point where detail disappears. The DPReview tests are not informative: one can always say 75% of Nyquist without even looking at the test results. Aliasing will occur only above the Nyquist limit and is manifested by false detail. At and slightly below Nyquist, the Bayer sensor will show no useful detail.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 01, 2007, 11:46:11 am
Quote
Firstly, I was talking about what I have seen (and heard from you) about actual lenses, not hypothetical ones that are purely diffraction limited at f/8.

Mainly, my rough reading of MTF curves is that for lenses other than very good and very long telephotos, the fall-off from center to edge is greater than the improvement on-axis in going from f/8 to any smaller f-stop. If so, then off-center performance at any f-stop is less than on axis performance at f/8 and so in turn worse than the limits set by diffraction alone at f/8. [a href=\"index.php?act=findpost&pid=115126\"][{POST_SNAPBACK}][/a]

BJL,
The problem here is that there's no information on the performance of Canon lenses at 60 & 70 lp/mm, the sorts of resolutions that FF sensors of 22mp and beyond should be able to record if the MTF at such resolutions is high enough.

However, there are a few lenses ranging from 50mm to 400mm that can resolve 40 lp/mm at 70% almost to the corners, according to Photodo test results. The cheapest of these is the 50/1.4 which, at f8, has a flat response of 70% out to 18mm from the centre. Some lenses such as the 135/2 fall only to 60% at the very corners.

Considering that the extreme corners are usually not significant in most compositions and especially not significant in shots where shallow DoF is sought, and considering the SQF criteria that Bill refers to, where good performance at 40 lp/mm would be the main requirement for sharp looking prints of 22"x33", I see no reason why good current lenses will not be up to the job.

However, because of QC lens variation, I would prefer that all lenses have a thorough MTF test, carried out either by the manufacturer or an approved agent, before being sold to the public. Such lenses should ship with detailed MTF results at a number of frequencies and should be graded and priced accordingly.

The astute buyer would then be in a position to select the grade of lens that best suits his/her requirements.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 01, 2007, 12:08:24 pm
Quote
It's not that simple. In your example you mention 40 lp/mm as a reference point, but other frequencies also come into play.

Bill,
Not as a reference point; merely as an example. The example can apply to as many frequency points you think necessary.

Quote
By the process of Reductio ad absurdum , with your line of reasoning, one could say that lens choice is not important for digital cameras since even a cheap lens will out resolve the camera. However, a good lens will often give better contrast at the important lower frequencies.

Not at all. That's not what I'm saying. Good lenses will always impart a certain quality to the image whatever the pixel density. I'm saying that a high pixel density sensor with even a modestly cheap lens, will deliver a sharper image than a low density sensor, at least in the central area of the image circle.

A high pixel density sensor will also deliver a sharper image with a very expensive lens, but perhaps also an even sharper image than the cheap lens, a higher contrast image and a better looking image in ways that sometimes might be difficult to define.

Quote
MTF and SQF complement each other. Whenever you reduce a complicated measurement to one number, you are losing information in the process of simplifying things.

I have no intention of simplifying things till they become meaningless. Let it all hang out, as long as it's relevant.

Quote
Bayer array sensors resolve about 70-80 percent of the Nyquist limit at low contrast, as can be seen by looking at the resolution tests on DPReview.

I thought the dpreview resolution tests were of high contrast line charts??
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 01, 2007, 08:01:05 pm
Quote
I thought the dpreview resolution tests were of high contrast line charts??
[a href=\"index.php?act=findpost&pid=115188\"][{POST_SNAPBACK}][/a]

They are, but at 10% MTF they become low contrast.  

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 01, 2007, 09:34:52 pm
Quote
They are, but at 10% MTF they become low contrast.   

[a href=\"index.php?act=findpost&pid=115259\"][{POST_SNAPBACK}][/a]

Then we have to wonder how lens resolution of 60 lp/mm and (say) 50% MTF becomes 60 lp/mm at a mere 10% MTF on a 20D which is effectively a cropped 21mp FF sensor. Is this due principally to lack of pixel density or too much read noise?
Title: f-stop limits for full sensor resolution
Post by: BJL on May 01, 2007, 11:07:28 pm
Quote
BJL,
The problem here is that there's no information on the performance of Canon lenses at 60 & 70 lp/mm...

However, there are a few lenses ranging from 50mm to 400mm that can resolve 40 lp/mm at 70% almost to the corners, according to Photodo test results. The cheapest of these is the 50/1.4 ...

... considering the SQF criteria that Bill refers to, where good performance at 40 lp/mm would be the main requirement for sharp looking prints of 22"x33"
[a href=\"index.php?act=findpost&pid=115187\"][{POST_SNAPBACK}][/a]
On the first point I agree that I am speculating by extrapolation from inconclusive data.

On the second, there are probably some lenses sharp enough to make good use of 30MP+, but I suspect too few for all but a rather special class of photography. Another guess of mine is that if the best standard zooms cannot make use of extra resolution, the market for it probably gets very small. But maybe still enough for one very high and 30MP model atop already very good 16-20MP high end models.

But what if you are right about 40 lp/mm being enough for sharp 22"x33" prints? Such resolution is probably comfortably provided already by the 16.5MP 1DsMkII sensor, so why push on to 5.5 micron pixel spacing, probably good for about 60lp/mm?

Canon talked a while ago about plans for some new higher resolution lenses (the first of which is the 16-35/2.8 II, it seems) for those who want high quality 13x19 prints.  What lp/mm and pixel density do you think is needed for that goal, more modest than the 22"x33" you mention?
Title: f-stop limits for full sensor resolution
Post by: Ray on May 01, 2007, 11:51:31 pm
Quote
On the second, there are probably some lenses sharp enough to make good use of 30MP+, but I suspect too few for all but a rather special class of photography. Another guess of mine is that if the best standard zooms cannot make use of extra resolution, the market for it probably gets very small. But maybe still enough for one very high and 30MP model atop already very good 16-20MP high end models.

I'm an optimist here. We know that QC variation in most (if not all) models of lenses is a serious problem. When people complain about their lenses not being good enough, I suspect that more often than not, it's simply due to the luck of the draw. They've got a dud, or at best a lens that falls into the lower 50% of the manufacturer's acceptable quality range. There's always a range for acceptable quality, set by the manufacturer.

My proposal is, let the consumer set the preferred range. Provide them with the information and let's see if the customer will pay a 50% premium for a lens that has sterling performance right to the edges, or a 20% premium for a lens of the same model that has moderately good performance to the edges.

There have been some disappointing reports of the new Canon 16-35. Who knows if such reports are due to QC variation. The consumer is being treated as a sucker. We're in the age of measurement. Even my 22 year old pair of loudspeakers (Celestion SL600) came with individual frequency response charts for each speaker.

What's the problem here? Beats me.

Quote
But what if you are right about 40 lp/mm being enough for sharp 22"x33" prints? Such resolution is probably comfortably provided already by the 16.5MP 1DsMkII sensor, so why push on to 5.5 micron pixel spacing, probably good for about 60lp/mm?

Micro detail. SQF is really a blunt instrument. Contrast at 40 lp/mm might well have an important effect on the perception of over-all sharpness in a 22x33" print, but those small leaves, high up in the tree need at least 70 lp/mm.
Title: f-stop limits for full sensor resolution
Post by: Claude Jodoin on May 06, 2007, 01:05:17 pm
Quote
These results more or less confirm the general consensus that f11 is the limit for stopping down with cropped 35mm format and and f16 the limit for full frame 35mm. Standard 50mm lenses are usually very sharp at f5.6 - f8. With the average zoom lens, the differences between f5.6 and f11 would be less.

Perhaps more relevant than using Imatest is to do real world comparisons with specific lenses. I've done extensive 'real world' comparisons of my 5D with 24-105 zoom and there's no significant resolution difference between f8 and f16, at the plane of focus. I haven't however done similar comparisons using my sharpest lens, the Canon 50/1.4.
[a href=\"index.php?act=findpost&pid=97247\"][{POST_SNAPBACK}][/a]

I was one of the "magnificent 7" photographers, brought to Foveon by the late, great, Dean Collins. He created the mentor program for the development of the original Prism Camera.

I had 6 different cameras from 1999-2001  and I can say, without any doubt, from a techno-geek perspective that the f/5.6-f/11 statement is true. The Foveon had a 6 micron well site pitch and optimum lens performance, verified by me with their chief engineer and optical enginneer, with all kinds of test targets was in that range.

The only time I was able to get any sort of moire pattern out of the Foveon (No moire was their claim to fame vs. the original Phase One Lightphase of the day) was at f/6.3. This turned out to be the peak MTF for the glass, allowing a sympathetic frequency of the clothing pattern against the well site pitch of the 3 chips micro-aligned onto a dichroic prism. Otherwise, the inability of the lens to reach 83 LPMM (required to resolve 6 microns) at other apertures made the lens act (effectively) as a low pass filter. Color aliasing was non existent since the 3 chips had full color sampling at every pixel, a Carver Mead philosophy which persists, even today in the lates Foveon X3 chips.

Claude Jodoin
Tech. Editor
Rangefinder/After Capture Mag.
Title: f-stop limits for full sensor resolution
Post by: Claude Jodoin on May 06, 2007, 01:09:07 pm
I forgot to mention that the camera had no moving parts. Only the aperture and focus motor in the lens moved. It could synch. to ANY studio flash at up to 1/8000 of a second, since it had an all-electronic shutter.

Planned obsolescence via electro-mechanical shutter wear persists today.

Claude Jodoin
Tech. Editor
Rangefinder/After Capture Mag.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 06, 2007, 09:57:51 pm
Quote
I had 6 different cameras from 1999-2001  and I can say, without any doubt, from a techno-geek perspective that the f/5.6-f/11 statement is true. The Foveon had a 6 micron well site pitch and optimum lens performance, verified by me with their chief engineer and optical enginneer, with all kinds of test targets was in that range.
[a href=\"index.php?act=findpost&pid=115963\"][{POST_SNAPBACK}][/a]

Claude,
I imagine that to equal the resolution of a Foveon type sensor with 6 micron spacing, a Bayer type sensor would need spacing even smaller than 5 microns. Right?

Perhaps the only way forward to maximise the full potential of existing 35mm lenses is the development of a full frame Foveon sensor with 6 micron pixel pitch, eventually progressing to 5 microns as better lenses are developed.  
Title: f-stop limits for full sensor resolution
Post by: BJL on May 07, 2007, 02:37:51 pm
Quote
Claude,
I imagine that to equal the resolution of a Foveon type sensor with 6 micron spacing, a Bayer type sensor would need spacing even smaller than 5 microns. Right?
[a href=\"index.php?act=findpost&pid=116054\"][{POST_SNAPBACK}][/a]
Right I think. According to both Foveon engineers and other sources including resolution tests I have seen, a Bayer CFA sensor needs about twice the pixel count of an "X3" type sensor to get equal resolution, or a factor about 1.4 in pixel spacing. So about 4 microns for Bayer CFA to match 6 micron X3.
Quote
Perhaps the only way forward to maximise the full potential of existing 35mm lenses is the development of a full frame Foveon sensor with 6 micron pixel pitch ...
[a href=\"index.php?act=findpost&pid=116054\"][{POST_SNAPBACK}][/a]
There is no fundamental problem making Bayer CFA photosites far smaller than the 4 microns needed to match 6 micron X3 photosites for resolution, or even the 3.5 micron ones needed to match 5 micron X3 photosites. So to claim an advantage for X3, you must assume that the X3 photosites would be superior in some way to Bayer CFA photosites of about half the area. Superior for example in sensitivity, noise levels at equal ISO, dynamic range and such.

Though that might seem intuitive, there is no evidence so far that it is true: the high ISO performance of Foveon's X3 sensors so far seems not to match that of good Bayer CFA photosites of half or less the area.

Perhaps this is because, at least with Foveon's version of X3 detection, far from all of the light of each of the three color ranges is detected by the appropriate layer of the three layer sensor. This is probably because Foveon's approach depends on the differential absorption in silicon of different wavelengths of light, which is very far from the ideal of, say, the top layer absorbing and detecting all blue light while all green and red light passes through, and so on.

Other X3 technologies are apparently being investigated by companies like Fujifilm and Canon, but it seems far too early to predict whether they will ever in fact perform better than the currently dominant Bayer CFA approach.
Title: f-stop limits for full sensor resolution
Post by: BJL on May 07, 2007, 02:54:19 pm
Quote
I'm an optimist here. ...
There have been some disappointing reports of the new Canon 16-35. ...
What's the problem here? Beats me.
[a href=\"index.php?act=findpost&pid=115282\"][{POST_SNAPBACK}][/a]
There might be an important point in there. Even with film, there was clearly room for improving overall image quality by improving on the quality of most or all Canon zoom lenses, so presumably Canon already had good reason to work on improving lens resolution. Thus it is unclear that the change to electronic sensors will lead Canon to achieve substantially more than it had already achieved as far as lens resolution, with zooms at least. Perhaps some primes are good enough that little would have been gained with film SLR's by improving them, but there is more room for improvement with current or future sensors of high enough resolution.

And as the 16-35 was the most often mentioned resolution limit of Canon's 35MM FF DSLR system, the most glaring weakness is any aspiration to take over the higher resolution sector previously dominated by medium format and as the new 16-35 is clearly intended as an important part of Canon's stated program of improving resolution limitations in its high end lens system, the "limited success" of that new lens as far as increased sharpness is an indication that 35mm format might already be close to the practical resolution limits of zoom lenses at least.

But I suppose that Canon can compete fairly well against medium format mostly with expensive prime lenses instead. So where are the new, sharper Canon primes, wide angles in particular?
Title: f-stop limits for full sensor resolution
Post by: BJL on May 07, 2007, 03:03:36 pm
Quote
I forgot to mention that the camera had no moving parts... It could synch. to ANY studio flash at up to 1/8000 of a second, since it had an all-electronic shutter.
[a href=\"index.php?act=findpost&pid=115965\"][{POST_SNAPBACK}][/a]
So no moving mirror: was this camera was a "rangefinder", with no TTL VF, or did it use a fixed beam splitter (pelicle mirror or such), or a live video viewfinder?

What was its maximum shutter speed and frame rate?

I am waiting for electronic shuttering to arrive in DSLR's, especially for high frame rate action, and for eliminating mirror and shutter vibration when used in conjunction with Live View (as now being offered in DSLR's by Olympus/Panasonic/Leica and Canon, and in a more limited form by Fujifilm.)
Title: f-stop limits for full sensor resolution
Post by: Ray on May 07, 2007, 05:35:47 pm
Quote
Though that might seem intuitive, there is no evidence so far that it is true: the high ISO performance of Foveon's X3 sensors so far seems not to match that of good Bayer CFA photosites of half or less the area.

Perhaps this is because, at least with Foveon's version of X3 detection, far from all of the light of each of the three color ranges is detected by the appropriate layer of the three layer sensor. This is probably because Foveon's approach depends on the differential absorption in silicon of different wavelengths of light, which is very far from the ideal of, say, the top layer absorbing and detecting all blue light while all green and red light passes through, and so on.

[a href=\"index.php?act=findpost&pid=116194\"][{POST_SNAPBACK}][/a]

This is my understanding also. However, we should not forget that the Bayer CFA deliberately blocks out a lot of light. Each of those red, green and blue filters is supposed to block out a large portion of the other 2 colors, is it not?

What proportion of light is blocked, do you know? Perhaps not as much as 2/3rds but maybe as much as a half. That doesn't sound particularly efficient to me.
Title: f-stop limits for full sensor resolution
Post by: BJL on May 08, 2007, 12:21:52 pm
Quote
... the Bayer CFA deliberately blocks out a lot of light. Each of those red, green and blue filters is supposed to block out a large portion of the other 2 colors, is it not?

What proportion of light is blocked, do you know? Perhaps not as much as 2/3rds but maybe as much as a half. That doesn't sound particularly efficient to me.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=116238\")
Indeed, that is the source of the hope that an X3 type sensor (one measuring three "colors" at each location) can have higher quantum efficiency and thus a sensitivity advantage at a given resolution level. The question is whether they can in fact achieve significantly higher QE than good CFA designs, given that there is some additional light loss due to the multiple absorption layers needed with X3. The inefficiencies of multiple layer sensors must be weighed against those of CFA's, and I trust experiment over the hand-waving theory of some X3 advocates to make such comparisons.

So here are some QE numbers, for Kodak sensors simply because Kodak lets it all hang out when it comes to sensor spec's, at [a href=\"http://www.kodak.com/US/en/dpq/site/SENSORS/name/ISSProductFamiliesRoot_product]http://www.kodak.com/US/en/dpq/site/SENSOR...iesRoot_product[/url]

The best I have seen so far is Kodak's new MF sensor with micro-lenses and 6.8 micron pixel pitch (most MF sensors lack micro-lenses, which about halves their sensitivity.) For overall luminosity sensitivity, the mid-spectrum green figure is probably the most indicative number.

KAF-31600, as in the H3-31 and the long expected Pentax DMF
QE 43% green, 37% red, 36% blue.

Another slightly older sensor with the same 6.8 micron pixel pitch:

KAF-10500, as in the Leica M8:
QE 40% green, 17% red, 32% blue.
That 17% red is anomolously low, and may be a typo.

An older sensor with smaller 5.4 micron pixel spacing:
KAF-8300, as in the (discontinued) Olympus E-300 and E-500:
Color version QE 40% green, 33% red, 33% blue.
Some new monochrome versions, to show how much light is lost to CFA's:
Monochrome, microlenses, no glass: 60%
Monochrome, microlenses, MAR glass (IR blocking?): 54%
Monochrome, microlenses, clear glass (IR blocking?): 52%
Monochrome, no microlenses, clear glass: 37%
Source: http://www.kodak.com/ezpres/business/ccd/g...300LongSpec.pdf (http://www.kodak.com/ezpres/business/ccd/global/plugins/acrobat/en/datasheet/fullframe/KAF-8300LongSpec.pdf)

A new interline CCD with "tiny" 4.75 micron pixel pitch:
KAI-10100:
QE 42% green, 32% red, 40% blue. (Aside: the highest blue QE I have ever seen.)


Monochrome sensors have higher QE, so the QE figures here are as a percentage of all light of all colors, suggesting that they are close to optimal. Being over 1/3 already sounds strange, but there are overlaps in the sensitivity curves, and perhaps "green" pixels in particular are made sensitive to more than one third of visible spectrum to make them more useful as the primary luminance measure, while R and B pixels are mostly used for "chroma" information.


The fact that monochrome sensors only have about 1.5x the sensitivity of CFA ones (60% vs 40% for the KAF-8300) suggest that the X3 advantage might not be so great after all. One possibility I see is that "green" pixels could effectively be "white pixels" giving optimum QE for luminance measurement, with R and B pixels solely for color information. Given the eye's greater resolution of luminance than color (rod vs cone density and cones being single color sensitive) this might be a smarter allocation of resources than X3's "misguided egalitarianism" in treating R, G and B as equally important.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 09, 2007, 09:26:42 am
Quote
The fact that monochrome sensors only have about 1.5x the sensitivity of CFA ones (60% vs 40% for the KAF-8300) suggest that the X3 advantage might not be so great after all. One possibility I see is that "green" pixels could effectively be "white pixels" giving optimum QE for luminance measurement, with R and B pixels solely for color information. Given the eye's greater resolution of luminance than color (rod vs cone density and cones being single color sensitive) this might be a smarter allocation of resources than X3's "misguided egalitarianism" in treating R, G and B as equally important.
[a href=\"index.php?act=findpost&pid=116392\"][{POST_SNAPBACK}][/a]

Interesting! But perhaps these differences are due to another issue apart from QE, and that's read noise which also affects sensitivity. Do we know that the worse noise performance, at high ISOs, of the Foveon sensor is due entirely to lower quantum efficiency?

For all I know, that higher noise could be entirely due to a lack of whatever technology Canon uses to reduce read noise. Perhaps the implementation of analog pre-amplification is either more difficult with a Foveon design, or would involve the use of patented processes held by Canon; or perhaps the group developing the Foveon sensors simply haven't cracked the problem of high read noise yet.

Nevertheless, revisiting dpreiew test results for the SD9 and SD10, I notice that the SD9 actually produces less noise than the Canon D60 and the SD10 less noise than the Canon 10D, although both Sigma cameras do not boast that highest ISO setting of the Canon cameras, ie. ISO 800 of the D60 and ISO 3200 of the 10D.

I suppose one could also argue that dpreview noise comparisons are on a pixel for pixel basis and that after taking into consideration the greater number of interpolated pixels of the D60 and 10D, the noise levels of the SD9 & SD10 are no better and perhaps slightly worse at the higher ISO setting.

The most obvious feature of the Foveon based cameras is that they're a bit behind in the pixel count race. The 4.6mp SD14 would be equivalent to about 8 or 9 Bayer type megapixels, wouldn't it?
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on May 09, 2007, 11:46:53 am
No, more like 6. Foveon pixels are better than interpolated Bayer pixels, but not 2x better. Maybe 1.5x...
Title: f-stop limits for full sensor resolution
Post by: Ray on May 09, 2007, 08:09:33 pm
Quote
No, more like 6. Foveon pixels are better than interpolated Bayer pixels, but not 2x better. Maybe 1.5x...
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=116579\")

That's not the impression I'm getting, Jonathan. You might like to have a look at the results of some fairly thorough tests of the SD14 by Mike Chaney, author of Qimage [a href=\"http://www.ddisoftware.com/sd14-5d/]here[/url]

The bottom line is, the Canon 5D has the resolution edge but the SD14 when compared with the 20D delivers more resolution.

In terms of pure resolution, Mike Chaney places the SD14 equivalent to a 10mp Bayer type sensor and in terms of 'less easily defined' image quality, closer to 12mp.
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on May 10, 2007, 12:42:44 pm
I've done my own comparisons, and simply don't buy the Foveon hype. The Foveon has better color accuracy per pixel, certainly, since no interpolation is being done. But in terms of resolution, and anti-aliasing filter does not halve the effective pixel count of the sensor. With proper sharpening technique, you can get pretty close to single-pixel detail from a Bayer sensor. Optimal sharpening for Foveon and Bayer sensors is different, and any comparison between them that sharpens both files exactly the same is intrinsically flawed.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 10, 2007, 08:59:09 pm
Quote
Optimal sharpening for Foveon and Bayer sensors is different, and any comparison between them that sharpens both files exactly the same is intrinsically flawed.
[a href=\"index.php?act=findpost&pid=116799\"][{POST_SNAPBACK}][/a]

I'm not sure I quite follow what you did you here. Are you saying:

(1) You compared the Bayer and Foveon sensors using the one sharpening routine which was actually optimal for the Bayer type sensor and therefore produced the result that the Foveon sensor appeared to be the equivalent of only 1.5x the number of Bayer interpolated pixels, or

(2) you compared both sensors using the most appropriate sharpening routine for each system, and despite this careful attention to methodology, found the Foveon sensor was equivalent to only 1.5x the number of Bayer pixels for resolution purposes.

If you did either or both (1) and (2), then one wonders what the result would have been if you had applied a sharpening routine to both sensors which was optimal for the Foveon sensor.

Setting aside such differences in sharpening procedures which will surely cloud the issue because there's probably no agreement on the best sharpening routine for either Bayer type or Foveon type images, it appears to be the case, from general resolution tests at dpreview, that it takes close to 3 Bayer pixels to record the width of one line pair.

If one of two sensors of the same dimensions has double the number of pixels of the other, then it has 1.4x the resolution and 1.4x the number of pixels along each dimension.

The Foveon sensor can record right up to the Nyquist limit (and beyond with a bit of aliasing apparently), ie. it takes just 2 Foveon pixels to record the width of one line pair. Therefore, logically, if it takes 1.4x2=2.8 Bayer pixels to record the same line, this would be approximately in accord with all the dpreview 'absolute' resolution tests for Bayer type systems that I've seen, and would tend to indicate that a Bayer type sensor needs double the pixel count to record the same detail as a Foveon sensor.

Your tests would tend to indicate it takes only 2.4 Bayer type pixels to record one line pair. You should publish the the results   .
Title: f-stop limits for full sensor resolution
Post by: Ray on May 10, 2007, 10:08:42 pm
Jonathan,
Just to to be sure I wasn't imagining these dpreview figures, I've just checked the lines-per-picture-height dpreview results for the 10mp Canon 400D.

We get a vertical resolution, across the 14.8mm height of the sensor, of 1800 lines.

(1) 1800/14.8/2= 60.8 lp/mm

(2) The number of pixels along this dimension is 2592.

(3) 2592/14.8/2=87.5 lp/mm

If the 400D sensor were a Foveon type, we could expect a resolution of 87.5 lp/mm which is actually 1.44x greater resolution than the 400D is actually delivering.

To put this another way, to increase resolution by 1.44x you need more than double the number of pixels.

The argument that these are lines right out to the edge of the sensor where resolution is not so good, might be relevant if one were counting vertical lines across the width of the 22.5mm sensor. This why I've looked at the horizontal lines in relation to the height of the sensor. Lenses like the Canon 50/1.4 have a remarkably flat MTF response, at 40 lp/mm and 70% contrast right out to 18mm from the centre, at their sharpest aperture of f8. On a sensor such as the 400D's, the middle of these horizontal lines would not extend beyond 7.4mm from the centre of the image circle.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 10, 2007, 10:36:56 pm
Quote
Setting aside such differences in sharpening procedures which will surely cloud the issue because there's probably no agreement on the best sharpening routine for either Bayer type or Foveon type images, it appears to be the case, from general resolution tests at dpreview, that it takes close to 3 Bayer pixels to record the width of one line pair.

If one of two sensors of the same dimensions has double the number of pixels of the other, then it has 1.4x the resolution and 1.4x the number of pixels along each dimension.

The Foveon sensor can record right up to the Nyquist limit (and beyond with a bit of aliasing apparently), ie. it takes just 2 Foveon pixels to record the width of one line pair. Therefore, logically, if it takes 1.4x2=2.8 Bayer pixels to record the same line, this would be approximately in accord with all the dpreview 'absolute' resolution tests for Bayer type systems that I've seen, and would tend to indicate that a Bayer type sensor needs double the pixel count to record the same detail as a Foveon sensor.

Your tests would tend to indicate it takes only 2.4 Bayer type pixels to record one line pair. You should publish the the results   .
[a href=\"index.php?act=findpost&pid=116882\"][{POST_SNAPBACK}][/a]

I performed Imatest analysis for the Sigma SD1 and Canon EOS 400 D in order to compare the Foveon 3.43 sensor with the Canon Bayer array sensor with up to date processing. Both images were downloaded from DPReview and primes at optimum apertures were used, limiting the lens part of the system.

[attachment=2470:attachment]

[attachment=2471:attachment]

The important criterion of perceived sharpness is line pairs/ picture height at 50% contrast (MTF 50) with correction for sharpening (shown in red). The 10MP Canon edges out the Foveon sensor, but not by a large margin (It handily beats the 6MP Nikon D100--not shown). At MTF 50, the Foveon resolves 0.435 cycles/pixel so 2.30 pixels are needed to resolve a line pair. This is outstanding performance, but not enough to make up for its lack of pixels.

The at MTF 50 the Canon resolves 0.368 cycles/pixel so 2.72 pixels are needed to resolve a line pair. It is resolving at about 75% of the Nyquist limit, which is typical good performance in the DPReviews.

As Ray mentioned the Foveon resolves well right up to Nyquist, but it has considerable aliasing since it lacks a blur filter, as shown by response beyond Nyquist, which is mainly due to aliasing but can also seen with excessive sharpening. The Foveon is subject to aliasing just as a Bayer sensor is, but with the Foveon the aliasing is monochrome and less noticeable.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 11, 2007, 12:07:10 am
Quote
The 10MP Canon edges out the Foveon sensor, but not by a large margin (It handily beats the 6MP Nikon D100--not shown).

Wait a minute, Bill. You are comparing a Foveon 3.43mp sensor with a Bayer 10mp sensor, right? The 10mp Bayer type edges out the 3.43mp Foveon. Quite understandable!

Extrapolating from the 3.43mp of the SD10 to the 4.6mp of the Sigma SD14, I think we could probably deduce that the SD14 is at least the equal of the 400D, resolution-wise. Right?

Noise is another issue and I'm disappointed there are no thorough comparisons yet(that I can find) of the 400D and SD14.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 11, 2007, 01:57:10 am
Quote
Just to to be sure I wasn't imagining these dpreview figures, I've just checked the lines-per-picture-height dpreview results for the 10mp Canon 400D.
[a href=\"index.php?act=findpost&pid=116893\"][{POST_SNAPBACK}][/a]

I'm afraid I have to respond to my own statements. When I'm wearing my philosophers's hat, I often realise that what I said in common language is quite absurd. How can I say, "to be sure I wasn't imagining these dpreview figures"?

Everything we know of and experience in any way, however subtle and however painful, or however wonderful, is imagined. If it's not imagined, it doesn't exist.

Those who claim there's a material reality that exists without our imagining it, are in deep trouble. The very claim itself is an act of imagination.

In other words, "I imagine there's a material world out there that is independent of my imagination", is an imaginative response which proves the statement wrong.  

(Just trying to raise the level of intellectual discussion here   ).
Title: f-stop limits for full sensor resolution
Post by: Jonathan Wienke on May 11, 2007, 08:53:35 am
Quote
I performed Imatest analysis for the Sigma SD1 and Canon EOS 400 D in order to compare the Foveon 3.43 sensor with the Canon Bayer array sensor with up to date processing. Both images were downloaded from DPReview and primes at optimum apertures were used, limiting the lens part of the system.

...

At MTF 50, the Foveon resolves 0.435 cycles/pixel so 2.30 pixels are needed to resolve a line pair. This is outstanding performance, but not enough to make up for its lack of pixels.

The at MTF 50 the Canon resolves 0.368 cycles/pixel so 2.72 pixels are needed to resolve a line pair. It is resolving at about 75% of the Nyquist limit, which is typical good performance in the DPReviews.

As Ray mentioned the Foveon resolves well right up to Nyquist, but it has considerable aliasing since it lacks a blur filter, as shown by response beyond Nyquist, which is mainly due to aliasing but can also seen with excessive sharpening. The Foveon is subject to aliasing just as a Bayer sensor is, but with the Foveon the aliasing is monochrome and less noticeable.

IMO this is much of the basis for the Foveon hype; people mistaking aliasing artifacts for true image detail. Since Foveon aliasing artifacts aren't neon colors, they aren't brightly-colored-obvious the way Bayer artifacts are. The figures cited here track pretty well with my observations and comparisons; pixel for pixel, a Foveon sensor resolves 1.18x more linear detail than a Bayer sensor, making it equivalent to a Bayer sensor with approximately 1.4x the pixel count. Throw in a little extra for color accuracy due to no interpolation, and 1.5x is pretty reasonable. But 2x is going way too far.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 11, 2007, 09:54:49 am
Quote
IMO this is much of the basis for the Foveon hype; [a href=\"index.php?act=findpost&pid=116942\"][{POST_SNAPBACK}][/a]

Maybe we should notify Phil Askey of dpreview about this. We don't want his spreading misinformation, do we?    

His vertical LPH test for the 3.43mp SD10 is 1550 lines. For the 6mp Canon 10D it's 1450.

If we look at 'extinction' resolution, which would include a lot of aliasing artifacts, it's > 2000 for the SD10 and 1850 for the 10D.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 11, 2007, 01:22:46 pm
Quote
Wait a minute, Bill. You are comparing a Foveon 3.43mp sensor with a Bayer 10mp sensor, right? The 10mp Bayer type edges out the 3.43mp Foveon. Quite understandable!

Extrapolating from the 3.43mp of the SD10 to the 4.6mp of the Sigma SD14, I think we could probably deduce that the SD14 is at least the equal of the 400D, resolution-wise. Right?

Noise is another issue and I'm disappointed there are no thorough comparisons yet(that I can find) of the 400D and SD14.
[a href=\"index.php?act=findpost&pid=116906\"][{POST_SNAPBACK}][/a]

Ray,

Yes I am comparing the 3.43MP SD10, since I don't have any test images from the 4.6 MP sensor. As Michael pointed out in his essay on sensors, you really have to double the MP to see a significance difference in linear resolution. The SD10 has a picture height of 1512 pixels, whereas the newer model has a PH of 1760. If one assumes that the newer sensor resolves the same 0.436 cycles/pixel of the older model, then the MTF 50 of the two systems would be 658 lp/ph and 767 lp/ph (1316 and 1534 line widths/picture height). The 400D MTF 50 is 1829 lw/ph, so the Canon wins here on this measure.

Bill
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 11, 2007, 01:43:45 pm
Quote
IMO this is much of the basis for the Foveon hype; people mistaking aliasing artifacts for true image detail. Since Foveon aliasing artifacts aren't neon colors, they aren't brightly-colored-obvious the way Bayer artifacts are. The figures cited here track pretty well with my observations and comparisons; pixel for pixel, a Foveon sensor resolves 1.18x more linear detail than a Bayer sensor, making it equivalent to a Bayer sensor with approximately 1.4x the pixel count. Throw in a little extra for color accuracy due to no interpolation, and 1.5x is pretty reasonable. But 2x is going way too far.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=116942\")

I agree entirely with Jonathan's analysis. The DPReview tests show the Foveon sensors resolving above Nyquist. In his report, Phil did mention that there was some discussion whether they were observing useful detail or aliasing garbage. He decided the former, but I would submit that the latter is more likely.

The Bayer sensors with a blur filter give little MTF at Nyquist as intended. As Norman Koren has pointed out, it does not really make sense to judge detail at the point it disappears. One should really look at the frequencies that most affect perceived sharpness at a given picture size, and these frequencies are lower than commonly thought as explained by [a href=\"http://bobatkins.com/photography/technical/mtf/mtf4.html]Bob Atkins[/url] in his article on subjective quality factor (SQF).

Here are SQF plots derived from Imatest for the Foveon sensor and the Canon EOS 400D:

[attachment=2475:attachment]

[attachment=2476:attachment]

The Canon gives considerably better SQF, but is more dependent on proper sharpening.

Bill
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 11, 2007, 01:59:02 pm
Quote
Maybe we should notify Phil Askey of dpreview about this. We don't want his spreading misinformation, do we?   

His vertical LPH test for the 3.43mp SD10 is 1550 lines. For the 6mp Canon 10D it's 1450.

If we look at 'extinction' resolution, which would include a lot of aliasing artifacts, it's > 2000 for the SD10 and 1850 for the 10D.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=116951\")

The Nyquist limit for the SD 10 is 1512 line widths/picture height and Phil's results are not credible. He is observing aliasing artifacts, which are quite prominent near Nyquist with the Foveon sensor, which lacks an anti-aliasing filter.

He should really upgrade his test methods to the 21st century. He is still using 1950's concepts. His noise tests are similarly non informative. Often he uses JPEGs that have noise reduction turned on and obscure not only the noise but also image detail. His standard deviations from raw captures converted with ACR are more informative, but pixel standard deviations do not take into account the frequency of the noise distribution resulting from differences of pixel density. For a given picture size, as the pixel count increases the noise becomes finer grained.

The test reports at [a href=\"http://www.imaging-resource.com/PRODS/EOS1DS2/1DS2IMATEST.HTM]Imaging Resource[/url] are more up to date.
Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 12, 2007, 02:45:32 am
Quote
The Nyquist limit for the SD 10 is 1512 line widths/picture height and Phil's results are not credible. He is observing aliasing artifacts, which are quite prominent near Nyquist with the Foveon sensor, which lacks an anti-aliasing filter.

Fair point, Bill. But Phil Askey has taken a lot of images with the SD10 and Canon 10D with which he compares it. I'm sure he wouldn't have included aliasing artifacts in his figures unless he thought they contibuted something meaningful to the image.

In situations like this where there might be some doubt as to what constitutes real detail as  opposed to artifacts, the printed page is an ideal test target. When the distances and print size are matched so that the print is barely legible, the camera that produces the most legible print is the one with the higher resolving power (assuming equal FoV and lens quality of course).

It matters little if any increased legibility is due to aliasing or includes alisaing or is present despite aliasing. If the print is more legible, or equally legible (whatever the case), that's all that counts.

It's difficult to say whether the SD10 samples in Phil Askey's review indicate that print legibility is the equal of the 10D. There's some indication that the 10D provides more detail than an interpolated SD10 image which has not been sharpened. I'm not sure how significant this is. I know you have mentioned that it's common prcatice to provide a bit of sharpening when interpolating images, but my own experience tends to indicate that sharpening cannot make invisible detail appear. Examing images at 100% to 400% on screen without sharpening is sufficient (for me) to determine the level of detail.

Without descending into nitpicking and extreme pixel peeping, I would be prepared to accept that the 3.43mp SD10 is the equal of the 6mp 10D, which makes the Foveon sensor equal to a Bayer type with 1.75x the number of (interpolated) pixels.

Jonathan reckons 1.5x, ie. a 5mp Bayer sensor. 5 or 6, neither here nor there.

Perhaps we could persuade BJL to hire an SD10 and compare with his 5mp E1   .
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 12, 2007, 10:23:00 am
Quote
Fair point, Bill. But Phil Askey has taken a lot of images with the SD10 and Canon 10D with which he compares it. I'm sure he wouldn't have included aliasing artifacts in his figures unless he thought they contibuted something meaningful to the image.
[a href=\"index.php?act=findpost&pid=117069\"][{POST_SNAPBACK}][/a]

In all fairness to Phil, he did report that there was much internal debate as to the significance of the "extra detail" offered by the Foveon sensor.

Quote
In situations like this where there might be some doubt as to what constitutes real detail as  opposed to artifacts, the printed page is an ideal test target. When the distances and print size are matched so that the print is barely legible, the camera that produces the most legible print is the one with the higher resolving power (assuming equal FoV and lens quality of course).

It matters little if any increased legibility is due to aliasing or includes alisaing or is present despite aliasing. If the print is more legible, or equally legible (whatever the case), that's all that counts.

[a href=\"index.php?act=findpost&pid=117069\"][{POST_SNAPBACK}][/a]

If you are photographing high contrast subjects with regularly repeating lines (such as the ISO test chart), then the aliasing of the Foveon sensor does seem to bring out detail, but so does judicious sharpening. In looking at the resolution target, I was struck by how cleanly defined the lines appeared near Nyquist with the Foveon sensor.

However, in our everyday photography we take pictures of naturally occurring scenes, not high contrast test charts with regularly repeating lines. Furthermore, MTF near the Nyquist frequency is of lesser importance to the appearance of the picture than the MTF at lower frequencies, at least for normally sized prints, as shown by the SQF graphs I provided. What did you think of them?

In naturally occurring subjects, aliasing may not be evident, since there are no uniformly repeating features. Indeed, the Leica M8 and medium format sensors often obtain excellent results with aliasing present but not readily seen except when the subject contains regularly repeating features such as seen in the fabric of clothing. However, the aliasing may have degraded the image or it may have actually enhanced the image.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 12, 2007, 08:12:08 pm
Quote
However, in our everyday photography we take pictures of naturally occurring scenes, not high contrast test charts with regularly repeating lines. Furthermore, MTF near the Nyquist frequency is of lesser importance to the appearance of the picture than the MTF at lower frequencies, at least for normally sized prints, as shown by the SQF graphs I provided. What did you think of them?
[a href=\"index.php?act=findpost&pid=117105\"][{POST_SNAPBACK}][/a]

Well, first, I'm not persuaded by the practical benefits of matching camera performance with 'normal' print size. I want the maximum quality I can get on the basis that too much quality for a small print is really no problem but too little quality for a large print definitely is a problem.

I adopt the same approach when scanning film. I'm not interested in scanning a slide at a low resolution suitable say for a postcard size print, in order to save time and storage, whilst putting myself in the position of having to rescan the slide if I or someone else wants a larger print at a later date.

Your SQF graphs seem a bit puzzling. If I'm reading them correctly, they seem to be saying:

(1) The unprocessed 400D image has a higher SQF at all print sizes than the same image sharpened.

(2) The 400D image, whether sharpened or not, has a higher SQF than the sharpened Sigma 10D image, at all print sizes, but the Sigma sharpened image has a higher SQF than its unsharpened version, which is what one would expect.

Generally though, one would not expect the 3.43mp SD10 to compete with the 10mp 400D. I don't think anyone is claiming this.

An often overlooked point about trying to be so precise about such matters, is lens variability. You can see the problem that Imaging Resource had when they tested the SD10. It appeared to be softer than the SD9. They had to repeat the test with another lens.

Strictly speaking, unless you are using the same lens when comparing sensor performance, the results can be no more than a rough guide.

Until recently, my sharpest lens was the el cheapo Canon 50/1.8 II. The question has been raised more than once on this forum whether the more expensive 50/1.4 is noticeably better. A number of contributors posted images comparing the two lenses. The 50/1.4 appeared to be noticeably sharper, even at f8.

On my last trip overseas I was able to pick up a 50/1.4 at a good price. I recently compared both lenses with test charts, textures and printed text at all apertures. My copy of the 50/1.4 is no better than the 50/1.8 II.

Whilst it's true the 50/1.4 gives me a half stop wider aperture, the resolution at f1.4 is truly awful, as is the resolution of the 50/1.8 II at f1.8. I'm just not into the taking of fuzzy images. I wouldn't bother to use either of these lenses at full aperture.
Title: f-stop limits for full sensor resolution
Post by: JeffKohn on May 13, 2007, 12:55:03 am
Quote
Generally though, one would not expect the 3.43mp SD10 to compete with the 10mp 400D. I don't think anyone is claiming this.
Sigma claims this.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 13, 2007, 04:25:46 am
Quote
Sigma claims this.
[a href=\"index.php?act=findpost&pid=117216\"][{POST_SNAPBACK}][/a]

I think you are confusing the discontinued 3.43mp SD10 with Sigma's latest model the 4.6mp SD14.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 13, 2007, 09:38:46 am
Quote
Well, first, I'm not persuaded by the practical benefits of matching camera performance with 'normal' print size. I want the maximum quality I can get on the basis that too much quality for a small print is really no problem but too little quality for a large print definitely is a problem.

I adopt the same approach when scanning film. I'm not interested in scanning a slide at a low resolution suitable say for a postcard size print, in order to save time and storage, whilst putting myself in the position of having to rescan the slide if I or someone else wants a larger print at a later date.

[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=117191\")

Ray,

Now you are talking like Nathan Myhrvold: he wants the best quality his camera is capable of achieving, not merely that need for good results at a given image size. In general, I think that is a good approach. Generally, a capture that produces an excellent large print will also do well at smaller print sizes, but as the SQF discussion by [a href=\"http://bobatkins.com/photography/technical/mtf/mtf4.html]Bob Atkins[/url] shows, this is not always the case.

Bill
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 14, 2007, 09:03:29 am
Quote
Your SQF graphs seem a bit puzzling. If I'm reading them correctly, they seem to be saying:

(1) The unprocessed 400D image has a higher SQF at all print sizes than the same image sharpened.

(2) The 400D image, whether sharpened or not, has a higher SQF than the sharpened Sigma 10D image, at all print sizes, but the Sigma sharpened image has a higher SQF than its unsharpened version, which is what one would expect.

[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=117191\")

Ray,

It is not unusual for the unprocessed image to have a higher SQF than the sharpened image because of the application of standardized sharpening. For over sharpened images, Imatest tries to undo the sharpening and then apply a normal amount of sharpening.

DPReivew and many other test sites use in camera JPEGs with camera defaults and the resulting images may be over sharpened. Over sharpening is more common with P&S and is less common with SLRs where sharpening may be left to the more sophisticated user.

For an example of an over sharpened image look at this figure in the [a href=\"http://www.imatest.com/docs/sqf.html]Imatest Documentation[/url]. Since the image was over sharpened, the SQF with standardized sharpening is lower than the image with standardized sharpening.

In the case of the Canon 400D, the image does not appear to be over sharpened since there is little overshoot in the edge plot and the image is reported as under sharpened by Imatest. When performing one's own tests, it is best to use raw apply no sharpening to the converted image.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 14, 2007, 09:47:35 am
Bill,
Perhaps these SQF procedures could be modified to produce an SDOFQF chart. (Subjective Depth of Field Quality Factor)   .

We could then refine standard DoF calculators by taking into consideration lens quality at the plane of focus as well as print size and viewing distance.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 14, 2007, 12:35:47 pm
Quote
Bill,
Perhaps these SQF procedures could be modified to produce an SDOFQF chart. (Subjective Depth of Field Quality Factor)   .

We could then refine standard DoF calculators by taking into consideration lens quality at the plane of focus as well as print size and viewing distance.
[a href=\"index.php?act=findpost&pid=117449\"][{POST_SNAPBACK}][/a]

Ray,

I take it you are being facetious?  


All of this information could be a bit difficult to implement at the time one is taking the picture. You might not know in advance the print size that will be used and other factors when you are adjusting the f/stop in order to optimize depth of field without running into diffraction limits.

Bill
Title: f-stop limits for full sensor resolution
Post by: Ray on May 14, 2007, 09:39:39 pm
Quote
I take it you are being facetious?   
All of this information could be a bit difficult to implement at the time one is taking the picture. You might not know in advance the print size that will be used and other factors when you are adjusting the f/stop in order to optimize depth of field without running into diffraction limits.
[a href=\"index.php?act=findpost&pid=117480\"][{POST_SNAPBACK}][/a]

Not completely facetious. I think it's doable   . There's at least one contributor to this site who claims to take into consideration print size factors when calculating distances and appropriate f stop for a certain DoF effect.

However, for a completley accurate result one needs to be able to measure distances accurately.

However, having got this completely accurate result in precise accordance with one's intentions, one then has the problem of communicating this result to the viewer.

In order to do this, might I suggest appropriately positioned viewing platforms in galleries for each print. Something along the lines, "Those with 20/20 vision please stand here to fully appreciate the intent of the author."  

Of course, we now have the complexities of individual vision acuity. Perhaps the photographer did not have 20/20 vision, so we need an algorithm to translate the photographers visual acuity into various standards that might correspond with the various standards of vision of the public at large.

Perhaps on the viewing platform we need various recommended positions for people with different eyesight problems.

Of course, the individual doesn't carry around such precise information about the condition of his/her eyesight.

That's fundamentally my point, without sarcasm.
Title: f-stop limits for full sensor resolution
Post by: jani on May 15, 2007, 05:54:35 am
Quote
Perhaps on the viewing platform we need various recommended positions for people with different eyesight problems.

Of course, the individual doesn't carry around such precise information about the condition of his/her eyesight.

That's fundamentally my point, without sarcasm.
This is fundamentally a problem with the artist's vision.

I hereby submit my latest work, "2", which does not have these problems. There is no recommended viewing distance, just view it.

[attachment=2495:attachment]
Title: f-stop limits for full sensor resolution
Post by: Ray on May 15, 2007, 07:09:42 am
Quote
This is fundamentally a problem with the artist's vision.

I hereby submit my latest work, "2", which does not have these problems. There is no recommended viewing distance, just view it.

[attachment=2495:attachment]
[a href=\"index.php?act=findpost&pid=117646\"][{POST_SNAPBACK}][/a]

Not sure about that, Jani. If the viewer is too far away, that dot might not be visible at all.  
Title: f-stop limits for full sensor resolution
Post by: BJL on May 15, 2007, 11:59:27 am
Quote
I agree entirely with Jonathan's analysis. The DPReview tests show the Foveon sensors resolving above Nyquist. In his report, Phil did mention that there was some discussion whether they were observing useful detail or aliasing garbage. He decided the former, but I would submit that the latter is more likely.[a href=\"index.php?act=findpost&pid=117000\"][{POST_SNAPBACK}][/a]
I also agree, and some test patterns show very clear, classic signs of aliasing: as you move down the pattern of nine narrowing lines, other sensors just have them fade to gray, but with the SD10, the lines fade in and out, and towards the small end, the number of black lines appears to be different than the true value of nine: aliasing of a higher spatial frequency to a lower one.

One irony of this is that some years ago, a Foveon engineer said in an interview that Foveon X3 type sensors have resolution comparable to a Bayer CFA sensor of about twice the pixel count, corresponding to 1.4x more "line pairs per pixel" (which fits with tests  have seen and theory of luminance resolution dominated by green pixel data) and yet some X3 enthusiasts persist in claiming better resolution that that Foveon engineer.

(This is not the first time that enthusiasts claim more for a product that even the technical people at the company that makes it, leaving me skeptical: the often claimed imminence of Canon 35mmFF DSLR's at "mainstream enthusiast" prices is another of course.)
Title: f-stop limits for full sensor resolution
Post by: Eric Myrvaagnes on May 15, 2007, 02:18:26 pm
Quote
This is fundamentally a problem with the artist's vision.

I hereby submit my latest work, "2", which does not have these problems. There is no recommended viewing distance, just view it.

[attachment=2495:attachment]
[a href=\"index.php?act=findpost&pid=117646\"][{POST_SNAPBACK}][/a]
The highlights appear just a bit blown to me, but otherwise fine. I feel that "3" might be a better title, helping to bring out the metaphysical subtleties.  

Eric
Title: f-stop limits for full sensor resolution
Post by: JeffKohn on May 16, 2007, 12:01:32 pm
Quote
I think you are confusing the discontinued 3.43mp SD10 with Sigma's latest model the 4.6mp SD14.
[a href=\"index.php?act=findpost&pid=117232\"][{POST_SNAPBACK}][/a]
No, Sigma marketed the SD10 as a 10mp camera and they're marketing the SD14 as a 14mp camera. They can argue that technically that might be true in a way, but I still think it's highly misleading advertising.
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on May 16, 2007, 02:00:32 pm
Quote
So here are some QE numbers, for Kodak sensors simply because Kodak lets it all hang out when it comes to sensor spec's, at http://www.kodak.com/US/en/dpq/site/SENSOR...iesRoot_product (http://www.kodak.com/US/en/dpq/site/SENSORS/name/ISSProductFamiliesRoot_product)

KAF-10500, as in the Leica M8:
QE 40% green, 17% red, 32% blue.
That 17% red is anomolously low, and may be a typo.[a href=\"index.php?act=findpost&pid=116392\"][{POST_SNAPBACK}][/a]

If that's the sensitivity of the red channel for white (full visible spectrum) light, then it is not atypical at all.  Most CFA cameras are 40 to 60% as sensitive to red as to green, with blue usually in-between.

The only thing odd is that this camera has almost no IR filtration, which is usually the cause of low red sensitivity.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 16, 2007, 07:18:06 pm
Quote
No, Sigma marketed the SD10 as a 10mp camera and they're marketing the SD14 as a 14mp camera. They can argue that technically that might be true in a way, but I still think it's highly misleading advertising.
[a href=\"index.php?act=findpost&pid=117912\"][{POST_SNAPBACK}][/a]

Ah! Yes, I see what you mean, but that's probably not the fault of Sigma. The confusion already existed in the use of the term 'pixel' before the SD9 was brought to market.

Before I even thought of buying my first digital camera, I understood that a pixel was a picture element consisting of a red, green and blue component. I was very unimpressed with the specs of those early 2mp and 3mp digital cameras until it was explained to me that those were all monochrome pixels that would be interpolated to what I understood as proper pixels.

I was still unimpressed, but a little less so. I knew that interpolated information is not completely accurate information.

If you are marketing a product and there's already ambiguity in the use of a term, it would be foolish (from a purely marketing point of view) to use the lower and less impressive figure.
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on May 16, 2007, 09:55:20 pm
Quote
Before I even thought of buying my first digital camera, I understood that a pixel was a picture element consisting of a red, green and blue component.[a href=\"index.php?act=findpost&pid=118037\"][{POST_SNAPBACK}][/a][/

That understanding was based on a limited context of color monitor "pixels".

If you turn down the saturation on a monitor, you can use each of its color pixels as 3 B&W pixels.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 16, 2007, 10:45:52 pm
Quote
That understanding was based on a limited context of color monitor "pixels".
[a href=\"index.php?act=findpost&pid=118072\"][{POST_SNAPBACK}][/a]

That may be true, but computer monitors were popular long before digital cameras.

I'm not commenting on the rightness or wrongnes of these different definitions but merely that they exist and are a source of confusion that wasn't created by Sigma.

Using the definition of pixel that is used when describing the number of pixels of a Bayer type sensor, Sigma are justified in describing the pixel count of their Foveon sensors by the same definition, but they would be wrong to claim that 10m of their pixels have the same resolving power as 10m Bayer type pixels. And I don't think they are claiming this.
Title: f-stop limits for full sensor resolution
Post by: Ray on May 17, 2007, 12:05:58 am
Quote
If you turn down the saturation on a monitor, you can use each of its color pixels as 3 B&W pixels.
[a href=\"index.php?act=findpost&pid=118072\"][{POST_SNAPBACK}][/a]

let's be clear about this. Are you implying that my monitor, currently set at its maximum resolution of 1600x1280 pixels, can in fact be converted to a B&W monitor with a resolution of 4800x3840 pixels?

Oops! Good job I've got time to edit this so I can avoid making a complete pratt of myself. My monitor resolution is 1600x1200 of course and 3x the number of pixels equates to a resolution of around 2768x2076.

So, John, I desaturate my monitor and thereby increase its resolution to 2768x2076 for B&W images, or do I simply increase the bit depth???
Title: f-stop limits for full sensor resolution
Post by: jani on May 18, 2007, 11:20:35 am
Quote
let's be clear about this. Are you implying that my monitor, currently set at its maximum resolution of 1600x1280 pixels, can in fact be converted to a B&W monitor with a resolution of 4800x3840 pixels?

Oops! Good job I've got time to edit this so I can avoid making a complete pratt of myself. My monitor resolution is 1600x1200 of course and 3x the number of pixels equates to a resolution of around 2768x2076.

So, John, I desaturate my monitor and thereby increase its resolution to 2768x2076 for B&W images, or do I simply increase the bit depth???
No, John is mistaken.

In RGB CRT displays, desaturation does not magically change the red, green and blue dots of phosphor into monochrome dots of phosphor.

RGB CRTs emulate greytones, just as RGB LCDs do.
Title: f-stop limits for full sensor resolution
Post by: BJL on May 18, 2007, 03:26:32 pm
Quote
If that's the sensitivity of the red channel for white (full visible spectrum) light, then it is not atypical at all.  Most CFA cameras are 40 to 60% as sensitive to red as to green, with blue usually in-between.

The only thing odd is that this camera has almost no IR filtration, which is usually the cause of low red sensitivity.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=117944\")
Still, the red QE of this sensor is eggregiously low compared to other Kodak FF CCD's with micro-lenses:
KAF-10500: 17% R 40% G  32% B
One of similar vintage, same pixel size, also with off-set micro-lenses:
KAF-31600: 37% R 43% G  36% B
The slightly older one in the Leica R back, of same size and pixel pitch as the KAF-10500
KAF-10010: 34% R 40% G  36% B
The older one with smaller pixels in the Olympus E-300 and E-500:
KAF-8300:  33% R 40% G  33% B
The far older one in the Olympus E-1 withe the same pixel pitch as the KAF-10500:
KAF-5100:  31% R 34% G  31% B
Every other one has about twice the red QE as the 17% for the KAF-10500.

But indeed the graphs on page 12 of the spec. document [a href=\"http://www.kodak.com/ezpres/business/ccd/global/plugins/acrobat/en/datasheet/fullframe/KAF-10500LongSpec.pdf]http://www.kodak.com/ezpres/business/ccd/g...500LongSpec.pdf[/url] confirm it, and make me think that this was deliberate, a consequence of an attempt to avoid the need for an IR filter by effectively doing it in the CFA filters. I say that because the KAF-10500 sensitivity curves are extended far further into the IR than in the specs for other Kodak FF CCD sensors (to 1100nm instead of 700nm) and show a long flat tail of very low IR sensitivity, where with many other sensors, sensitivity picks up a bit at some point in the near IR, even in the B and G channels.

So maybe
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on May 19, 2007, 09:58:15 am
Quote
No, John is mistaken.

In RGB CRT displays, desaturation does not magically change the red, green and blue dots of phosphor into monochrome dots of phosphor.
[a href=\"index.php?act=findpost&pid=118386\"][{POST_SNAPBACK}][/a]


You're right, it doesn't.  I was thinking of something else from the past when I wrote that, where I simulated ~2560 shades of grey with the 3:6:1 R:G:B luminance ratios, and forgot that they did not increase virtual resolution.
My main point, however, is that a Pixel represents a 2D location and has nothing to do with color, whatsoever.  B&W displays have pixels, IR B&W cameras have pixels, etc.  "Pixels" have nothing generically to do with "RGB".
Title: f-stop limits for full sensor resolution
Post by: Ray on May 19, 2007, 08:45:40 pm
Quote
My main point, however, is that a Pixel represents a 2D location and has nothing to do with color, whatsoever.  B&W displays have pixels, IR B&W cameras have pixels, etc.  "Pixels" have nothing generically to do with "RGB".
[a href=\"index.php?act=findpost&pid=118525\"][{POST_SNAPBACK}][/a]

John,
Nevertheless, if I say my color monitor has a resolution of 1600x1200 pixels, I mean that it has 1600 groups of 3 phosphors in the horizontal direction and 1200 groups of 3 phosphors in the vertical direction. Each of the three phosphors of each group is designed to emit only one primary color when excited by the electron gun. They are color specific.

My knowledge of electronics is very limited, but I presume when I turn down the saturation of my monitor, I'm merely putting in place a set of instructions that causes all 3 phosphors in each group to be excited equally. It's the group of 3 that is called a pixel. Is this not the case?
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on May 19, 2007, 11:24:28 pm
Quote
John,
Nevertheless, if I say my color monitor has a resolution of 1600x1200 pixels, I mean that it has 1600 groups of 3 phosphors in the horizontal direction and 1200 groups of 3 phosphors in the vertical direction. Each of the three phosphors of each group is designed to emit only one primary color when excited by the electron gun. They are color specific.

My knowledge of electronics is very limited, but I presume when I turn down the saturation of my monitor, I'm merely putting in place a set of instructions that causes all 3 phosphors in each group to be excited equally. It's the group of 3 that is called a pixel. Is this not the case?
[a href=\"index.php?act=findpost&pid=118635\"][{POST_SNAPBACK}][/a]

It's the case, but only because it's an RGB monitor, intended to be used to simulate a single RGB pixel.  Technically, each pixel is actually in three discreet 2D locations.  If it were representing a CFA of the same geometrical shape (without demosaicing), it could conceivably be considered to have 3x as many pixels, and have more spatial resolution.  Demosaicing is probably unnecessary when resolution is above the range of individual pixel recognition.
Title: f-stop limits for full sensor resolution
Post by: bjanes on May 20, 2007, 08:02:49 am
Quote
Demosaicing is probably unnecessary when resolution is above the range of individual pixel recognition.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=118647\")

John,

That is an interesting speculation. If no demosaicing were done, wouldn't it be necessary to have equal numbers of RBG elements so that the colors would be balanced? [a href=\"http://en.wikipedia.org/wiki/Demosaicing]Wikipedia[/url] has an illustration of an undemosiaced Bayer image. Note the greenish cast in the white character shadow areas.

Bill
Title: f-stop limits for full sensor resolution
Post by: John Sheehy on May 20, 2007, 08:24:19 pm
Quote
John,

That is an interesting speculation. If no demosaicing were done, wouldn't it be necessary to have equal numbers of RBG elements so that the colors would be balanced? Wikipedia (http://en.wikipedia.org/wiki/Demosaicing) has an illustration of an undemosiaced Bayer image. Note the greenish cast in the white character shadow areas.
[a href=\"index.php?act=findpost&pid=118674\"][{POST_SNAPBACK}][/a]

One of my stipulations was that the CFA on the camera had to match the Color pattern of the monitor.  So, the camera in this hypothetical situation would have equal amounts each of red, green and blue pixels.  The color pixels would display the luminance as it was witnessed through a corresponding sensor pixel, altered, of course, for color balance (and local color profiling, if the spectral properties were not equal).  You could do color balancing, profiling, saturation, etc, without demosaicing.

I'm sure we'll never see something like this, though.