Luminous Landscape Forum
Equipment & Techniques => Digital Cameras & Shooting Techniques => Topic started by: dwdallam on July 08, 2008, 05:03:38 am
-
Interesting thread on MF forum. Here is an excerpt:
Thread Location:
http://luminous-landscape.com/forum/index....ic=26025&st=160 (http://luminous-landscape.com/forum/index.php?showtopic=26025&st=160)
Excerpt:
http://www.openphotographyforums.com/forum...p?t=6103&page=3 (http://www.openphotographyforums.com/forum...p?t=6103&page=3)
See reply 23
It's an extreme sample, but sort like stories have been appearing on several sites.
I tested a 1DsIII a while ago and also found that f11 was noticable softer that f8 and this was with a 70-200 f2.8 L IS.
*
Frank,
Makes sense to me. When I moved from the 8mp 20D to the 5D, I was surprised I could use F16 with minimal loss of resolution with the 5D. With the 20D I would hesitate to stop down beyond F8 because I knew there would be a trade-off. I'd be sacrificing sharpness for the benefit of additional DoF.
The 1Ds3 has similar pixel density to the 20D. Resolution (ie. lp/mm) does not care about sensor size, only pixel density or pixel pitch. The same principles that apply to the 20D will apply to the 1Ds3. Softening of the image will appear at the same F stops using the same lens.
-
Effect of diffraction in combo with pixel density?
Interesting thread on MF forum. Here is an excerpt:
Thread Location:
http://luminous-landscape.com/forum/index....ic=26025&st=160 (http://luminous-landscape.com/forum/index.php?showtopic=26025&st=160)
Excerpt:
http://www.openphotographyforums.com/forum...p?t=6103&page=3 (http://www.openphotographyforums.com/forum...p?t=6103&page=3)
See reply 23
It's an extreme sample, but sort like stories have been appearing on several sites.
I tested a 1DsIII a while ago and also found that f11 was noticable softer that f8 and this was with a 70-200 f2.8 L IS.
*
Frank,
Makes sense to me. When I moved from the 8mp 20D to the 5D, I was surprised I could use F16 with minimal loss of resolution with the 5D. With the 20D I would hesitate to stop down beyond F8 because I knew there would be a trade-off. I'd be sacrificing sharpness for the benefit of additional DoF.
The 1Ds3 has similar pixel density to the 20D. Resolution (ie. lp/mm) does not care about sensor size, only pixel density or pixel pitch. The same principles that apply to the 20D will apply to the 1Ds3. Softening of the image will appear at the same F stops using the same lens.
[a href=\"index.php?act=findpost&pid=206365\"][{POST_SNAPBACK}][/a]
-
Effect of diffraction in combo with pixel density?
[a href=\"index.php?act=findpost&pid=206381\"][{POST_SNAPBACK}][/a]
It's just the matter of diffraction.
In perfect, diffraction-limited lens, using ideal values for the Rayleigh limit, and applying the contrast theory of Kühler, resolution for the average wavelength of light (0,555 micron) is:
Aperture resolution
1,4 550 lp/mm
2,0 385 lp/mm
2,8 263 lp/mm
4,0 185 lp/mm
5,6 135 lp/mm
8,0 94 lp/mm
11 69 lp/mm
16 48 lp/mm
22 30 lp/mm
32 21 lp/mm
So theoretically, in case of 1Ds3 sensor (which resolution is 78 lp/mm), diffraction should affect image quality for apretures smaller than f8.
-
It's just the matter of diffraction.
In perfect, diffraction-limited lens, using ideal values for the Rayleigh limit, and applying the contrast theory of Kühler, resolution for the average wavelength of light (0,555 micron) is:
Aperture resolution
1,4 550 lp/mm
2,0 385 lp/mm
2,8 263 lp/mm
4,0 185 lp/mm
5,6 135 lp/mm
8,0 94 lp/mm
11 69 lp/mm
16 48 lp/mm
22 30 lp/mm
32 21 lp/mm
So theoretically, in case of 1Ds3 sensor (which resolution is 78 lp/mm), diffraction should affect image quality for apretures smaller than f8.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=206456\")
Just for anyone unaware of this topic, you can read about it here:
[a href=\"http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm]http://www.cambridgeincolour.com/tutorials...photography.htm[/url]
Given the above quote, the "practical" diffraction is dependent upon print size or monitor size and distance viewed from image. So you could use f16 and get an image where your eyes cannot see the diffraction, such as shooting a 1ds3 at f16 and printing it in a magazine at 4x6 inches (or whatever size cancels the eye perceiving the diffraction). From my experience, and from reading, the sweet spot is virtually always between around 5.6 and 11, and most full frame cameras, and maybe even MF (If I recall correctly) is around f8--please correct if wrong. It's just what I remember and what I go by.
Since most people here are interested in landscapes, this would mean figuring out from the above link what your camera's best aperture is, and using it when you can, where "when you can" is the operative phrase. For instance, if you're aiming at the ground 3 feet in front of you and you want the ground and background mountains crisp, you won't be using f8 at 30mm (It may not happen whatever you do, but you get my point).
Diffraction simply means, for practical application, that you will be getting a softer image, all things being equal, which they are not.
---------
The reason I started this thread is because I am interested in what people had to say about diffraction issues and the comparison between the 20D, 5D, and 1DS MKIII. The reason I am interested in this topic is because I do a little product photography, and the higher the f stop the better when rendering an entire product in focus. If the above is true, then I'm wondering if the 5D would be a better camera for product photography than the 1DS3 in the studio, where DoF and sharpness are most important, in respect to print size and detail quality?
I just wanted to be clear about what I was thinking on this topic and what my interest is for posting it.
-
Yes, 5d is a better camera for product image, I can work with 16 aperture on 5d, 1dsIII is only until 11 or 8, as for crisp image with small apertures - 5d or even full frame with 6mp like light Phase
-
Doesn't this all depend on out put size and not 100% screen viewing for lpm? I mean smaller denser pixel count ceases to be denser if it has to be stretched further right? Same with D of F.
Just my feeling, I stopped worrying about the mathematics of photography when I stopped accounting for bellows extension.
Kevin.
-
I stopped worrying about the mathematics of photography when I stopped accounting for bellows extension.
VERY WELL SAID.
-
Yes, 5d is a better camera for product image, I can work with 16 aperture on 5d, 1dsIII is only until 11 or 8
[a href=\"index.php?act=findpost&pid=206602\"][{POST_SNAPBACK}][/a]
Not really: for equal sized prints at f/16 (or any aperture) the 1DsIII will give sharper images: equal diffraction effects (l/mm), higher sensor resolution (l/mm).
Diffraction merely limits the improvement in resolution that you get at a given f-stop from an increase in sensor resolution, it never decreases resolution or sharpness in any real sense like "lines per mm", in either the image recorded by the sensor at the focal plane or on prints of equal size.
-
Not really: for equal sized prints at f/16 (or any aperture) the 1DsIII will give sharper images: equal diffraction effects (l/mm), higher sensor resolution (l/mm).
Diffraction merely limits the improvement in resolution that you get at a given f-stop from an increase in sensor resolution, it never decreases resolution or sharpness in any real sense like "lines per mm", in either the image recorded by the sensor at the focal plane or on prints of equal size.
[a href=\"index.php?act=findpost&pid=206672\"][{POST_SNAPBACK}][/a]
Bingo. If you're always shooting at f/22 the 1DsMk3 may not provide any resolution advantage in actual output, but to think that it will actually be worse than the 5D makes no sense.
-
The reason I am interested in this topic is because I do a little product photography, and the higher the f stop the better when rendering an entire product in focus. If the above is true, then I'm wondering if the 5D would be a better camera for product photography than the 1DS3 in the studio, where DoF and sharpness are most important, in respect to print size and detail quality?
I just wanted to be clear about what I was thinking on this topic and what my interest is for posting it.
Wouldn't a 90mm TS-E lens solve whatever DoF issues there might be in dealing with the diffraction of the 1Ds3? A 1Ds3 + 90mm TS-E for product shots seems like the best of all worlds.
CHeers!
-
This website has produced or referenced a considerable amount of material on these issues recently. For example you may wish to have a look at my recent contribution:Noise About Noise (http://www.luminous-landscape.com/tutorials/noise.shtml). While this was in the works, Michael published: Martinec (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html); and also see more recently Osuna and Garcia (http://luminous-landscape.com/tutorials/resolution.shtml)
The diffraction effect of narrow apertures is a lens issue and completely separate from the resolution of a sensor. As for the latter, yes the 1Dsmk3 has smaller photosites than a 1Ds or 5d, but the image detail using the same lens and the same f/stop is AT LEAST as good, as far as I can see from all the testing I've done.
-
Not really: for equal sized prints at f/16 (or any aperture) the 1DsIII will give sharper images: equal diffraction effects (l/mm), higher sensor resolution (l/mm).
Diffraction merely limits the improvement in resolution that you get at a given f-stop from an increase in sensor resolution, it never decreases resolution or sharpness in any real sense like "lines per mm", in either the image recorded by the sensor at the focal plane or on prints of equal size.
[a href=\"index.php?act=findpost&pid=206672\"][{POST_SNAPBACK}][/a]
BJL,
Could you provide further explanation for this point of view.
If we accept that at F16 all good 35mm lenses are very substantially diffraction limited (it wouldn't be too inaccurate to state that all lenses are equal at F16), then it's clear, as you state, that diffraction at F16 will limit any improvement in resolution from an increase in sensor resolution.
Yet you also state that at F16 the 1Ds3 will still provide sharper images (than the 5D, for example) . I agree that a higher resolution sensor will never provide less resolution, whatever the F stop. However, that it might still provide more resolution despite this strong diffraction limitation at F16 is interesting.
Since I already have a bunch of Minolta lenses, there's a strong possibility I might buy the Sony 24mp A900 that will probably be shown at Photokina this year and available before the end of the year.
It would be interesting to see some comparisons at F16, between the 5D (or D3) and the 1Ds3. I suspect that any resolution advantage of the 1Ds3 at F16 would be insignificant at any print size, after appropriate interpolation and sharpening.
-
Yet you also state that at F16 the 1Ds3 will still provide sharper images (than the 5D, for example) . I agree that a higher resolution sensor will never provide less resolution, whatever the F stop. However, that it might still provide more resolution despite this strong diffraction limitation at F16 is interesting.
Go read a book on how MTF curves work. System MTF is the product of the MTFs of each component in the system, not the smallest individual component's value. If f/16 limits lens MTF to 50% at a given linear frequency, then increasing sensor MTF from 50% to 60% will still increase image MTF by 5%. There is obviously a principle of diminishing returns involved, but the amount of benefit from upgrading is non-zero. There's also other improvements in play (less noise, etc.) that make the degree of benefit more significant.
-
Go read a book on how MTF curves work. System MTF is the product of the MTFs of each component in the system, not the smallest individual component's value. If f/16 limits lens MTF to 50% at a given linear frequency, then increasing sensor MTF from 50% to 60% will still increase image MTF by 5%. There is obviously a principle of diminishing returns involved, but the amount of benefit from upgrading is non-zero. There's also other improvements in play (less noise, etc.) that make the degree of benefit more significant.
[a href=\"index.php?act=findpost&pid=206846\"][{POST_SNAPBACK}][/a]
Which book do you recommend that deals with sensor MTF response? I've seen many MTF curves of film resolution in my time, but few of sensor resolution.
My impression is that sensor MTF does not fall off so rapidly as film does with increasing resolution, so the basic formula 1/S=1/F+1/L does not necessarilly apply. This is certainly the impression I get when viewing line charts I've shot to compare resolution. The lines remain strong and clear almost to the cut-off point.
However, if the sensor MTF of the 1Ds3 at 40 lp/mm is noticeably greater than that of the 5D or D3, then one would expect a marginally contrastier image from the 1Ds3 at F16.
On the other hand, since it is reported that the 5D has a weaker AA filter than the 1Ds3 and since the D3 has lower pixel noise than the 1Ds3, I'd like to see the comparisons. Seeing is believing.
-
On the other hand, since it is reported that the 5D has a weaker AA filter than the 1Ds3 and since the D3 has lower pixel noise than the 1Ds3, I'd like to see the comparisons. Seeing is believing.
[a href=\"index.php?act=findpost&pid=206850\"][{POST_SNAPBACK}][/a]
Ray, "it is reported" by who? Someone who makes the filters? Someone who knows how to take them apart and appraise them? Someone who was told something by one of these parties?
How do you know the D3 has lower pixel noise than the 1Ds3? I'd be curious to know where this is reported? Much of this depends on exposure, as I demonstrated in my article on this website.
-
Which book do you recommend that deals with sensor MTF response?
Please note I said MTF in general, not sensor MTF. Since there are multiple factors involved--lens, AA filter, sensor MTF, sensor noise, etc.--a comparison should look at all of them.
-
Which book do you recommend that deals with sensor MTF response? I've seen many MTF curves of film resolution in my time, but few of sensor resolution.
My impression is that sensor MTF does not fall off so rapidly as film does with increasing resolution, so the basic formula 1/S=1/F+1/L does not necessarilly apply. This is certainly the impression I get when viewing line charts I've shot to compare resolution. The lines remain strong and clear almost to the cut-off point.
However, if the sensor MTF of the 1Ds3 at 40 lp/mm is noticeably greater than that of the 5D or D3, then one would expect a marginally contrastier image from the 1Ds3 at F16.
On the other hand, since it is reported that the 5D has a weaker AA filter than the 1Ds3 and since the D3 has lower pixel noise than the 1Ds3, I'd like to see the comparisons. Seeing is believing.
[a href=\"index.php?act=findpost&pid=206850\"][{POST_SNAPBACK}][/a]
No need to read a book, only a little thought is required. The sensor can't resolve anything beyond Nyquist, so MTF must drop to essentially zero at that point; it resolves quite well any contrast variation of more than a few pixels in spatial wavelength, and so the sensor MTF will be a steeply decreasing function of spatial frequency starting from say 1/3 to 1/2 Nyquist, and which vanishes around Nyquist.
Thus, all other things being equal, an increase in sensor resolution (via decreasing pixel pitch) will result in the spatial scale corresponding to the Airy disk diameter being pushed further from Nyquist and thus increasing overall system MTF, since the sensor MTF ceases to be the weakest link in the chain. As Jonathan says, at some point the law of diminishing returns sets in, and further decrease in pixel pitch results in only marginal gains in system MTF, because the sensor MTF curve flattens out at about 1/3 to 1/2 Nyquist.
One can also probably formulate these notions in terms of a quantization error of spatial discretization of the signal, which would lead one to a quantitative formulation of sensor MTF. I'll have to think about that at some point...
-
So which camera will provide more detail at f16, the 5D or 1DS3?
-
So which camera will provide more detail at f16, the 5D or 1DS3?
[a href=\"index.php?act=findpost&pid=206901\"][{POST_SNAPBACK}][/a]
I dunno; all other things are not equal (in particular the AA filter).
-
Ray, "it is reported" by who? Someone who makes the filters? Someone who knows how to take them apart and appraise them? Someone who was told something by one of these parties?
How do you know the D3 has lower pixel noise than the 1Ds3? I'd be curious to know where this is reported? Much of this depends on exposure, as I demonstrated in my article on this website.
[a href=\"index.php?act=findpost&pid=206858\"][{POST_SNAPBACK}][/a]
Mark,
I recall Michael commented on this after his initial appraisal of the 1Ds3. Some time later, Jack Flesher reported that he was surprised the 1Ds3 images required so much sharpening, more sharpening than the 5D and 1Ds2, and concluded that the 1Ds3 AA filter was unusually strong.
As regards the pixel noise of the D3, it would have to be less than that of the 1Ds3 otherwise Michael would not have been able to report that D3 images have lower noise than any other 35mm camera on the market.
I understand from Emil Martinec's article on noise that the greater number of 1Ds3 pixels compensates for the slightly higher noise of the individual 1Ds3 pixels so that the 1Ds3 image as a whole has about equal noise to the D3 image.
-
Please note I said MTF in general, not sensor MTF. Since there are multiple factors involved--lens, AA filter, sensor MTF, sensor noise, etc.--a comparison should look at all of them.
[a href=\"index.php?act=findpost&pid=206861\"][{POST_SNAPBACK}][/a]
Jonathan, I too was curious about your commendation to read a book on this. There is tons of (free) material on the web about MTF insofar as it relates to lenses, and some but less on sensor resolution issues, but I guess much like Ray I haven't seen all this brought together from a systemic perspective in a book. But you apparently would think it has given your recommendation to Ray. So if you happen to know the title of a book which does this, it could be of more general interest to a number of us.
-
So which camera will provide more detail at f16, the 5D or 1DS3?
[a href=\"index.php?act=findpost&pid=206901\"][{POST_SNAPBACK}][/a]
Doug, with all the factors involved (lens, scene, pixel pitch, sensor design, AA filter, firmware, blah, blah) I think it's an impossible question to answer without doing empirical tests. From the work I did, frankly I'd be surprised if there were "hit-you-in-the-faces" differences at f/16 using say a lens with maximum aperture of f/2 ~ f/5.6. The main advantage of the 21 MP sensor is the additional 35% or so resolution - nice elbow room for cropping when we can't fill the frame with the desired composition at capture.
-
On the other hand, since it is reported that the 5D has a weaker AA filter than the 1Ds3 and since the D3 has lower pixel noise than the 1Ds3, I'd like to see the comparisons. Seeing is believing.
[a href=\"index.php?act=findpost&pid=206850\"][{POST_SNAPBACK}][/a]
Surely that would not make sense and would make having much higher pixel density pointless. I suspect that what the report means is that when images are viewed at 100%, the effect of the 1DsIII AA filter is stronger. That would make much more sense.
People tend to confuse viewing an image at 100%, and the appearance at a given print size.
-
Mark,
I recall Michael commented on this after his initial appraisal of the 1Ds3. Some time later, Jack Flesher reported that he was surprised the 1Ds3 images required so much sharpening, more sharpening than the 5D and 1Ds2, and concluded that the 1Ds3 AA filter was unusually strong.
As regards the pixel noise of the D3, it would have to be less than that of the 1Ds3 otherwise Michael would not have been able to report that D3 images have lower noise than any other 35mm camera on the market.
I understand from Emil Martinec's article on noise that the greater number of 1Ds3 pixels compensates for the slightly higher noise of the individual 1Ds3 pixels so that the 1Ds3 image as a whole has about equal noise to the D3 image.
[a href=\"index.php?act=findpost&pid=206941\"][{POST_SNAPBACK}][/a]
Well, where do the results I reported in my Noise article stand in relation to these observations about D3 vs 1Ds3 noise? (Hint: "it depends"...............)
As well, you may recall Michael's first comment on the 1ds3 AA filter was not the last word on that matter.
As for Jack Flesher's observation about the amount of sharpening needed for 1Ds3 images - this of course is rather subjective so I'm not disputing Jack's findings - especially as he is referencing the 1DsMk2 which I haven't tested myself, but I find my 1Ds3 images sharpen just as well as my previous 1Ds images did using the default settings of PK Sharpener Pro - i.e. I haven't found myself wanting to increase the opacities of those sharpening layers for the 1Ds3. And I recall at the time the 1DsMk2 appeared, some users complained the images weren't quite as sharp as the old 1Ds images.................etc., etc..
-
Jonathan, I too was curious about your commendation to read a book on this. There is tons of (free) material on the web about MTF insofar as it relates to lenses, and some but less on sensor resolution issues, but I guess much like Ray I haven't seen all this brought together from a systemic perspective in a book. But you apparently would think it has given your recommendation to Ray. So if you happen to know the title of a book which does this, it could be of more general interest to a number of us.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=206953\")
[a href=\"http://www.normankoren.com/Tutorials/MTF.html]Norman Koren [/url]discusses some of these factors in a tutorial on his web site. The classical additive reciprocal formula for combining MTFs works only at low contrast, around 10%, and does not apply to a more useful contrast such as 50%. If you had the MTF50s for the various components in a system, you couldn't just multiply the individual values to obtain the system MTF 50 since the individual component MTFs above and below 50% also affect the system response at 50%.
As Norman explains, you would need to perform a Fourier transform to separate the individual MTFs into the frequency domain and multiply these components. Having done this, you would then need to convert back to the spatial domain using a complicated process known as convolution. The process would be further complicated by the fact that the MTF of real lenses is different in the meridional and saggital planes.
In practice, it is simpler to determine the system MTF 50 by observation using such tools as Imatest. For 35mm style digital, many such results are posted on PhotoZone. For example, Klaus has tested the Zeiss 50 mm f/2.0 Makro-Planar (http://www.photozone.de/Reviews/46-nikon--nikkor-aps-c/259-zeiss-zf-makro-planar-t-50mm-f2-review--test-report) lens on the Nikon D200.
This plot incorporates the test results and also shows the MTF 50 for an ideal lens as well as the Nyquist limit of the camera. Optimal MTF for this excellent lens is at f/4.0. For other lenses, the optimum may be at f/5.6 or even f/8. Beyond that, diffraction is the limiting factor.
Bill
[attachment=7398:attachment]
-
I'm travelling in Northern Queensland at the moment, sheltering from the colder winter of Brisbane and taking a few photos. I'm using a 5D as my primary camera and the 40D as back-up. The 40D is also attached most of the time to my 100-400 zoom. At F8 it should provide sharper and more detailed results than my 5D when a 600mm lens is required with full frame.
If the 40D were full frame, it would be a 26mp camera with a slightly greater pixel count than the 1Ds3 and pretty close to the much anticipated Sony A900.
Since I have my eye on the 24mp Sony as a possible future purchase, I thought I would do a bit of testing and comparisons between the 5D and 40D whilst out shooting today.
I used the 100-400 at 400mm, the TS-E 90/2.8, the 50/1.4 and the Sigma 15-30, all at F16, and other apertures, and all at the same focal length and from the same position.
It's not looking good. My suspicions so far are confirmed. You can't get more resolution than the 5D at F16, no matter how many pixels, unless you increase sensor size, or remove the AA filter (perhaps).
The following scene was chosen for its low contrast. I upressed the 5D shot, cropped to the 40D FoV, to the same file size as the 40D. I applied no further sharpening after upressing. Both images were converted with mostly identical settings, default sharpening, 50 clarity, 30 vibrance, but I forgot to equalise WB.
I'm searching for any low contrast detail in the 40D image that is not present in the upressed 5D image. I can't find any, but it will take some time to work through all the comparison shots I took.
Here's the first comparison, at 400mm and F16.
[attachment=7399:attachment] [attachment=7400:attachment]
I'm working on an uncalibrated laptop, so apologies if the color sucks or the brightness/contrast is off.
-
In practice, it is simpler to determine the system MTF 50 by observation using such tools as Imatest.
Bill
[attachment=7398:attachment]
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=206973\")
Interesting - as you were writing (I think) I was looking at Imatest - some very interesting results there - though dated - still relevant to this discussion:
[a href=\"http://www.imatest.com/docs/sharpness_comparisons.html]Imatest Results 1Ds, 1DsMk2, etc[/url]
-
I'm travelling in Northern Queensland at the moment, sheltering from the colder winter of Brisbane and taking a few photos. I'm using a 5D as my primary camera and the 40D as back-up. The 40D is also attached most of the time to my 100-400 zoom. At F8 it should provide sharper and more detailed results than my 5D when a 600mm lens is required with full frame.
If the 40D were full frame, it would be a 26mp camera with a slightly greater pixel count than the 1Ds3 and pretty close to the much anticipated Sony A900.
Since I have my eye on the 24mp Sony as a possible future purchase, I thought I would do a bit of testing and comparisons between the 5D and 40D whilst out shooting today.
I used the 100-400 at 400mm, the TS-E 90/2.8, the 50/1.4 and the Sigma 15-30, all at F16, and other apertures, and all at the same focal length and from the same position.
It's not looking good. My suspicions so far are confirmed. You can't get more resolution than the 5D at F16, no matter how many pixels, unless you increase sensor size, or remove the AA filter (perhaps).
The following scene was chosen for its low contrast. I upressed the 5D shot, cropped to the 40D FoV, to the same file size as the 40D. I applied no further sharpening after upressing. Both images were converted with mostly identical settings, default sharpening, 50 clarity, 30 vibrance, but I forgot to equalise WB.
I'm searching for any low contrast detail in the 40D image that is not present in the upressed 5D image. I can't find any, but it will take some time to work through all the comparison shots I took.
Here's the first comparison, at 400mm and F16.
[attachment=7399:attachment] [attachment=7400:attachment]
I'm working on an uncalibrated laptop, so apologies if the color sucks or the brightness/contrast is off.
[a href=\"index.php?act=findpost&pid=206977\"][{POST_SNAPBACK}][/a]
Ray, from what you describe above it's not clear to me you are doing this, but in case you aren't, let me suggest you'd get more definitive answers to your testing doing things (both capture and post-capture) one variable at a time, all else equal, and using test scenes which have the kind of surface texture and edges allowing you to get a quite reliable visual impression of sharpness and resolution.
Also, there may be issues trying to infer the quality of a 24MP Sony image from a Canon D40 sensor as more factors than MP determine pixel quality. If you're not in a panic and you want to be sure you're making the right purchase, wait till the 24MP beast hits the market and gets tested by the usual gurus - then you'll know "for sure".
I found it hard to assess the images you posted, because of the subject matter and their size at my display resolution (1200*1600).
-
If the 40D were full frame, it would be a 26mp camera with a slightly greater pixel count than the 1Ds3 and pretty close to the much anticipated Sony A900. Since I have my eye on the 24mp Sony as a possible future purchase, I thought I would do a bit of testing and comparisons between the 5D and 40D whilst out shooting today.
I used the 100-400 at 400mm, the TS-E 90/2.8, the 50/1.4 and the Sigma 15-30, all at F16, and other apertures, and all at the same focal length and from the same position.
It's not looking good. My suspicions so far are confirmed. You can't get more resolution than the 5D at F16, no matter how many pixels, unless you increase sensor size, or remove the AA filter (perhaps).
I'm searching for any low contrast detail in the 40D image that is not present in the upressed 5D image. I can't find any, but it will take some time to work through all the comparison shots I took.
Here's the first comparison, at 400mm and F16.
[a href=\"index.php?act=findpost&pid=206977\"][{POST_SNAPBACK}][/a]
I'm not sure what you're looking at, the 40D crop on the second pair has loads more detail than the uprezzed 5D shot.
The other question to ask yourself is whether you always intend to shoot at apertures narrower than f8. A 5µ sensel is not going to be diffraction limited at f5.6 or wider, and there will be a definite benefit to the higher resolution.
-
I'm not sure what you're looking at, the 40D crop on the second pair has loads more detail than the uprezzed 5D shot.
The other question to ask yourself is whether you always intend to shoot at apertures narrower than f8. A 5µ sensel is not going to be diffraction limited at f5.6 or wider, and there will be a definite benefit to the higher resolution.
[a href=\"index.php?act=findpost&pid=206989\"][{POST_SNAPBACK}][/a]
Part of the problem is the number of variables. Up-rezzing itself introduces yet another variable.
-
Interesting - as you were writing (I think) I was looking at Imatest - some very interesting results there - though dated - still relevant to this discussion:
Imatest Results 1Ds, 1DsMk2, etc (http://www.imatest.com/docs/sharpness_comparisons.html)
[a href=\"index.php?act=findpost&pid=206980\"][{POST_SNAPBACK}][/a]
Interesting indeed. If we compare the 1Ds2 with the 20D using the DPR data analyzed by Koren (so same lens (50mm) at same aperture (f9) and hopefully other testing parameters similar), the 20D has 23% higher resolution in lp/mm, despite having only 12% smaller pixel pitch.
-
Interesting indeed. If we compare the 1Ds2 with the 20D using the DPR data analyzed by Koren (so same lens (50mm) at same aperture (f9) and hopefully other testing parameters similar), the 20D has 23% higher resolution in lp/mm, despite having only 12% smaller pixel pitch.
[a href=\"index.php?act=findpost&pid=206993\"][{POST_SNAPBACK}][/a]
Yes, and that I think raises a more general proposition about ranges and limits. Within certain ranges of pixel pitch, one can generalize that more PPI (higher MP per sensor size) will outresolve fewer PPI despite the incremental downsides of somewhat smaller pixels, the question being at what lower limit of pixel pitch, all else equal (if it ever is in practice) does this become the decisive variable.
-
Ray, from what you describe above it's not clear to me you are doing this, but in case you aren't, let me suggest you'd get more definitive answers to your testing doing things (both capture and post-capture) one variable at a time, all else equal, and using test scenes which have the kind of surface texture and edges allowing you to get a quite reliable visual impression of sharpness and resolution.
Also, there may be issues trying to infer the quality of a 24MP Sony image from a Canon D40 sensor as more factors than MP determine pixel quality. If you're not in a panic and you want to be sure you're making the right purchase, wait till the 24MP beast hits the market and gets tested by the usual gurus - then you'll know "for sure".
I found it hard to assess the images you posted, because of the subject matter and their size at my display resolution (1200*1600).
[a href=\"index.php?act=findpost&pid=206985\"][{POST_SNAPBACK}][/a]
Mark,
This is just one shot; the first scene I used for a test, on my way up the mountain. I'll proceed to post additional shots. If I don't find any difference of significance in any of my comparisons, I'll shoot newspapers. Okay!
-
Okay! The next comparison of the top of Wallaman Falls near Ingham (the tallest waterfall in Australia, incidentally) was taken with the Canon 50/1.4 at F16.
I've concentrated on the rocks in sunlight. 100% crops still show no greater detail from the 40D.
[attachment=7401:attachment] [attachment=7402:attachment]
-
Mark,
This is just one shot; the first scene I used for a test, on my way up the mountain. I'll proceed to post additional shots. If I don't find any difference of significance in any of my comparisons, I'll shoot newspapers. Okay!
[a href=\"index.php?act=findpost&pid=207018\"][{POST_SNAPBACK}][/a]
Looking forward to your future findings Ray!
-
Okay! The next comparison of the top of Wallaman Falls near Ingham (the tallest waterfall in Australia, incidentally) was taken with the Canon 50/1.4 at F16.
I've concentrated on the rocks in sunlight. 100% crops still show no greater detail from the 40D.
[a href=\"index.php?act=findpost&pid=207022\"][{POST_SNAPBACK}][/a]
OK, now you're into the kind of subject matter that makes it easier to see what's going on or not going on. The issue here, going back to one of my previous posts (discussing ranges and limits), however, is whether by using f/16 the diffraction of the lens has become the binding constraint, such that differences between the sensors get over-ridden by the limitation of the lens. It would be, perhaps, more telling to see a comparison of these rock-face shots if the aperture were somewhere in the range of f/2.8~ f/4 (based on the "rule-of-thumb" that optimal image quality resides about 2 f/stops above the maximum aperture - maybe someone else has better data on the optimal f/stop range for this lens).
-
Interesting - as you were writing (I think) I was looking at Imatest - some very interesting results there - though dated - still relevant to this discussion:
Imatest Results 1Ds, 1DsMk2, etc (http://www.imatest.com/docs/sharpness_comparisons.html)
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=206980\")
[a href=\"http://www.imx.nl/photo/technique/technique/page40.html]Erwin Putts[/url] has posted some interesting test results showing image sharpness with and without image stabilization. He compares the Imatest resolution handheld vs that obtained with the use of a tripod.
It would be interesting to extend these observations to a high resolution camera such as the 1DsMIII to determine how much sharpness you can actually get in the field with hand-held and image stabilization. IOW, can you really make use of the resolution of this camera under field conditions?
Bill
-
Erwin Putts (http://www.imx.nl/photo/technique/technique/page40.html) has posted some interesting test results showing image sharpness with and without image stabilization. He compares the Imatest resolution handheld vs that obtained with the use of a tripod.
It would be interesting to extend these observations to a high resolution camera such as the 1DsMIII to determine how much sharpness you can actually get in the field with hand-held and image stabilization. IOW, can you really make use of the resolution of this camera under field conditions?
Bill
[a href=\"index.php?act=findpost&pid=207030\"][{POST_SNAPBACK}][/a]
Good article. I think he's by and large correct. I'm not a great fan of a tripod in field conditions - I find it an encumbrance, and only use it when it is just plain unavoidable. I can't complain about the sharpness of my 1Ds3 shots magnified to 13*19 or with cropping considerably larger, particularly those using the 24~105 lens. Stabilization works VERY well. I have not done a stationary eyeball-to-eyeball test of exactly the same scene handheld vs tripod, but they would be darn close at shutter speeds down to 1/100th with the 24~105 - that I have no doubt about. Using the 70~300 is a different story - hand-holding at the higher focal lengths is dicier, so with that lens I use a tripod below 1/300th or so. Given how clean the 1Ds3 images are with real ETTR exposure, in lower lighting I'd pump the ISO so I could increase the shutter speed before resorting to the tripod. But I know there's a fair bit of skepticism out there about this approach. BTW the images I shot for the Noise article were all done WITH a tripod. For testing it's just plain dumb not to, unless one is testing the need for the tripod !
-
OK, now you're into the kind of subject matter that makes it easier to see what's going on or not going on. The issue here, going back to one of my previous posts (discussing ranges and limits), however, is whether by using f/16 the diffraction of the lens has become the binding constraint, such that differences between the sensors get over-ridden by the limitation of the lens. It would be, perhaps, more telling to see a comparison of these rock-face shots if the aperture were somewhere in the range of f/2.8~ f/4 (based on the "rule-of-thumb" that optimal image quality resides about 2 f/stops above the maximum aperture - maybe someone else has better data on the optimal f/stop range for this lens).
[a href=\"index.php?act=findpost&pid=207025\"][{POST_SNAPBACK}][/a]
Hey! This was incidental testing on my trip to Australia's highest waterfall. There's a limit to the amount of testing I'm prepared to do.
It seems fairly clear, so far, if I buy the 24mp Sony A900 I'm going to have to hone my skills at focus bracketing if I want to improve on the DoF and resolution that the 5D can provide at F16.
-
Okay! The next comparison of the top of Wallaman Falls near Ingham (the tallest waterfall in Australia, incidentally) was taken with the Canon 50/1.4 at F16.
I've concentrated on the rocks in sunlight. 100% crops still show no greater detail from the 40D.
[attachment=7401:attachment] [attachment=7402:attachment]
[a href=\"index.php?act=findpost&pid=207022\"][{POST_SNAPBACK}][/a]
Ray, I'm still confused as to why you say that the 40D has no greater detail. I downloaded your 100% crops and did a spatial frequency analysis. Here is the Fourier transform of a 256x256 portion of the same area from each image:
(http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/dpr/RayLL.png)
Nyquist is out at the edge of the square, low frequencies are in the middle. Black is no power, bright is lots of power. What you see is that the 40D has much more power at high spatial frequencies, indicating more detail (and it's not just noise power, that is a soft grey background that goes all the way out to the edges; I'm talking about the extent of the bright blob above that background). If the 40D were not resolving any more than the 5D, the extent of the bright disk would be the same for the two images. This just reaffirms what my eyes are telling me, that the 40D is getting way more detail.
-
Thanks for this thread. It allowed me to update my ignore user list.
-
Hey! This was incidental testing on my trip to Australia's highest waterfall. There's a limit to the amount of testing I'm prepared to do.
It seems fairly clear, so far, if I buy the 24mp Sony A900 I'm going to have to hone my skills at focus bracketing if I want to improve on the DoF and resolution that the 5D can provide at F16.
[a href=\"index.php?act=findpost&pid=207072\"][{POST_SNAPBACK}][/a]
Ray, take it easy - don't fall off the mountainside. Sounds like a nice place. To keep in mind if I ever get to that part of the world.
Ya - in those conditions I agree - limits to testing - but twirling the aperture ring a few notches ain't a big-time stress factor, even in the upper atmosphere. Anyhow, whatever, however. As you can and want.
I don't believe the sensor is the determinative factor for what you want from good focusing technique - for any quality DSLR. And I remain to be convinced that what you see from the 40D is necessarily a reliable guide to what you will get from a Sony A900. That remains to be seen.
-
I used the 100-400 at 400mm, the TS-E 90/2.8, the 50/1.4 and the Sigma 15-30, all at F16, and other apertures, and all at the same focal length and from the same position.
... You can't get more resolution than the 5D at F16, no matter how many pixels ...
[a href=\"index.php?act=findpost&pid=206977\"][{POST_SNAPBACK}][/a]
Are you saying that none of those lenses shows any improvement in resolution as one opens up from f/16, even with the 40D? If so, I suggest that you need to seriously reconsider your choice of lenses! There is good evidence that in at least one other SLR system, most or all lenses have higher resolution at their optimal apertures (between f/4 and f/8) than is allowed by diffraction alone at f/16.
-
Are you saying that none of those lenses shows any improvement in resolution as one opens up from f/16, even with the 40D? If so, I suggest that you need to seriously reconsider your choice of lenses! There is good evidence that in at least one other SLR system, most or all lenses have higher resolution at their optimal apertures (between f/4 and f/8) than is allowed by diffraction alone at f/16.
[a href=\"index.php?act=findpost&pid=207093\"][{POST_SNAPBACK}][/a]
Exactly. Back to ranges and limits again: what's the binding constraint: whether we're talking a 40D, a 5D, a 1Ds3, a D3, a D300 - the sensor is most unlikely to be the binding constraint, but f/16 probably is, and if so, these comparisons aren't saying much about either absolute or comparative resolution for these sensors.
-
Exactly. Back to ranges and limits again: what's the binding constraint: whether we're talking a 40D, a 5D, a 1Ds3, a D3, a D300 - the sensor is most unlikely to be the binding constraint, but f/16 probably is, and if so, these comparisons aren't saying much about either absolute or comparative resolution for these sensors.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207097\")
Yes, if you go back to that Zeiss D200 plot I posted, the Nyquist of the camera is 82 lp/mm and maximal observed MTF50 was 71 lp/mm at f/4. At f/16 the ideal lens MTF50 is 48 lp/mm (for green light) and the observed MTF50 was 56 lp/mm. The MTF above the ideal is most likely due to sharpening, which has a marked effect on MTF50.
The link below is from my own test using the D200 and Imatest. The results are shown without sharpening (uncorrected) and with standardized sharpening (corrected). If you want to compare lenses or diffraction effects, it is best to disable sharpening. Generally speaking, the best you can do with a Bayer arrar camera is 75-80% of Nyquist. The best results with this lens were at f/5.6 and I suspect the results for Ray's Canon 50mm f/1.4 would be similar. You can look up the results with that lens and the EOS 350 on PhotoZone.
Bill
[a href=\"http://bjanes.smugmug.com/photos/53712086_UaE9x-O-1.gif]http://bjanes.smugmug.com/photos/53712086_UaE9x-O-1.gif[/url]
-
Yes, if you go back to that Zeiss D200 plot I posted, the Nyquist of the camera is 82 lp/mm and maximal observed MTF50 was 71 lp/mm at f/4. At f/16 the ideal lens MTF50 is 48 lp/mm (for green light) and the observed MTF50 was 56 lp/mm. The MTF above the ideal is most likely due to sharpening, which has a marked effect on MTF50.
The link below is from my own test using the D200 and Imatest. The results are shown without sharpening (uncorrected) and with standardized sharpening (corrected). If you want to compare lenses or diffraction effects, it is best to disable sharpening. Generally speaking, the best you can do with a Bayer arrar camera is 75-80% of Nyquist. The best results with this lens were at f/5.6 and I suspect the results for Ray's Canon 50mm f/1.4 would be similar. You can look up the results with that lens and the EOS 350 on PhotoZone.
Bill
http://bjanes.smugmug.com/photos/53712086_UaE9x-O-1.gif (http://bjanes.smugmug.com/photos/53712086_UaE9x-O-1.gif)
[a href=\"index.php?act=findpost&pid=207114\"][{POST_SNAPBACK}][/a]
OK, if one posits that diffraction is setting in somewhere between f8 and f11, shouldn't we attribute the downslope there and beyond to diffraction effects, or is there more going on?
If it's just diffraction, then the uncorrected MTF50 drops by about 1/3 between f11 and f22, while the aperture is dropping by 1/2. The effect of stopping down is to change the size of the Airy disk relative to Nyquist, so if diffraction limitation were a hard cutoff one should see MTF50 drop in proportion to f number rather than more slowly as it does.
Assuming that the limiting effect is diffraction and not any optical aberrations of the lens, if the effect of diffraction on MTF50/Nyquist is dependent only on pixel spacing/Airy disk size, and I don't see offhand why it shouldn't be (the question is how big is the diffraction blur on the scale of the sampling array), then stopping down by two stops should have the same effect as halving the pixel size -- both change the ratio of the pixel spacing to the Airy disk size by the same amount.
Thus halving the pixel size at f11 should have the same effect on MTF50 relative to Nyquist as keeping the same pixel size and doubling f number, and as we have seen the MTF50 goes down by less than a factor of two. This says to me that increasing pixel density even beyond the point where diffraction limitation sets in, will result in a net gain of resolution (as I and others were arguing before), at least for a while. Less than double the resolution, for sure, but more than nothing (the simple linearized model I am suggesting would say that one gets 4/3 the resolving power). But maybe the model is too naive?
-
OK, if one posits that diffraction is setting in somewhere between f8 and f11, shouldn't we attribute the downslope there and beyond to diffraction effects, or is there more going on?
But maybe the model is too naive?
[a href=\"index.php?act=findpost&pid=207128\"][{POST_SNAPBACK}][/a]
Emil,
I don't know if the model is linear or what, but am merely reporting observed results. Since you are the physicist, I would accept your interpretation. In any event, if you stop down to f/16 with the tested camera and lens, you lose MTF. If you look at the results on PhotoZone, you will see that this is generally the case with similar pixel spacings.
Bill
-
and so the sensor MTF will be a steeply decreasing function of spatial frequency starting from say 1/3 to 1/2 Nyquist, and which vanishes around Nyquist.
The power spectrum of natural images decreases in proportion to more or less the inverse of the square of the frequency. The highest value, is of course, the DC entry.
If the power spectrum did not fall we could kiss compression schemes such as wavelet, mpeg, h.264, etc., good bye.
-
I've concentrated on the rocks in sunlight. 100% crops still show no greater detail from the 40D.
[attachment=7401:attachment] [attachment=7402:attachment]
[a href=\"index.php?act=findpost&pid=207022\"][{POST_SNAPBACK}][/a]
Just came across this thread...
Ray, in these shots of the rock face the 40D looks clearly sharper to me. ejmartin's analysis confirms this. Do they not look different to you?
-
The power spectrum of natural images decreases in proportion to more or less the inverse of the square of the frequency. The highest value, is of course, the DC entry.
If the power spectrum did not fall we could kiss compression schemes such as wavelet, mpeg, h.264, etc., good bye.
[a href=\"index.php?act=findpost&pid=207141\"][{POST_SNAPBACK}][/a]
Yes, but I presume that Bill was shooting a test target, which have sharp edges or sinusoidal variation with spectral power out to high frequency. A proper test target used in a proper test methodology should not in and of itself limit the MTF as a function of spatial frequency, otherwise the testing methodology is flawed. Are you saying that Bill's measured MTF50 is inaccurate?
MTF simply tells you how much of the spectral power of the scene is transferred to the recording medium. It is not a function of the spectral power distribution of any particular scene you wish to photograph (including a test chart), and doesn't care whether that power is large or small, it just says how much of that large or small power gets through.
-
Yes, but I presume that Bill was shooting a test target, which have sharp edges or sinusoidal variation with spectral power out to high frequency. A proper test target used in a proper test methodology should not in and of itself limit the MTF as a function of spatial frequency, otherwise the testing methodology is flawed. Are you saying that Bill's measured MTF50 is inaccurate?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207146\")
Imatest uses a [a href=\"http://www.imatest.com/docs/sfr_instructions.html]Slanted Edge Target[/url], which is a standard ISO test method. It is not necessary to have variable frequency bars and the camera to target distance is not critical. If you have the point spread function of one transition, the remaining data can be calculated. Imatest is fairly idiot proof, and there could be some inaccuracy in my results, but the overall trends should be fairly accurate. The newer versions can also use a Siemens Star to calculate MTF along multiple axes.
Users of ImageJ can download a plugin to perform the calculation from that excellent freeware program. You can download some test images from the DPReview tests or I could upload a couple of images if someone is interested in the raw files.
ImageJ Plugin (http://rsbweb.nih.gov/ij/plugins/se-mtf/index.html)
Bill
-
Imatest uses a Slanted Edge Target (http://www.imatest.com/docs/sfr_instructions.html), which is a standard ISO test method. It is not necessary to have variable frequency bars and the camera to target distance is not critical. If you have the point spread function of one transition, the remaining data can be calculated. Imatest is fairly idiot proof, and there could be some inaccuracy in my results, but the overall trends should be fairly accurate. The newer versions can also use a Siemens Star to calculate MTF along multiple axes.
Users of ImageJ can download a plugin to perform the calculation from that excellent freeware program. You can download some test images from the DPReview tests or I could upload a couple of images if someone is interested in the raw files.
ImageJ Plugin (http://rsbweb.nih.gov/ij/plugins/se-mtf/index.html)
Bill
[a href=\"index.php?act=findpost&pid=207153\"][{POST_SNAPBACK}][/a]
Cool! ImageJ is one of my favorite analysis tools.
-
Are you saying that none of those lenses shows any improvement in resolution as one opens up from f/16, even with the 40D? If so, I suggest that you need to seriously reconsider your choice of lenses! There is good evidence that in at least one other SLR system, most or all lenses have higher resolution at their optimal apertures (between f/4 and f/8) than is allowed by diffraction alone at f/16.
[a href=\"index.php?act=findpost&pid=207093\"][{POST_SNAPBACK}][/a]
No. So far, I'm saying that at F16 I can see no worthwhile improvement, but that might be partly due to the inadequacy of my Dell Inspiron 710 laptop.
However, despite my using a laptop, a 100% crop represents a pretty large print.
-
No. So far, I'm saying that at F16 I can see no worthwhile improvement, but that might be partly due to the inadequacy of my Dell Inspiron 710 laptop.
However, despite my using a laptop, a 100% crop represents a pretty large print.
[a href=\"index.php?act=findpost&pid=207175\"][{POST_SNAPBACK}][/a]
Two reasons why you may not be seeing the improvement are (1) diffraction from the use of f/16 for the captures and (2) the resolution limit of your laptop screen. I often see detail in prints (Epson 3800) that don't show as well on my high-res laptop (LaCie 321 1600*1200, resolution 94 PPI - i.e. 1600/17 in. wide).
-
Just came across this thread...
Ray, in these shots of the rock face the 40D looks clearly sharper to me. ejmartin's analysis confirms this. Do they not look different to you?
[a href=\"index.php?act=findpost&pid=207145\"][{POST_SNAPBACK}][/a]
Nick,
Having looked more closely this morning, the 40D crop does appear slightly sharper and more 'integrated', especially in the upper left region.
However, what I'm looking for is more detail; some smidgen or speck that is clearly defined in the 40D shot but not in the 5D shot.
Accutance is something one can create to taste in accordance with intended print size. The 5D crop has been interpolated to more than double its size without further sharpening. It's not ready for print and because it's been interpolated it should, I believe, have more sharpening than the uninterpolated 40D shot, before printing.
From a practical point of view, what counts is the print on the wall after appropriate processing, and I can't test that in my present circumstances.
-
Ray, I'm still confused as to why you say that the 40D has no greater detail. I downloaded your 100% crops and did a spatial frequency analysis. Here is the Fourier transform of a 256x256 portion of the same area from each image:
(http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/dpr/RayLL.png)
Nyquist is out at the edge of the square, low frequencies are in the middle. Black is no power, bright is lots of power. What you see is that the 40D has much more power at high spatial frequencies, indicating more detail (and it's not just noise power, that is a soft grey background that goes all the way out to the edges; I'm talking about the extent of the bright blob above that background). If the 40D were not resolving any more than the 5D, the extent of the bright disk would be the same for the two images. This just reaffirms what my eyes are telling me, that the 40D is getting way more detail.
[a href=\"index.php?act=findpost&pid=207083\"][{POST_SNAPBACK}][/a]
Emil,
No greater detail that I consider significant, as far as I can judge on my laptop and bearing in mind that the upressed 5D crop has not been sharpened.
It looks as though the results might be more conclusive if I shoot newspapers. There's nothing more convincing than legible text.
-
It looks as though the results might be more conclusive if I shoot newspapers. There's nothing more convincing than legible text.
[a href=\"index.php?act=findpost&pid=207185\"][{POST_SNAPBACK}][/a]
Fresh currency notes are a good choice; crisp detail down to very fine scales.
-
I think I finally found what I was looking for. DPReview's new lens tests do resolutions tests for the 70-200/2.8L on both the 5D and 40D out to f32:
http://www.dpreview.com/lensreviews/canon_...m_c16/page4.asp (http://www.dpreview.com/lensreviews/canon_70-200_2p8_is_usm_c16/page4.asp)
A quick eyeball of the results shows that, for a decrease in pixel pitch by a factor of 1.45, the resolution in lp/mm improves at f5.6 by nearly the difference of their Nyquist frequencies; at f8 the 40D has about a 25% advantage, at f11 about a 20% advantage; then the difference narrows to about a 10% advantage for the 40D for all apertures f16 and narrower, all the way out to f32.
So again, the improvement is not as great as the ratio of pixel pitches once diffraction sets in, there is a law of diminishing returns, but one still gets about a quarter of the change of Nyquist in terms of resolution improvement (FWIW -- probably not much -- my crude little model was predicting about 1/3 of the change in Nyquist) even at very narrow apertures.
-
Having looked at a few more tests at various apertures, it's clear the 26mp sensor provides a substantial improvement in detail at F8, as expected, compared with the upressed 5D crop at F8.
At F22, I have to admit that, even after applying a bit of additional sharpening to the interpolated 5D image, the 40D image looks better, slightly finer grained and perhaps even has a greater 3D effect, although I'm still doubtful if such subtleties would be noticed in a normal size print of, say, 22"x33".
A 100% crop of a 26mp image on this laptop represents a print size of about 5 feet by 3 1/2 feet.
Here's the F8 comparison, followed by the F22 comparison with the interpolated 5D crop sharpened 100% at pixel radius 0.8.
[attachment=7412:attachment] [attachment=7413:attachment]
-
I created a replicable, reliable test set with the 40D. The scenery is very exciting, see below (a moskito screen). I made a crop from the center of all full stop apertures of the EF-S 17-55mm f/2.8 at 55mm.
The top row contains the "compound" image (four identical pixels for the four pixels of a Bayer subarray), the bottom row shows it raw pixel for raw pixel; both in 5x magnification.
The threads of the screen are 4 pixels wide at the best resolution. At f/16, and particularly at f/22 one can see JPEG-like artifacts. They are not from JPEG, but from diffraction.
-
I think it's generally accurate to say that the larger the sensor compared to the pixel density the better the camera will resolve, such as comparing the 1D to the 1DS.
So, the question remains: If you had to shoot products in the studio and wanted the best DoF you could get with the best detail, would you use either a 5D or a 1DS3 (given a print size of no more than 12 x 18 or for web usage)? I don't have any other cameras so those options are irrelevant for my practical purposes.
There is something more to this also, unless I'm wrong, but when you shoot a 2D subject parallel to the film plane, such as a currency note or flat laser printed text page, perhaps that is not as telling as it might be.
Could a lower MP camera perhaps have less detail in the areas further away from the focal point when shooting something like a ruler from near to far--you would have the ruler at one inch towards the lens and the other end facing away from the lens, with the HF point at the 12" mark on a 3' yard stick?
-
Could a lower MP camera perhaps have less detail in the areas further away from the focal point when shooting something like a ruler from near to far--you would have the ruler at one inch towards the lens and the other end facing away from the lens, with the HF point at the 12" mark on a 3' yard stick?
[a href=\"index.php?act=findpost&pid=207233\"][{POST_SNAPBACK}][/a]
I think the indications are that the lower MP camera will have less detail at the plane of focus but closer to the same amount of detail away from the plane of focus. Reducing resolution at the plane of focus will have the effect of increasing the perception of DoF if the print is sufficiently large, but I don't think that's a good idea.
I'm very pleased that a camera such as the 24mp A900 is likely to provide at least some improvement in detail at F16 because I use F16 quite a lot. What I intend to examine now is the possibility that my simulated 26mp full frame DSLR at F22 might be close enough in resolution to the 5D at F16 whilst providing the increased DoF that one expects from F22.
-
Having looked at a few more tests at various apertures, it's clear the 26mp sensor provides a substantial improvement in detail at F8, as expected, compared with the upressed 5D crop at F8.
At F22, I have to admit that, even after applying a bit of additional sharpening to the interpolated 5D image, the 40D image looks better, slightly finer grained and perhaps even has a greater 3D effect, although I'm still doubtful if such subtleties would be noticed in a normal size print of, say, 22"x33".
A 100% crop of a 26mp image on this laptop represents a print size of about 5 feet by 3 1/2 feet.
Here's the F8 comparison, followed by the F22 comparison with the interpolated 5D crop sharpened 100% at pixel radius 0.8.
[attachment=7412:attachment] [attachment=7413:attachment]
[a href=\"index.php?act=findpost&pid=207219\"][{POST_SNAPBACK}][/a]
Ray, this set of results is VERY interesting - and also much as one would expect. Both images are clearly sharper at f/8 than at f/22, indicating once again that the theory about diffraction has a practical impact, and at least in the f/8 example, one clearly sees as you say that the 40D outresolves the 5D. However, in the f/22 comparison, it is much harder to see this difference because it is smothered by the diffraction hit in both cases.
PS. The up-rezzing of the 5D crop is an additional factor that may be contributing to its relative unsharpness. Can you judge what the uprezzing may be doing?
-
I think it's generally accurate to say that the larger the sensor compared to the pixel density the better the camera will resolve, such as comparing the 1D to the 1DS.
So, the question remains: If you had to shoot products in the studio and wanted the best DoF you could get with the best detail, would you use either a 5D or a 1DS3 (given a print size of no more than 12 x 18 or for web usage)? I don't have any other cameras so those options are irrelevant for my practical purposes.
There is something more to this also, unless I'm wrong, but when you shoot a 2D subject parallel to the film plane, such as a currency note or flat laser printed text page, perhaps that is not as telling as it might be.
Could a lower MP camera perhaps have less detail in the areas further away from the focal point when shooting something like a ruler from near to far--you would have the ruler at one inch towards the lens and the other end facing away from the lens, with the HF point at the 12" mark on a 3' yard stick?
[a href=\"index.php?act=findpost&pid=207233\"][{POST_SNAPBACK}][/a]
Doug - as both cameras have the same sensor dimensions the lens focal length, aperture and shooting distance will determine the DoF. With aggressive ETTR exposure, all else equal, you should get more detail from the 1Ds3 at apertures in proximity to the optimum for the lens.
-
Ray I hope you don't mind that I modified your images a little
[attachment=7417:attachment]
Cheers,
J
-
Ray I hope you don't mind that I modified your images a little
[attachment=7417:attachment]
Cheers,
J
[a href=\"index.php?act=findpost&pid=207294\"][{POST_SNAPBACK}][/a]
I'm not sure what I'm supposed to conclude here. All I'm seeing from your modification of the 5D image is an image which continues to have less detail resolution than the 40D image, but now has higher accutance than the one that Ray originally posted. Sharpness is not detail.
-
I'm not sure what I'm supposed to conclude here. All I'm seeing from your modification of the 5D image is an image which continues to have less detail resolution than the 40D image, but now has higher accutance than the one that Ray originally posted. Sharpness is not detail.
[a href=\"index.php?act=findpost&pid=207356\"][{POST_SNAPBACK}][/a]
Hi,
Nothing has been done to the images themselves except that now it's easier to compare the 5D shots to the 40D shot as both of the 5D images are seen at the same time (on top of each other). Why there's so little difference between the f8 and f22 5D images is a bit weird and actually it seems the f22 is a bit sharper. And as already concluded, the 40D image has a lot more detail. The point being that when comparing these images, 40D @ f22 has more detail than interpolated 5D @ either f8 or f22.
Cheers,
J
-
Hi,
Nothing has been done to the images themselves except that now it's easier to compare the 5D shots to the 40D shot as both of the 5D images are seen at the same time (on top of each other). Why there's so little difference between the f8 and f22 5D images is a bit weird and actually it seems the f22 is a bit sharper. And as already concluded, the 40D image has a lot more detail. The point being that when comparing these images, 40D @ f22 has more detail than interpolated 5D @ either f8 or f22.
Cheers,
J
[a href=\"index.php?act=findpost&pid=207378\"][{POST_SNAPBACK}][/a]
OK, got it. I was misinterpreting what you were trying to do, sorry for that.
-
OK, got it. I was misinterpreting what you were trying to do, sorry for that.
[a href=\"index.php?act=findpost&pid=207381\"][{POST_SNAPBACK}][/a]
No prob. I wasn't clear myself. But that actually proved my original point which was that it's really futile to compare images by trying to remember what they look like at this hyper-pixel-peeping level. Side-by-side is the way to go.
Anyway, maybe Ray could give some more details how the images were prepared because as mentioned it's quite weird that there are little difference between the f8 and f22 5D images.
Cheers,
J
-
Ray, this set of results is VERY interesting - and also much as one would expect. Both images are clearly sharper at f/8 than at f/22, indicating once again that the theory about diffraction has a practical impact, and at least in the f/8 example, one clearly sees as you say that the 40D outresolves the 5D. However, in the f/22 comparison, it is much harder to see this difference because it is smothered by the diffraction hit in both cases.
PS. The up-rezzing of the 5D crop is an additional factor that may be contributing to its relative unsharpness. Can you judge what the uprezzing may be doing?
[a href=\"index.php?act=findpost&pid=207259\"][{POST_SNAPBACK}][/a]
Mark,
I'll have to wait till I get back to base to check out the impact of these subtle differences in a print. I'm no expert on sharpening techniques. I use Focus Magic quite a lot but I don't have that program on my laptop which, with CS3E is proving to have rather inadequate processing power, and as you pointed out, is probably not revealing everything I would see on a desktop monitor. The graphics uses shared memory, for example, and I've noticed when printing from this laptop I get more detail in the highlights on the print than I see on the laptop screen.
Glad you find the results interesting .
-
No prob. I wasn't clear myself. But that actually proved my original point which was that it's really futile to compare images by trying to remember what they look like at this hyper-pixel-peeping level. Side-by-side is the way to go.
Anyway, maybe Ray could give some more details how the images were prepared because as mentioned it's quite weird that there are little difference between the f8 and f22 5D images.
Cheers,
J
[a href=\"index.php?act=findpost&pid=207385\"][{POST_SNAPBACK}][/a]
Juicy,
I've always found the 5D to be insignificantly worse at F16 than at F8 for general landscapes. However, shooting a model's sharp eyelashes might reveal more noticeable differences.
As far as I recall, all images were exposed to the same sharpening procedures in ACR, but the 5D at F22 has had additional sharpening after conversion, of 100% at pixel radius 0.8.
ACR sharpening consisted of 50 clarity, 40 sharpening at pixel radius 1 and detail 60.
I should add that I rarely use F22 with the 5D because I do see a loss of detail compared with F8, which is why I decided to give the F22 image a bit of help with additional sharpening which I think is appropriate after significant interpolation.
-
Ray, just a point on testing technique - I think there are two ways of handling the sharpening issue: (1) don't do any so you see only and exactly what the combination of lens settings and sensor is telling you - i.e. the "intrinsic, raw quality", or (2) sharpen the way you would for a "real world" print and see what that tells you, because in the "real world" this is what you would do to achieve the image you will sell/live with. I think both of them have their place, because they each answer a different question; therefore it is good to do it each way. Needless to say - when you get back to your studio - I'm impressed by how you are managing to do what you've already been doing from a laptop on a mountainside and posting it all here.
-
ACR sharpening consisted of 50 clarity, 40 sharpening at pixel radius 1 and detail 60.
[a href=\"index.php?act=findpost&pid=207431\"][{POST_SNAPBACK}][/a]
Just a point of technique...
There's not much point in using anything less than 100 for detail in these circumstances. Similarly, masking should be 0. For lens and CoC testing why would you want to reduce the detail or mask areas? That would cloud the issues at hand.
BTW and IMHO, I find detail and masking be be essentially useless for any sort of image where full detail is needed, like landscapes. Maybe masking for a noisy high ISO blue sky but reduced detail, I don't think so.
-
Doug - as both cameras have the same sensor dimensions the lens focal length, aperture and shooting distance will determine the DoF. With aggressive ETTR exposure, all else equal, you should get more detail from the 1Ds3 at apertures in proximity to the optimum for the lens.
[a href=\"index.php?act=findpost&pid=207263\"][{POST_SNAPBACK}][/a]
Well what I had in mind was using the same lens on both cameras at f16 at the same distance, which is fairly close for a studio shot, and f16 is not optimum for any Canon lens. So something like shooting a ruler front to back at say 4 feet to the focal point on the ruler, or yard stick. At the focal point, I'd expect the 1DS3 to out resolve the 5D, but what about as the DoF decreases from the focal point in each direction?
-
Well what I had in mind was using the same lens on both cameras at f16 at the same distance, which is fairly close for a studio shot, and f16 is not optimum for any Canon lens. So something like shooting a ruler front to back at say 4 feet to the focal point on the ruler, or yard stick. At the focal point, I'd expect the 1DS3 to out resolve the 5D, but what about as the DoF decreases from the focal point in each direction?
[a href=\"index.php?act=findpost&pid=207514\"][{POST_SNAPBACK}][/a]
You're talking about the comparative resolution of blurred images shot with the same lens, same settings between these two cameras? Have I understood the question? The extent of blur should be identical and I'd be hard-put to imagine visible differences of resolution through the blur between these cameras in the conditions you describe. But why not just test it and see?
-
Ray, just a point on testing technique - I think there are two ways of handling the sharpening issue: (1) don't do any so you see only and exactly what the combination of lens settings and sensor is telling you - i.e. the "intrinsic, raw quality", or (2) sharpen the way you would for a "real world" print and see what that tells you, because in the "real world" this is what you would do to achieve the image you will sell/live with. I think both of them have their place, because they each answer a different question; therefore it is good to do it each way. Needless to say - when you get back to your studio - I'm impressed by how you are managing to do what you've already been doing from a laptop on a mountainside and posting it all here.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207474\")
Mark makes a good point, which is amplified on by Norman Koren (author of Imatest). If you are using the same camera to measure lens performance, you should not use any sharpening. However, if you are comparing different cameras you have to use sharpening. Otherwise, the camera with the strongest blur filter will be at a disadvantage.
Here is an Imatest plot I made with the D200 and the 50 mm f/1.8 lens at f/5.6 I used autofocus and there may be some focusing error since the maximal MTF is not quite as good as one would expect. The plot shows MTF with and without sharpening. Without sharpening, the MTF falls rapidly with increasing frequency, but with sharpening, the MTF at lower frequencies is improved. At the highest frequencies, the sharpening makes little difference.
[attachment=7426:attachment]
SQF (subjective quality factor) adds information, since the perceived image quality is affected by relatively low frequencies. For example, [a href=\"http://bobatkins.com/photography/technical/mtf/mtf4.html]Bob Atkins[/url] has calculated that for an 8 by 12 inch print made from a 35mm negative and with normal viewing conditions, the most important MTFs on the film are 4-16 lp/mm. For a 16 by 24 inch print, the corresponding MTFs would be 8-32 lp/mm.
This is an Imatest plot for SQF with the D200 and the same conditions.
[attachment=7427:attachment]
Bill
-
So, the question remains: If you had to shoot products in the studio and wanted the best DoF you could get with the best detail, would you use either a 5D or a 1DS3 (given a print size of no more than 12 x 18 or for web usage)? I don't have any other cameras so those options are irrelevant for my practical purposes.
You want better DoF and prints only at 12x18, and get to use as much light as you want?
You might seriously consider a Rebel XSi. No wait, hear me out.
12MP is about as many pixels as the 5D, the bare resolution of the 5D is clearly potentially acceptable to you, the 1.6x crop sensor gives you a better DOF at the same "effective focal length" and aperture, I seem to recall means you get a 1.6x increase in DOF.
Ta da!
-
12MP is about as many pixels as the 5D, the bare resolution of the 5D is clearly potentially acceptable to you, the 1.6x crop sensor gives you a better DOF at the same "effective focal length" and aperture, I seem to recall means you get a 1.6x increase in DOF.
Sorry, this was kind of muddled.
My point was you say you want two things, best resolution in general *and* best DOF. If the former is good enough with a 5D, its' good enough with a Rebel XSi, and the Rebel XSi leaves you a grand or more to buy a 90 TSe with in addition, *and* a free 1.6x increase in depth of field at the same aperture and effective focal length.
What's not to like?
--Joe
-
You're talking about the comparative resolution of blurred images shot with the same lens, same settings between these two cameras? Have I understood the question? The extent of blur should be identical and I'd be hard-put to imagine visible differences of resolution through the blur between these cameras in the conditions you describe. But why not just test it and see?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207568\")
Actually, I answered this question earlier in the thread, but nobody picked up on it:
[a href=\"http://luminous-landscape.com/forum/index.php?showtopic=26420&view=findpost&p=207216]http://luminous-landscape.com/forum/index....ndpost&p=207216[/url]
(BTW, there is a nice feature to put the FF and APS-C side-by-side for comparison). Now, the comparison there is between the 40D and 5D, but having these two data points I think one can extrapolate to the in-between pixel size of the 1Ds3.
I also did a similar analysis for Nikon in a DPR thread:
http://forums.dpreview.com/forums/read.asp...essage=28590555 (http://forums.dpreview.com/forums/read.asp?forum=1021&message=28590555)
The upshot is that decreasing the pixel pitch increases the resolution, even well into the diffraction-dominated regime; it's just that the resolution gain is rather less than the ratio of pixel pitches.
-
Actually, I answered this question earlier in the thread, but nobody picked up on it:
[a href=\"index.php?act=findpost&pid=207683\"][{POST_SNAPBACK}][/a]
Emil, I read your post but I didn't pick up on it in relation to Doug's latest question because he SEEMS (if I understood correctly) to be concerned about the resolution of out-of-focus image areas and this makes me wonder what the point is - seems to me that the resolution at which one sees these out of focus regions is hardly determinative of image quality.
-
Emil, I read your post but I didn't pick up on it in relation to Doug's latest question because he SEEMS (if I understood correctly) to be concerned about the resolution of out-of-focus image areas and this makes me wonder what the point is - seems to me that the resolution at which one sees these out of focus regions is hardly determinative of image quality.
[a href=\"index.php?act=findpost&pid=207691\"][{POST_SNAPBACK}][/a]
I see, I didn't read his post carefully enough. I agree with what you say.
-
Yes, but I presume that Bill was shooting a test target, which have sharp edges or sinusoidal variation with spectral power out to high frequency. A proper test target used in a proper test methodology should not in and of itself limit the MTF as a function of spatial frequency, otherwise the testing methodology is flawed. Are you saying that Bill's measured MTF50 is inaccurate?
In theory, the MTF of a sensor can't be defined because such a system is not shift invariant. I know that even in this case people like to call the Fourier response of the sampled output of a point input as MTF, though it is not correct. However, solutions do exist that provide rigor to such traditional MTF methods by taking into account the shift-variant dependences that sampling produces by defining shift-invariant sampling MTF as an ensemble average over all possible positions of the scene with respect to the sampling locations.
-
You're talking about the comparative resolution of blurred images shot with the same lens, same settings between these two cameras? Have I understood the question? The extent of blur should be identical and I'd be hard-put to imagine visible differences of resolution through the blur between these cameras in the conditions you describe. But why not just test it and see?
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207568\")
No. What I mean is when you shoot at the HFD about 1/3 into the subject for a given lens length and aperture (or whatever exact calculation you get using charts), you expect to get "reasonable" focus 1/2 the distance in front of the HFD measured from the camera to the HFD and to infinity in back of the HFD, for each focal length and aperture--as you know. This is the "DoF" for that given camera, lens length, and aperture. Again, as you know.
So my question is within that DoF for any lens and focal length together with any aperture--and we're talking f16, not shallow DoF apertures--which camera will give the best detail in the areas in front and back of the HFD, the 5D or 1DS3?
A real word example:
You shoot a widget that is 3 feet long and 2" high at a 45 degree angle using the HFD at f16 and 35mm with X lens on both cameras at 5 feet (or whatever you need to shoot it at to maintain the widget in the DoF. Which camera will give the best detail in the DoF area, the 5D or the 1DS3?
illustration:
[a href=\"http://www.dofmaster.com/hyperfocal.html]http://www.dofmaster.com/hyperfocal.html[/url]
-
No. What I mean is when you shoot at the HFD about 1/3 into the subject for a given lens length and aperture (or whatever exact calculation you get using charts), you expect to get "reasonable" focus 1/2 the distance in front of the HFD measured from the camera to the HFD and to infinity in back of the HFD, for each focal length and aperture--as you know. This is the "DoF" for that given camera, lens length, and aperture. Again, as you know.
So my question is within that DoF for any lens and focal length together with any aperture--and we're talking f16, not shallow DoF apertures--which camera will give the best detail in the areas in front and back of the HFD, the 5D or 1DS3?
A real word example:
You shoot a widget that is 3 feet long and 2" high at a 45 degree angle using the HFD at f16 and 35mm with X lens on both cameras at 5 feet (or whatever you need to shoot it at to maintain the widget in the DoF. Which camera will give the best detail in the DoF area, the 5D or the 1DS3?
illustration:
http://www.dofmaster.com/hyperfocal.html (http://www.dofmaster.com/hyperfocal.html)
[a href=\"index.php?act=findpost&pid=207775\"][{POST_SNAPBACK}][/a]
OK - it's the reverse: anything relatively IN-focus. In this case I think the gist of the discussion in this thread would suggest that notwithstanding the diffraction hit at narrower apertures, the higher resolution sensor will still have an advantage in rendering fine detail.
-
Emil, I read your post but I didn't pick up on it in relation to Doug's latest question because he SEEMS (if I understood correctly) to be concerned about the resolution of out-of-focus image areas and this makes me wonder what the point is - seems to me that the resolution at which one sees these out of focus regions is hardly determinative of image quality.
[a href=\"index.php?act=findpost&pid=207691\"][{POST_SNAPBACK}][/a]
Mark,
I think the issue that Doug seems to be concerned about is the possibility that a 5D image on a certain size print might have an appearance of greater (more extensive) DoF at a particular F stop as a result of that part of the image which is in the plane of focus not being as sharp as it might be if he's used the 1Ds3 at the same F stop.
-
With regard to the sharpening I've applied, I'll reconvert my test images later with no sharpening whatsoever and re-examine the results.
I'd also like to shoot some more test images to compare noise. It's often said that larger pixels have greater dynamic range. I tend to be of the view it's the larger sensor that results in greater DR, whatever the size of the pixel.
I would expect a full frame sensor comprised of 40D pixels to have better DR than the 5D, and probably at least as good as that of the D3.
-
Mark,
I think the issue that Doug seems to be concerned about is the possibility that a 5D image on a certain size print might have an appearance of greater (more extensive) DoF at a particular F stop as a result of that part of the image which is in the plane of focus not being as sharp as it might be if he's used the 1Ds3 at the same F stop.
[a href=\"index.php?act=findpost&pid=207811\"][{POST_SNAPBACK}][/a]
Quite possibly - he'll tell us. As such, I THINK it's answered - go for the higher resolution sensor (i.e. more PPI).
-
With regard to the sharpening I've applied, I'll reconvert my test images later with no sharpening whatsoever and re-examine the results.
I'd also like to shoot some more test images to compare noise. It's often said that larger pixels have greater dynamic range. I tend to be of the view it's the larger sensor that results in greater DR, whatever the size of the pixel.
I would expect a full frame sensor comprised of 40D pixels to have better DR than the 5D, and probably at least as good as that of the D3.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=207814\")
Ray - nothing I've read suggests that sensor size alone is a factor in this. It is the size and design of the photosites and the camera's firmware which affect both dynamic range and the apparent noise.
Once you've done your noise tests it would be interesting to see a comparison of your results with mine [a href=\"http://www.luminous-landscape.com/tutorials/noise.shtml]Noise About Noise[/url]
-
Ray - nothing I've read suggests that sensor size alone is a factor in this. It is the size and design of the photosites and the camera's firmware which affect both dynamic range and the apparent noise.
[a href=\"index.php?act=findpost&pid=207834\"][{POST_SNAPBACK}][/a]
The size of the photosites is rather uncorrelated to image level noise and DR. If we're talking about raw data, firmware is irrelevant. What most people tend to look at however, is pixel level noise and DR, which is of course strongly correlated to pixel size -- we have all seen how much noisier small pixels are than large ones (just look at a digicam and a DSLR at 100% pixel level view, even the 40D and 5D will show a big difference). But when we print an image we are looking at it not at the pixel level but at the image level, and smaller pixels comprise a smaller sample of the image. When pixels are combined, noise goes down; even when pixels are not combined, noise per unit area is largely independent of pixel size for a fixed level of sensor technology.
A fair comparison for the purposes of noise/DR, assuming pixels of a given size were tiling a full frame sensor, would be to perform a blur to a common level of absolute resolution, say 10 microns. So for each image you want to compare, do a gaussian blur by a radius amount 10/pixel pitch in microns (40D - 5.7µ; 5D - 8.4µ; 1Ds3 - 6.4µ; 1D3 - 7.2µ; D3 - 8.45µ; D300 - 5.5µ). Then measure the noise.
You will find that all of these cameras do about the same, with the D3 coming out a little bit ahead; but the differences are far, far less than the differences in pixel-level noise might lead you to believe.
-
Ray - nothing I've read suggests that sensor size alone is a factor in this. It is the size and design of the photosites and the camera's firmware which affect both dynamic range and the apparent noise.
Both DR and SNR roughly increase as the square root of the pixel size. However, it not always immediately clear that what is an optimal pixel size. A small pixel size helps in higher MTF and spatial resolution, where as a large pixel size helps in better DR and SNR. Therefore there must exist a pixel size that strikes a compromise between high DR and SNR on the one hand, and high spatial resolution and MTF on the other. However, it is not always clear how to trade off DR and SNR with spatial resolution and MTF, and especially, how to relate these parameters to image quality.
It is possible to determine that "optimal" pixel size using certain methods that is a joint optimization of DR and SNR with MTF and spatial response. It has also been found that the "optimal" size kind of scales with technology (i.e., becomes smaller with a particular micron technology), however its rate of shrinking is slower than that of the technology. Additionally, image quality using certain measures degrades as technology scales.
-
The size of the photosites is rather uncorrelated to image level noise and DR. If we're talking about raw data, firmware is irrelevant. What most people tend to look at however, is pixel level noise and DR, which is of course strongly correlated to pixel size -- we have all seen how much noisier small pixels are than large ones (just look at a digicam and a DSLR at 100% pixel level view, even the 40D and 5D will show a big difference). But when we print an image we are looking at it not at the pixel level but at the image level, and smaller pixels comprise a smaller sample of the image. When pixels are combined, noise goes down; even when pixels are not combined, noise per unit area is largely independent of pixel size for a fixed level of sensor technology.
A fair comparison for the purposes of noise/DR, assuming pixels of a given size were tiling a full frame sensor, would be to perform a blur to a common level of absolute resolution, say 10 microns. So for each image you want to compare, do a gaussian blur by a radius amount 10/pixel pitch in microns (40D - 5.7µ; 5D - 8.4µ; 1Ds3 - 6.4µ; 1D3 - 7.2µ; D3 - 8.45µ; D300 - 5.5µ). Then measure the noise.
You will find that all of these cameras do about the same, with the D3 coming out a little bit ahead; but the differences are far, far less than the differences in pixel-level noise might lead you to believe.
[a href=\"index.php?act=findpost&pid=207888\"][{POST_SNAPBACK}][/a]
Emil, if the size of the photosite is correlated to the size of the pixel and the size of the pixel is correlated with pixel level noise, then by logical inference the size of the photosite is correlated with the level of noise, ceteris paribus, is it not? The reason why I mentioned firmware is that the camera's firmware - AFAIK - does the analog to digital conversion and the manner in which that happens should have some impact on the qualities of DR and apparent noise when we open the image in Camera Raw - would it not? (i.e. please clarify what I may ne misunderstanding here.)
I don't understand your last para - "do about the same" is in regard to what? The D3 is coming out slightly ahead in respect of what? Does the last part of the statement mean that the differences in noise are far less than the differences in pixel pitch between these cameras?
Sorry for all the questions, but your message just made them come to mind!
-
Ray - nothing I've read suggests that sensor size alone is a factor in this. It is the size and design of the photosites and the camera's firmware which affect both dynamic range and the apparent noise.
Once you've done your noise tests it would be interesting to see a comparison of your results with mine Noise About Noise (http://www.luminous-landscape.com/tutorials/noise.shtml)
[a href=\"index.php?act=findpost&pid=207834\"][{POST_SNAPBACK}][/a]
Mark,
No, of course not. I meant 'primarily', given modern pixels and the latest technology.
Emil has explained very well what I always sensed was the case. It's the one area where the DB has an undeniable advantage over 35mm and, for the same reasons, why the Olympus 4/3rds system will probably never match the DR and low noise of FF 35mm.
Your article, "Noise about Noise", is very thorough but the crops demonstrating resolution at various F stops seem too small to show any significant differences. All I see clearly is that at f22, crops are noticeably softer than at F6.3.
Since my printer is the 24" wide Epson 7600, which I shall probably eventually upgrade to the 7880, for resolution comparisons I should perhaps use a crop enlargement that is representative of a 24"x36" print, since this is the largest print I would make from a single, unstitched image file. But this would entail comparing crops on the monitor at considerably less than 100%.
I suppose this is where the validity of pixel peeping can be questioned. It's necessary to pixel peep in order to find out if differences exist. However, having done that, it's perhaps necessary to make allowances for the significance of such differences in real-world applications, such as the making prints of a particular size.
I'll have to give this matter more thought.
-
OK - it's the reverse: anything relatively IN-focus. In this case I think the gist of the discussion in this thread would suggest that notwithstanding the diffraction hit at narrower apertures, the higher resolution sensor will still have an advantage in rendering fine detail.
[a href=\"index.php?act=findpost&pid=207807\"][{POST_SNAPBACK}][/a]
Yep, that's it. It's good to know and bad at the same time. I was going to leave my 5D in the studio bolted down to a boom and take teh 1DS3 with me. Now I'll have to break it down each time I want to change from studio to location, but on the other hand, it's good to know that an 8, 000US camera will out resolve the 5D--indeed.
Thanks for those who responded, and for those discussing a similar but different topic, good luck to you also.
Unsubscribed
-
Mark,
No, of course not. I meant 'primarily', given modern pixels and the latest technology.
Emil has explained very well what I always sensed was the case. It's the one area where the DB has an undeniable advantage over 35mm and, for the same reasons, why the Olympus 4/3rds system will probably never match the DR and low noise of FF 35mm.
Your article, "Noise about Noise", is very thorough but the crops demonstrating resolution at various F stops seem too small to show any significant differences. All I see clearly is that at f22, crops are noticeably softer than at F6.3.
Since my printer is the 24" wide Epson 7600, which I shall probably eventually upgrade to the 7880, for resolution comparisons I should perhaps use a crop enlargement that is representative of a 24"x36" print, since this is the largest print I would make from a single, unstitched image file. But this would entail comparing crops on the monitor at considerably less than 100%.
I suppose this is where the validity of pixel peeping can be questioned. It's necessary to pixel peep in order to find out if differences exist. However, having done that, it's perhaps necessary to make allowances for the significance of such differences in real-world applications, such as the making prints of a particular size.
I'll have to give this matter more thought.
[a href=\"index.php?act=findpost&pid=207962\"][{POST_SNAPBACK}][/a]
Ray, for clarity in my own mind - and I assume this is how you would see it too- the reason why the DB is advantageous relative to FF35mm and the latter relative to 4:3 is that either or both more/larger pixels can be packed-in as the sensor gets larger - it's not the size of the sensor per se, but what it allows. Hence the interest of a previous post about optimizing the number of photosites and their size for any given sensor size.
Yes, I agree with your observation that in my article the magnification of the images is a bit small to fully appreciate the differences of apparent sharpness at the intermediate f/stops. There was a trade-off here - I didn't want to introduce more variables by resampling or over-magnifying. However, I did make prints of all those cases and the written observations are the result of two pairs of eyes examining the prints closely. Even so, until you get past f/11 the differences are really subtle even on paper.
The latest issue of PhotoshopUser magazine has an interesting article by Deke McClelland on how to make your monitor simulate how the details in an image will look when printed. It's a 10 step procedure which I haven't tried yet, but it looks like the first serious attempt (at least as far as I've seen) to softproof sharpening, which is usually considered very hard to do on a display. This may help you examining resolution as well on the display. But in the final analysis, I would still recommend actually making e.g. A4-size prints using crops at the same PPI (without resampling) as applicable for your largest print size of the full image.
Before thinking too long and hard about an Epson 7880, if you're not itching to renew your printer urgently, you may wish to wait a bit - see Michael's "What's New" for May 30th.
-
Emil, if the size of the photosite is correlated to the size of the pixel and the size of the pixel is correlated with pixel level noise, then by logical inference the size of the photosite is correlated with the level of noise, ceteris paribus, is it not? The reason why I mentioned firmware is that the camera's firmware - AFAIK - does the analog to digital conversion and the manner in which that happens should have some impact on the qualities of DR and apparent noise when we open the image in Camera Raw - would it not? (i.e. please clarify what I may ne misunderstanding here.)
I don't understand your last para - "do about the same" is in regard to what? The D3 is coming out slightly ahead in respect of what? Does the last part of the statement mean that the differences in noise are far less than the differences in pixel pitch between these cameras?
[a href=\"index.php?act=findpost&pid=207940\"][{POST_SNAPBACK}][/a]
The size of the photosite is correlated with the *pixel* level noise, what you see when you view the image at 100% on a monitor. Noise has a scale dependence. Suppose we combine the output of a 2x2 block of photosites. The signal is additive -- one just adds the photon counts from all four. Noise adds as RMS -- the sqrt of the sum of the squares of the pixel noises. This means if all are about the same, signal goes up by four, but the noise only goes up by two, and so the S/N ratio improves if we halve the resolution.
It is easy to see this for yourself. Open a new canvas in photoshop, fill it with middle gray and add a bunch of gaussian noise (Filter>Noise>Add Noise). Now go to the Gaussian blur filter and watch what happens to the width of the histogram as you change the radius of the blur. Increasing the blur radius decreases the image resolution; it also decreases the noise.
So if you properly resample say a 1Ds3 image to the resolution of a D3, the apparent one stop pixel level noise advantage of the D3 drops to something on the order of 20%. That 20% can be understood from the fact that the D3 is 20% more efficient per unit area at collecting photons than the 1Ds3. So the 1Ds3 has higher *pixel* level noise, but *image* level noise -- noise measured at the same spatial scale -- is not all that different. Now, that near equivalence is there whether we resample/blur the 1Ds3 image or not, since the same photons were collected by the sensor whether we bin the pixel samples together or not. Since noise goes as sqrt of signal, that 20% difference in photons is only about a 10% difference in noise amplitude to the 1Ds3's detriment; on the other side of the balance sheet is a 32% increase in linear resolution.
It is thus important when comparing noise figures to compare them at the same scale; any pixel level noise measurement should be divided by the square root of the MP count to normalize relative to a consistent and comparable percentage of the frame size. If one wants to extrapolate -- eg imagine how a full frame sensor tiled with 40D pixels would perform -- one should multiply the pixel level noise of the cameras to be compared by the respective pixel pitches. By this measure, the 40D, 1Ds3, and 1D3 have about the same noise at comparable scales, even though their pixel level noises are quite different (and larger the smaller the photosite is). One way to understand this is that Canon did their job right, and all three sensors are capturing the same number of photons per unit area. BTW, some people have speculated that the reason the D3 does about 20% better is that its microlenses purportedly have about 20% better area coverage.
As for analog-to-digital conversion, that is done in hardware, not firmware; the only way that firmware could have an effect is if the camera is doing some processing of the sensor data before it is written to raw (eg noise reduction), and I have seen no evidence of that under ordinary shooting conditions (not at all in Canons with default settings; Nikon does some NR on raw for exposures of 1/4 sec and longer but not otherwise).
The reason MFDB's do so well is that they have twice the sensor area of 35mm, so gather more photons; the size of individual pixels matters little in this regard.
Sorry for the lengthy diatribe...
-
Before thinking too long and hard about an Epson 7880, if you're not itching to renew your printer urgently, you may wish to wait a bit - see Michael's "What's New" for May 30th.
Mark,
I meant the 7990 due in Australia around October this year; the one with reduced ink costs, more cartridges and easy swapping between Photo Black and Matte Black.
This could be an expensive year with the 5D upgrade and/or Sony A900 out soon; another trip to Nepal, Thailand and Cambodia to test my new camera; and a new printer with increased color gamut, reduced bronzing and easy changing from gloss paper to matte.
I really think that Canon should release a 26mp 5D upgrade in order to trump Sony's offering of a 24mp DSLR, although I guess owners of a 1Ds3 might feel a bit peeved if Canon were to do that.
Ray, for clarity in my own mind - and I assume this is how you would see it too- the reason why the DB is advantageous relative to FF35mm and the latter relative to 4:3 is that either or both more/larger pixels can be packed-in as the sensor gets larger - it's not the size of the sensor per se, but what it allows. Hence the interest of a previous post about optimizing the number of photosites and their size for any given sensor size.
I wonder if this is just semantics. I prefer to see it from the perspective of the composition being photographed. It has a specific FoV and DoF. At a given ISO the sensor will receive an amount of light in proportion to its size, assuming correct exposure. The greater amount of light that the larger sensor requires for proper exposure ensures greater DR, irrespective of pixel size, but not of course irrespective of pixel quality and factors such as quantum efficiency.
The latest issue of PhotoshopUser magazine has an interesting article by Deke McClelland on how to make your monitor simulate how the details in an image will look when printed. It's a 10 step procedure which I haven't tried yet, but it looks like the first serious attempt (at least as far as I've seen) to softproof sharpening, which is usually considered very hard to do on a display. This may help you examining resolution as well on the display. But in the final analysis, I would still recommend actually making e.g. A4-size prints using crops at the same PPI (without resampling) as applicable for your largest print size of the full image.
Thanks. I'll look into that.
-
The size of the photosite is correlated with the *pixel* level noise, what you see when you view the image at 100% on a monitor. Noise has a scale dependence. Suppose we combine the output of a 2x2 block of photosites. The signal is additive -- one just adds the photon counts from all four. Noise adds as RMS -- the sqrt of the sum of the squares of the pixel noises. This means if all are about the same, signal goes up by four, but the noise only goes up by two, and so the S/N ratio improves if we halve the resolution.
It is easy to see this for yourself. Open a new canvas in photoshop, fill it with middle gray and add a bunch of gaussian noise (Filter>Noise>Add Noise). Now go to the Gaussian blur filter and watch what happens to the width of the histogram as you change the radius of the blur. Increasing the blur radius decreases the image resolution; it also decreases the noise.
So if you properly resample say a 1Ds3 image to the resolution of a D3, the apparent one stop pixel level noise advantage of the D3 drops to something on the order of 20%. That 20% can be understood from the fact that the D3 is 20% more efficient per unit area at collecting photons than the 1Ds3. So the 1Ds3 has higher *pixel* level noise, but *image* level noise -- noise measured at the same spatial scale -- is not all that different. Now, that near equivalence is there whether we resample/blur the 1Ds3 image or not, since the same photons were collected by the sensor whether we bin the pixel samples together or not. Since noise goes as sqrt of signal, that 20% difference in photons is only about a 10% difference in noise amplitude to the 1Ds3's detriment; on the other side of the balance sheet is a 32% increase in linear resolution.
It is thus important when comparing noise figures to compare them at the same scale; any pixel level noise measurement should be divided by the square root of the MP count to normalize relative to a consistent and comparable percentage of the frame size. If one wants to extrapolate -- eg imagine how a full frame sensor tiled with 40D pixels would perform -- one should multiply the pixel level noise of the cameras to be compared by the respective pixel pitches. By this measure, the 40D, 1Ds3, and 1D3 have about the same noise at comparable scales, even though their pixel level noises are quite different (and larger the smaller the photosite is). One way to understand this is that Canon did their job right, and all three sensors are capturing the same number of photons per unit area. BTW, some people have speculated that the reason the D3 does about 20% better is that its microlenses purportedly have about 20% better area coverage.
As for analog-to-digital conversion, that is done in hardware, not firmware; the only way that firmware could have an effect is if the camera is doing some processing of the sensor data before it is written to raw (eg noise reduction), and I have seen no evidence of that under ordinary shooting conditions (not at all in Canons with default settings; Nikon does some NR on raw for exposures of 1/4 sec and longer but not otherwise).
The reason MFDB's do so well is that they have twice the sensor area of 35mm, so gather more photons; the size of individual pixels matters little in this regard.
Sorry for the lengthy diatribe...
[a href=\"index.php?act=findpost&pid=207996\"][{POST_SNAPBACK}][/a]
Dont apologise- that was an excellent and easy to understand explanation.
Thanks for posting.
-
The size of the photosite is correlated with the *pixel* level noise, what you see when you view the image at 100% on a monitor. Noise has a scale dependence. ..................
[a href=\"index.php?act=findpost&pid=207996\"][{POST_SNAPBACK}][/a]
Thanks Emil - this is a post to keep in one's archives. Very useful.
-
..............
The greater amount of light that the larger sensor requires for proper exposure ensures greater DR, irrespective of pixel size, ...................................
[a href=\"index.php?act=findpost&pid=208000\"][{POST_SNAPBACK}][/a]
I don't profess any expertise on the subject of how photosites register the tonal range of a scene, but this does not come across as intuitively correct to me. I would like to be educated on that point.
-
I don't profess any expertise on the subject of how photosites register the tonal range of a scene...............[a href=\"index.php?act=findpost&pid=208044\"][{POST_SNAPBACK}][/a]
Nor do I, yet it does seem intuitive to me that a larger sensor requires more light than a smaller sensor requires, in order to record the same scene. If you stitch together two 1Ds3 sensors, you need double the amount of light to fully expose that doubled sensor area, and as a consequence DR is increased by approximately one stop.
In fact Mark, it seems to me that this increase in DR would apply whatever the initial sensor size. Double the area of a 5D sensor and presumably you'd still get approximately one stop increase in DR. Double the area of a G9 or Olympus E3 sensor and you'd expect to get the same increase in DR despite the fact that the pixels are a different size in each case.
-
Nor do I, yet it does seem intuitive to me that a larger sensor requires more light than a smaller sensor requires, in order to record the same scene. If you stitch together two 1Ds3 sensors, you need double the amount of light to fully expose that doubled sensor area, and as a consequence DR is increased by approximately one stop.
In fact Mark, it seems to me that this increase in DR would apply whatever the initial sensor size. Double the area of a 5D sensor and presumably you'd still get approximately one stop increase in DR. Double the area of a G9 or Olympus E3 sensor and you'd expect to get the same increase in DR despite the fact that the pixels are a different size in each case.
[a href=\"index.php?act=findpost&pid=208051\"][{POST_SNAPBACK}][/a]
Ray, I'm skeptical about this. If you need twice the amount of light to cover twice the sensor area, it would seem to me that each photosite is receiving the same amount of light it would have received whether the sensor was half the size or twice the size as long as the relationship between total light and total sensor area remains the same. So how does the DR change? I think we need one of our Forum physicists to step in here!
-
If you need twice the amount of light to cover twice the sensor area, it would seem to me that each photosite is receiving the same amount of light it would have received whether the sensor was half the size
[a href=\"index.php?act=findpost&pid=208072\"][{POST_SNAPBACK}][/a]
Read Emil's excellent explanation above. In brief, at equal exposure index (ISO speed)
- Doubling sensor area with equal pixel count doubles the light per pixel, and so can improve S/N ratio (by about 1/2 stop).
- Doubling sensor area with equal pixel size can given equal per pixel S/N ratio but twice as many pixels: then downsampling to the lower pixel count can increase the S/N raito over what the smaller sensor gives (again) by about 1/2 stop.
Either way one can probably buy an extra half stop of D/R and one stop of usable ISO speed for each doubling of sensor area.
Aside: this gathering of twice as much light requires either (1) a longer exposure time and/or (2) a larger aperture diameter and less DOF, as with equal f-stop and longer focal length. Larger sensors of "equal technology" offer no free lunch in this IQ comparison.
-
Read Emil's excellent explanation above. In brief, at equal exposure index (ISO speed)
- Doubling sensor area with equal pixel count doubles the light per pixel, and so can improve S/N ratio (by about 1/2 stop).
- Doubling sensor area with equal pixel size can given equal per pixel S/N ratio but twice as many pixels: then downsampling to the lower pixel count can increase the S/N raito over what the smaller sensor gives (again) by about 1/2 stop.
Either way one can probably buy an extra half stop of D/R and one stop of usable ISO speed for each doubling of sensor area.
Aside: this gathering of twice as much light requires either (1) a longer exposure time and/or (2) a larger aperture diameter and less DOF, as with equal f-stop and longer focal length. Larger sensors of "equal technology" offer no free lunch in this IQ comparison.
[a href=\"index.php?act=findpost&pid=208097\"][{POST_SNAPBACK}][/a]
BJL,
Is it as little as 1/2 a stop of improved S/N resulting from a doubling of light gathering capacity? If so, why?
If one were to stitch tegether four 1Ds3 sensors, one would get an 84mp sensor of dimensions 72mmx48mm. Are you saying that such a huge sensor would have merely a one stop S/N advantage over the 1Ds3?
I'm not arguing you are wrong. It just seems too conservative a figure, intuitively.
-
BJL,
Is it as little as 1/2 a stop of improved S/N resulting from a doubling of light gathering capacity? If so, why?
If one were to stitch tegether four 1Ds3 sensors, one would get an 84mp sensor of dimensions 72mmx48mm. Are you saying that such a huge sensor would have merely a one stop S/N advantage over the 1Ds3?
I'm not arguing you are wrong. It just seems too conservative a figure, intuitively.
[a href=\"index.php?act=findpost&pid=208267\"][{POST_SNAPBACK}][/a]
Two typical noise sources are present: Photon shot noise and the noise in the electronics that processes the data (so-called read noise). Photon noise rises as the square root of the number of photons collected, so if you quadruple the number of photons collected by stitching together four sensors, you double the amount of photon noise. S/N goes up by 4/2=2 when area goes up by four. It's the same for read noise. Four pixels were doing the same job that one was before; the read noise per pixel is constant, and noise adds as RMS as I mentioned above, and so doubles for the combined four pixels.
So indeed doubling the area (one stop more light) is only a half stop more S/N (since stops are powers of two, and 1/2 stop means square root two).
-
Read Emil's excellent explanation above. In brief, at equal exposure index (ISO speed)
- Doubling sensor area with equal pixel count doubles the light per pixel, and so can improve S/N ratio (by about 1/2 stop).
- Doubling sensor area with equal pixel size can given equal per pixel S/N ratio but twice as many pixels: then downsampling to the lower pixel count can increase the S/N raito over what the smaller sensor gives (again) by about 1/2 stop.
Either way one can probably buy an extra half stop of D/R and one stop of usable ISO speed for each doubling of sensor area.
Aside: this gathering of twice as much light requires either (1) a longer exposure time and/or (2) a larger aperture diameter and less DOF, as with equal f-stop and longer focal length. Larger sensors of "equal technology" offer no free lunch in this IQ comparison.
[a href=\"index.php?act=findpost&pid=208097\"][{POST_SNAPBACK}][/a]
Re your first bullet - OK, but I don't think Ray is talking about doubling the sensor area leaving the pixel count unchanged.
Second bullet OK, but I don't see how this contributes to greater D/R if by D/R you mean Dynamic Range.
-
Two typical noise sources are present: Photon shot noise and the noise in the electronics that processes the data (so-called read noise). Photon noise rises as the square root of the number of photons collected, so if you quadruple the number of photons collected by stitching together four sensors, you double the amount of photon noise. S/N goes up by 4/2=2 when area goes up by four. It's the same for read noise. Four pixels were doing the same job that one was before; the read noise per pixel is constant, and noise adds as RMS as I mentioned above, and so doubles for the combined four pixels.
So indeed doubling the area (one stop more light) is only a half stop more S/N (since stops are powers of two, and 1/2 stop means square root two).
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208329\")
Emil's analysis applies to current dSLRs, but not to sceintific sensors where pixel binning can be performed in hardware before digitization. Consider a CCD with 2 by 2 pixel binning as described [a href=\"http://www.photomet.com/pm_solutions/library_encyclopedia/index.php]here[/url].
Without pixel binning, each pixel has a read noise and if 4:1 downsizing is done in software subsequent to digitization, one still has 4 read noises. However, with 2 by 2 pixel binning, the superpixel can be read with only one read noise, so the S:N is 4:1 as compared to the single pixels.
CCDs typically have a high fill factor, but this factor is considerably in CMOS sensors such as now used in the best 35 mm style DSLRs as explained here (http://www.dalsa.com/sensors/products/dsc.asp). This is due to the extra circuitry added to each pixel for signal processing. If the overall sensor size is held constant and the pixel count is quadrupled, this circuitry is also quadrupled, so the fill factor must decrease as compared to the larger pixel before the increase in the pixel count. Also, there is a certain amount of dead space between pixels which also decreases the fill factor.
I see that Michael has previewed a new Phase One digital back with variable resolution. I would presume this is a CCD with variable pixel binning.
Bill
-
Two typical noise sources are present: Photon shot noise and the noise in the electronics that processes the data (so-called read noise). Photon noise rises as the square root of the number of photons collected, so if you quadruple the number of photons collected by stitching together four sensors, you double the amount of photon noise. S/N goes up by 4/2=2 when area goes up by four. It's the same for read noise. Four pixels were doing the same job that one was before; the read noise per pixel is constant, and noise adds as RMS as I mentioned above, and so doubles for the combined four pixels.
So indeed doubling the area (one stop more light) is only a half stop more S/N (since stops are powers of two, and 1/2 stop means square root two).
[a href=\"index.php?act=findpost&pid=208329\"][{POST_SNAPBACK}][/a]
Hmm! Doesn't seem much, does it! I wonder if I should take the trouble to test this experimentally with the 40D. I could bracket a few exposures at F8 using a 50mm lens, then bracket a few more exposures at F13 using an 80mm lens (same scene, same position), then examine details in the shadows and match the appropriate overexposure at F8 and 50mm that visually has the same amount of shadow noise as the correct exposure at F13 and 80mm (or what would have been the correct exposure if the sensor had been full frame).
-
Emil's analysis applies to current dSLRs, but not to sceintific sensors where pixel binning can be performed in hardware before digitization. Consider a CCD with 2 by 2 pixel binning as described here (http://www.photomet.com/pm_solutions/library_encyclopedia/index.php).
Without pixel binning, each pixel has a read noise and if 4:1 downsizing is done in software subsequent to digitization, one still has 4 read noises. However, with 2 by 2 pixel binning, the superpixel can be read with only one read noise, so the S:N is 4:1 as compared to the single pixels.
[a href=\"index.php?act=findpost&pid=208338\"][{POST_SNAPBACK}][/a]
I've always assumed it's meaningful to compare equal size files or equal size prints when comparing noise. Is there necessarily much difference in the final noise outcome between binning 4 pixels in hardware as opposed to reducing the image file size to a quarter, or increasing the smaller image file by a factor of 4?
I've been of the opinion that downsampling results in a reduction of noise and upsampling results in an increase in noise.
-
Hmm! Doesn't seem much, does it! I wonder if I should take the trouble to test this experimentally with the 40D. I could bracket a few exposures at F8 using a 50mm lens, then bracket a few more exposures at F13 using an 80mm lens (same scene, same position), then examine details in the shadows and match the appropriate overexposure at F8 and 50mm that visually has the same amount of shadow noise as the correct exposure at F13 and 80mm (or what would have been the correct exposure if the sensor had been full frame).
[a href=\"index.php?act=findpost&pid=208342\"][{POST_SNAPBACK}][/a]
I don't see why one needs to go to all this trouble. Just take a single exposure and see what happens to the noise when you reduce the resolution by binning 2x2 bunches of pixels. What's being considered is local on the sensor. If you want the reduction in noise, you have to give up some resolution; if you then make the sensor bigger, you get the same resolution relative to frame height with the binned pixels.
I've always assumed it's meaningful to compare equal size files or equal size prints when comparing noise. Is there necessarily much difference in the final noise outcome between binning 4 pixels in hardware as opposed to reducing the image file size to a quarter, or increasing the smaller image file by a factor of 4?
I've been of the opinion that downsampling results in a reduction of noise and upsampling results in an increase in noise.
[a href=\"index.php?act=findpost&pid=208348\"][{POST_SNAPBACK}][/a]
The business about pixel binning only refers to read noise; it has no effect on photon noise, which is dominant in midtones and highlights for low to moderate ISO. I think that suggestion is a bit of a tangent to the topic under discussion.
-
Some replies to Mark, Ray and Bill.
To MarkDS:
Second bullet OK, but I don't see how this contributes to greater D/R
[a href=\"index.php?act=findpost&pid=208336\"][{POST_SNAPBACK}][/a]
Because averaging the signal from several photosites (by binning or downsampling) reduces the RMS noise level while not reducing the signal level, and that increases the ratio of maximum signal to noise floor, which is the definition of dynamic range.
In other words, at any given signal level, S/N is better, and so you can go to a lower signal level (darker parts of the scene) before the per-pixel S/N level gets down to the same threshold of acceptable S/N, such as the 10:1 suggested by Kodak as the minimum acceptable.
To Ray: I should have said a minimum of 1/2 stop, and perhaps more but less than a full stop.
A half stop is what one would get if photon shot noise (and perhaps dark current noise) are the main noise sources, because these sources follow a square root law: noise increases in proportion to the square root of signal (photon or electron count).
Some other noise sources might not increase as fast with pixel and sensor size, though all data I have seen on the read noise of sensors show some noticeable increase in electrons of noise with photosite area. Thus total noise increases, but at most by the square root trend, and so S/N ratio and DR increase by between a factor between 1 and sqrt(2) for a doubling of sensor area, which is an improvement of between a half and one stop at equal exposure level.
The better sensor technology gets, the more that noise is dominated by "square root law" sources like photon shot noise, and so the closer one gets to the 1/2 stop I mentioned.
To Bill Janes: some CCD read noise comes from the photosites themselves (dark current noise) and this part combines by the square root law. So I would expect 2x2 binning to improve DR and S/N by a factor of between 2 and 4. One indication of this is the Kodak CCD's of similar era show a clear trend of increasing read noise with increasing photosite area, fitting a square root of area trend fairly well so that S/N ratio and DR grows roughly as the square root of photosite area.
-
BJL - thanks - that's helpful.
Mark
-
To Bill Janes: some CCD read noise comes from the photosites themselves (dark current noise) and this part combines by the square root law. So I would expect 2x2 binning to improve DR and S/N by a factor of between 2 and 4. One indication of this is the Kodak CCD's of similar era show a clear trend of increasing read noise with increasing photosite area, fitting a square root of area trend fairly well so that S/N ratio and DR grows roughly as the square root of photosite area.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208399\")
BJL,
Dark current noise (thermal noise) becomes significant only at exposures of seconds and does not contribute significantly to most normal photographic situations. Perhaps you did not read the Photometrics link about read noise. In the real world, 2by2 binning may yield an SNR improvement of less than 4:1. Dark current noise is not helped by binning.
Read noise expressed in electrons does not correlate well with sensor pixel size (see [a href=\"http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/]Roger Clark[/url]). However, when the electron count is converted to a data number (raw pixel value), the larger sensor has an advantage because of the increased gain (electrons per DN).
Photometrics (http://www.photomet.com/pm_solutions/library_encyclopedia/index.php)
"The primary benefit of binning is higher SNR due to reduced read noise contributions. CCD read noise is added during each readout event and in normal operation, read noise will be added to each pixel. However, in binning mode, read noise is added to each superpixel, which has the combined signal from multiple pixels. In the ideal case, this produces SNR improvement equal to the binning factors (4x in the above example). The figure below shows the effect of 2x2 binning for a four-pixel region. This example assumes that 10 photoelectrons have been collected in each pixel and the read noise is 10 electrons. If this region is read out in normal mode the SNR will be 1:1 and the signal will be lost in the noise. However, with 2x2 binning, the SNR becomes 4:1, which is sufficient to observe this weak signal."
Bill
-
I don't see why one needs to go to all this trouble. Just take a single exposure and see what happens to the noise when you reduce the resolution by binning 2x2 bunches of pixels. What's being considered is local on the sensor. If you want the reduction in noise, you have to give up some resolution; if you then make the sensor bigger, you get the same resolution relative to frame height with the binned pixels.
The business about pixel binning only refers to read noise; it has no effect on photon noise, which is dominant in midtones and highlights for low to moderate ISO. I think that suggestion is a bit of a tangent to the topic under discussion.
[a href=\"index.php?act=findpost&pid=208398\"][{POST_SNAPBACK}][/a]
Emil,
The experimental procedure I outlined above was not for the purpose of checking noise reduction effects resulting only from downsampling. I've seen clear examples of this effect comparing equal FoV images from the P30 and 5D. Same size crops (but at different magnification) look about equally noisy, but the P30 image displays more detail because it's comprised of more pixels.
After downsampling the P30 crop of the shadows to the same file size as the 5D crop of the same shadows, the resolution in both images is more or less equalised, but the P30 crop suddenly appears a lot cleaner, viewed side by side on the monitor.
Perhaps first we should define what constitutes a stop of dynamic range before I do this experiment.
My understanding is as follows. I compare two images of equal FoV and file size that have been correctly exposed according to the same ETTR standards. Call them images A and B.
If image A shows more noise in the shadows than image B, then image A clearly has less DR than image B.
If I have to overexpose image A by (for example) one whole stop in order that the shadows in image A look as clean as the shadows in image B (a process which also result in image A having blown highlights), then it is true to say that the camera that produced image B has one stop more dynamic range than the camera that produced image A. Is this correct?
-
BJL,
Dark current noise (thermal noise) becomes significant only at exposures of seconds and does not contribute significantly to most normal photographic situations. Perhaps you did not read the Photometrics link about read noise.
Actually, dark current is significant at small intervals of time. In standard models, output referred quantities (such as shot noise) may be referred back to their input referred values to be compared with input quantities (such as dark current). For e.g., under certain assumptions it can be shown that the output referred noise is reduced by square of integration time when referred to input.
Dark current calibration is a very significant part of any sensor device and may not be ignored for even smaller intervals of time than in seconds.
Furthermore, DR and SNR are not the same things -- sensors typically quote 2 different numbers for these two. They do correlate in the sense that if we consider SNR to be a good measure of image quality, then high DR, can be equally regarded as a good measure of image quality.
-
Emil,
The experimental procedure I outlined above was not for the purpose of checking noise reduction effects resulting only from downsampling. I've seen clear examples of this effect comparing equal FoV images from the P30 and 5D. Same size crops (but at different magnification) look about equally noisy, but the P30 image displays more detail because it's comprised of more pixels.
After downsampling the P30 crop of the shadows to the same file size as the 5D crop of the same shadows, the resolution in both images is more or less equalised, but the P30 crop suddenly appears a lot cleaner, viewed side by side on the monitor.
Perhaps first we should define what constitutes a stop of dynamic range before I do this experiment.
My understanding is as follows. I compare two images of equal FoV and file size that have been correctly exposed according to the same ETTR standards. Call them images A and B.
If image A shows more noise in the shadows than image B, then image A clearly has less DR than image B.
If I have to overexpose image A by (for example) one whole stop in order that the shadows in image A look as clean as the shadows in image B (a process which also result in image A having blown highlights), then it is true to say that the camera that produced image B has one stop more dynamic range than the camera that produced image A. Is this correct?
[a href=\"index.php?act=findpost&pid=208513\"][{POST_SNAPBACK}][/a]
OK, you've got me thoroughly confused as to what you want to do. Yes it's true that two images with equivalent absolute exposure have different noise in shadows (when converted with the same tone curve), then the one with less noise at a given tonal value in deep shadows will be exhibiting more DR. If you overexpose image C and convert it with the same tone curve as A and B, the noise at the same tonal value as before will not have changed. What you will have done is to move the parts of the image that used to be at that tonal value up to a higher tonal value, where there is less noise. In ETTR, one then typically changes the tone curve to restore that part of the image to its original tonal "intent". That part of the image then has less noise than it would have had with no ETTR and the original tone curve. But for the purposes of testing, one should always use the same tone curve (preferably linear) to compare apples to apples.
If you're not going to resample one or the other of your 40D images, you're not testing any property of the sensor. The sensor doesn't care what lens you put in front of it, at whatever f-stop, it always responds the same way to photons in a given absolute exposure.
-
Actually, dark current is significant at small intervals of time, and additionally, in standard models, output referred quantities (such as shot noise) should be referred back to their input referred values to be compared with input quantities (such as dark current). For e.g., under certain assumptions it can be shown that the output referred noise is reduced by square of integration time when referred to input.
Dark current calibration is a very significant part of any sensor device and may not be ignored for even smaller intervals of time than in seconds.
Furthermore, DR and SNR are not the same things -- sensors typically quote 2 different numbers for these two. They do correlate in the sense that if we consider SNR to be a good measure of image quality, then high DR, can be equally regarded as a good measure of image quality.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208516\")
You seem to be disagreeing with the standard noise model as for instance laid out in
[a href=\"http://learn.hamamatsu.com/articles/ccdsnr.html]http://learn.hamamatsu.com/articles/ccdsnr.html[/url]
or perhaps I'm not understanding what you mean by output/input referred values (do you just mean noise in ADU vs noise in electrons?). I would have thought, regardless how you normalize it, dark current noise variance is linear in integration time, not decreasing with it. And certainly any black frame noise I've ever measured for exposure times of say a tenth or less have had negligible contributions from dark current (ie they are reasonably independent of exposure time).
I do agree that DR and SNR are two different things. I think people are too hung up on DR; it's the level of S/N over the range that is more important, and that's where MFDB's have the edge -- gathering more photons as a percentage of frame area means higher S/N throughout the range, even if the range itself is the same as FF DSLR's.
-
But for the purposes of testing, one should always use the same tone curve (preferably linear) to compare apples to apples.
Emil,
This I shall do, as well as zero everything, including sharpening.
If you're not going to resample one or the other of your 40D images, you're not testing any property of the sensor. The sensor doesn't care what lens you put in front of it, at whatever f-stop, it always responds the same way to photons in a given absolute exposure.
I am going to resample one or the other. Comparing equal size images is the only meaningful comparison for the photographer. I shall compare both procedures; downsampling the larger file, and upsampling the smaller file.
-
You seem to be disagreeing with the standard noise model as for instance laid out in
http://learn.hamamatsu.com/articles/ccdsnr.html (http://learn.hamamatsu.com/articles/ccdsnr.html)
or perhaps I'm not understanding what you mean by output/input referred values (do you just mean noise in ADU vs noise in electrons?). I would have thought, regardless how you normalize it, dark current noise variance is linear in integration time, not decreasing with it.
Mapping of output parameters in an electrical/electronics circuit/network is frequently done to relate a parameter as observed on the output of an electronics circuit/model as the equivalent quantity as what would have been observed inside. (Not to be confused with ADU -- do you mean ADC by ADU?).
Using standard symbols (and some standard assumptions), the noise power may be represented as q(i_ph + i_d) * t_int + sigma_r^2, where t_int is the integration time and i_d is the dark current. Suppose f is the functional that represents the accumulated charge obtained from the current i over integration time t_int. Then, if I have to model the noise as input referred; first realizing that under the assumption that input referred noise is small compared to signal and may be estimated by the first order approximation
f(i + N_i) ~= f(i) + N_i * f'(i),
where, f' is the derivative, the average power of the equivalent input referred noise may be seen as:
(q(i_ph + i_d) * t_int + sigma_r^2) / f'^2 = (q(i_ph + i_d) * t_int + sigma_r^2) / t_int^2,
where the last parameter t_int^2 is what I mentioned as the square in the input referred noise.
And certainly any black frame noise I've ever measured for exposure times of say a tenth or less have had negligible contributions from dark current (ie they are reasonably independent of exposure time).
What we refer to as "raw" signal is normally not raw. Typically extensive calibration is applied to properly offset the signal inside the hardware. I don't know if the equipment that you are using is applying these calibrations (it should), and assuming that it does, then perhaps dark current degradation may have already been corrected.
We have measured the dark current degradations right off the hardware without calibration, and some correction is always needed to make the image quality better.
-
Actually, dark current is significant at small intervals of time, and additionally, in standard models, output referred quantities (such as shot noise) should be referred back to their input referred values to be compared with input quantities (such as dark current). For e.g., under certain assumptions it can be shown that the output referred noise is reduced by square of integration time when referred to input.
Dark current calibration is a very significant part of any sensor device and may not be ignored for even smaller intervals of time than in seconds.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208516\")
Rather than discussing the matter in the abstract and obscuring the issues with needless jargon and mathematics (as in your reply to Emil), you might want to refer to actual examples. For example, the [a href=\"http://www.photomet.com/pm_solutions/library_encyclopedia/library_enc_dark.php]Photometrics[/url] web site has a good section in dark current and noise. In their example, they conclude:
"Thus, the dark current noise generated in a 4-second exposure has virtually no effect on the total camera system noise. Similarly, for a 30-second exposure we find that the total system noise equals 14.1 electrons. Again, even at a 30-second exposure, dark current noise barely contributes to the total camera system noise."
In another example, Roger Clark (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html) examines the noise characteristics of the Canon 1 D Mark II. The sensor has a read noise of 3.8 electrons. He found that average dark currents per pixel were 0.013 to 0.02 electrons/second, but that some pixels had dark currents as high as about 0.25 electrons/second. For an exposure of one second, the dark noise is negligible as compared to the read noise.
Furthermore, DR and SNR are not the same things -- sensors typically quote 2 different numbers for these two. They do correlate in the sense that if we consider SNR to be a good measure of image quality, then high DR, can be equally regarded as a good measure of image quality.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208516\")
That is true, but I don't know why you bring up the matter in your reply to me. I did not mention DR. DR is defined as the full well in electrons divided by the read noise, also expressed in electrons. Shot noise, the most important source of noise over most of the range of exposures, does not enter into the equation.
SNR includes all sources of noise and may be calculated for various levels of exposure, as described by [a href=\"http://www.imatest.com/docs/noise.html]Norman Koren[/url] on his Imatest web site.
Bill
-
Mapping of output parameters in an electrical/electronics circuit/network is frequently done to relate a parameter as observed on the output of an electronics circuit/model as the equivalent quantity as what would have been observed inside. (Not to be confused with ADU -- do you mean ADC by ADU?).
Using standard symbols (and some standard assumptions), the noise power may be represented as q(i_ph + i_d) * t_int + sigma_r^2, where t_int is the integration time and i_d is the dark current. Suppose f is the functional that represents the accumulated charge obtained from the current i over integration time t_int. Then, if I have to model the noise as input referred; first realizing that under the assumption that input referred noise is small compared to signal and may be estimated by the first order approximation
f(i + N_i) ~= f(i) + N_i * f'(i),
where, f' is the derivative, the average power of the equivalent input referred noise may be seen as:
(q(i_ph + i_d) * t_int + sigma_r^2) / f'^2 = (q(i_ph + i_d) * t_int + sigma_r^2) / t_int^2,
where the last parameter t_int^2 is what I mentioned as the square in the input referred noise.
What we refer to as "raw" signal is normally not raw. Typically extensive calibration is applied to properly offset the signal inside the hardware. I don't know if the equipment that you are using is applying these calibrations (it should), and assuming that it does, then perhaps dark current degradation may have already been corrected.
We have measured the dark current degradations right off the hardware without calibration, and some correction is always needed to make the image quality better.
[a href=\"index.php?act=findpost&pid=208579\"][{POST_SNAPBACK}][/a]
OK, now I understand where you're coming from (and BTW, ADU is a common abbreviation for quantization step, ie raw level).
What is the utility of using input-referred quantities? I think the thrust of bjanes' reply and my previous post is that dark current is a negligible component of the output noise in all but very long time exposures (several seconds or more). I'm also puzzled why one would want to extrapolate this back to a hypothetical input noise which, as far as I can tell, is not any actual noise of any actual component of the capture process.
-
Okay! Enough theory! Time for real world experiments!
Today was a dull day in Brisbane. I wondered what might be a good scene for checking the DR variation between the cropped format 40D and a full frame sensor consisting of 40D pixels.
I decided to remove a 24"x36" print from the wall and place it in the centre of the window, through which is a view of the Brisbane river and grey skies. The dynamic range is great, but not excessive.
The print is a 15mm shot of Ta Prohm in the Angkor Wat area (taken with the 5D and Sigma lens). It actually looks a lot more vibrant than the crops in this experiment would suggest.
I used the Canon 24-105mm zoom, initially at 24mm amd F8. To simulate a doubling of sensor area, I would need to multuply that focal length by 1.4, ie. 33.6mm. To simulate a full frame 35mm sensor, I would need to multiply the focal length by 1.6, ie. 38.4mm. I can't get such precision with zoom adjustments and/or ACR read-outs, so I settled for a compromise of 35mm. That's the first approximation.
To adjust F stop for equal DoF, I would need F11.2 with double the sensor area and F12.8 with FF 35mm. I used F13, so that's the second approximation.
The conclusion first. Doubling sensor area containing same size and same quality pixels seems to increase DR by close to one F stop but probably not quite one F stop in view of the above stated approximations.
Immediately below is the over all scene with a bit of adjustment, so it doesn't appear totally flat, followed by 200% and 100% comparison crops of totally flat conversions with zero settings in ACR, no sharpening and linear tone curve.
[attachment=7470:attachment]
First, F8 at 1/30th (24mm) compared with F13 at 1/10th (35mm). Perhaps that's the third approximation. I can't get any more precise than that.
[attachment=7471:attachment] [attachment=7472:attachment]
For the benefit of Mark, who seems a bit skeptical about this relationship between sensor size and DR, the Full Frame 40D clearly has less noise. There's no doubt about it. The only doubt is, by just how much?
The next crops compare the 24mm shot overexposed by one stop, with the correctly exposed 35mm shot. To my eyes, they are about equal. But this is extreme pixel peeping. How such differences relate to real world prints is another experiment.
[attachment=7473:attachment] [attachment=7474:attachment]
In the final crop comparisons, the crop of the cropped format shot has been upsmapled to the same size as the full frame shot.
[attachment=7475:attachment]
-
you might want to refer to actual examples.
And,
I think the thrust of bjanes' reply and my previous post is that dark current is a negligible component of the output noise in all but very long time exposures (several seconds or more).
It all depends upon what sensor and what integration time you are using. As I mentioned before that in some of the sensors we have tried, we have not found dark current to be insignificant for exposures as small as ~30 milli seconds, when signal is captured right off the sensor without any calibration. Where as typically people will acquire "raw" after all the calibrations, adjustments, etc. have been made.
It may be possible that for really small integration times one may not observe a lot of dark current, but as I mentioned above, for ~30 milli seconds we have not found that with at least those sensors we tested.
What is the utility of using input-referred quantities?
I'm also puzzled why one would want to extrapolate this back to a hypothetical input noise which, as far as I can tell, is not any actual noise of any actual component of the capture process.
Input level quantities may be used to derive equations for DR, SNR, etc. and give a good comparison of noise with dark current, etc.
BTW, the whole discussion on this input thingy started from a side remark of mine and originally I had no intention of going into the details of it as it is indeed irritating for some users.
-
Okay! Enough theory! Time for real world experiments!
Today was a dull day in Brisbane. I wondered what might be a good scene for checking the DR variation between the cropped format 40D and a full frame sensor consisting of 40D pixels.
I decided to remove a 24"x36" print from the wall and place it in the centre of the window, through which is a view of the Brisbane river and grey skies. The dynamic range is great, but not excessive.
The print is a 15mm shot of Ta Prohm in the Angkor Wat area (taken with the 5D and Sigma lens). It actually looks a lot more vibrant than the crops in this experiment would suggest.
I used the Canon 24-105mm zoom, initially at 24mm amd F8. To simulate a doubling of sensor area, I would need to multuply that focal length by 1.4, ie. 33.6mm. To simulate a full frame 35mm sensor, I would need to multiply the focal length by 1.6, ie. 38.4mm. I can't get such precision with zoom adjustments and/or ACR read-outs, so I settled for a compromise of 35mm. That's the first approximation.
To adjust F stop for equal DoF, I would need F11.2 with double the sensor area and F12.8 with FF 35mm. I used F13, so that's the second approximation.
The conclusion first. Doubling sensor area containing same size and same quality pixels seems to increase DR by close to one F stop but probably not quite one F stop in view of the above stated approximations.
Immediately below is the over all scene with a bit of adjustment, so it doesn't appear totally flat, followed by 200% and 100% comparison crops of totally flat conversions with zero settings in ACR, no sharpening and linear tone curve.
[attachment=7470:attachment]
First, F8 at 1/30th (24mm) compared with F13 at 1/10th (35mm). Perhaps that's the third approximation. I can't get any more precise than that.
[attachment=7471:attachment] [attachment=7472:attachment]
For the benefit of Mark, who seems a bit skeptical about this relationship between sensor size and DR, the Full Frame 40D clearly has less noise. There's no doubt about it. The only doubt is, by just how much?
The next crops compare the 24mm shot overexposed by one stop, with the correctly exposed 35mm shot. To my eyes, they are about equal. But this is extreme pixel peeping. How such differences relate to real world prints is another experiment.
[attachment=7473:attachment] [attachment=7474:attachment]
In the final crop comparisons, the crop of the cropped format shot has been upsmapled to the same size as the full frame shot.
[attachment=7475:attachment]
[a href=\"index.php?act=findpost&pid=208670\"][{POST_SNAPBACK}][/a]
I think what's happening in the comparisons where one can see a difference is the effect of diffraction between f/8 (less) and f/11 (more).
-
Furthermore, DR and SNR are not the same things -- sensors typically quote 2 different numbers for these two.
[a href=\"index.php?act=findpost&pid=208516\"][{POST_SNAPBACK}][/a]
True, and there are also at least two different meanings of each phrase, some of which overlap.
Sensor spec. sheets do indeed often state both a dynamic range (in dB) and a signal to noise ratio (in a form like 1000:1), but these are different statements of the same information: the ratio between maximum output signal (well capacity) to dark noise level. The DR in dB is computed as 20*log10(SN), so that for example 1000:1 given 60dB. I call this S/N ratio the "global" one.
But most often the noise measure that we care about is the "local S/N ratio" at various parts of the image: the ratio of signal level at a particular part of the image to the noise level there. Then the relevant signal level is usually far less than maximum, and the noise is often more, as it includes photon shot noise as well as dark noise. So local S/N ratio is far lower, and a Kodak document suggests that about 40:1 is excellent and 10:1 is barely acceptable.
For comparison the "global" S/N ratio probably needs to be about 1000:1 (60dB) or more for decent dynamic range, and MF sensors offer about 4000:1 (72dB).
By the way, Dalsa's spec's for DR in stops are simply the log base 2 of the S:N ratio, so 12.5 stops means a S/N ratio of about 5600:1 or 75dB. This is a valid engineering spec. if understood correctly, but the bottom three or so stops of that range will have excessive visible noise levels.
-
Thanks BJL, you explained it very well. I could not have done that.
-
I think what's happening in the comparisons where one can see a difference is the effect of diffraction between f/8 (less) and f/11 (more).
[a href=\"index.php?act=findpost&pid=208747\"][{POST_SNAPBACK}][/a]
Mark,
It's true that the simulated 26mp full frame image (at 35mm) appears only marginally more detailed than the smaller format image at 24mm. This is probably due to a number of factors such as zero sharpening and a linear tone curve which tends to reduce accutance differences; diffraction effects at F13, and the degraded nature of image quality in dark shadows which are not ideal targets for displaying resolution. However, here I was not trying to demonstrate resolution differences but noise differences.
It's clear on my monitor (a Sony 19" CRT at 1600x1200) that the 24mm correctly exposed F8 shot has a lot more noise in the shadows than the 35mm F13 shot which has been exposed by the same standards within the accuracy limitations of the camera's exposure values, ie. F8 at 1/30th = F13 at 1/10th.
It's also clear that an F8 shot overexposed by one stop (ie. 1/15th instead of 1/30th) has about the same degree of shadow noise as the F13 shot correctly exposed.
Below are two sets of 3 smaller shadow crops showing, from left to right; (1) the correctly exposed F8 shot, (2) the F8 shot overexposed by one stop, (3) the correctly exposed F13 shot.
Of course I've had to lighten the shadows a lot using 'levels' in order for such differences to be apparent. The excessive noise in the first image of both sets hits me in the face at this high degree of magnification. Do you see a flaw in the methodology? It was an overcast day with even lighting. All shots were taken within a 5 minute time frame.
[attachment=7491:attachment] [attachment=7490:attachment]
-
Ray, yes I know you were demonstrating differences in noise - it's all part of the same package I had in mind in the sense that more diffraction means less apparent resolution and less apparent noise.
Your examples are consistent with this generality. At f/8 you are getting less diffraction, more resolution and more apparent noise than you are at f/13. By increasing the exposure of the f/8 shot, you are adding light which improves the S/N hence reducing apparent noise toward the lower level of the f/13 (versus the f/8) shot. Of course lightening shadows with Levels simply "shines a light" on the noise that is there. I don't see any methodological issues so far, except that I probably would have chosen a real scene with the appropriate ingredients rather than a print for generating the scene data. But what you are using seems to be working for the purpose at hand.
-
Ray, yes I know you were demonstrating differences in noise - it's all part of the same package I had in mind in the sense that more diffraction means less apparent resolution and less apparent noise.
Your examples are consistent with this generality. At f/8 you are getting less diffraction, more resolution and more apparent noise than you are at f/13. [a href=\"index.php?act=findpost&pid=208802\"][{POST_SNAPBACK}][/a]
Mark,
I don't see it that way. The F8 shot is a cropped format 10mp shot and actually has slightly less resolution than the F13 shot which is 26mp full frame (simulated). If I had used F8 with the 26mp full frame at 35mm, then instead of a barely perceptable increase in resolution, I would have got a significant increase in resolution in the plane of focus, which was the print, and therefore, according to your reasoning, the improvement in noise (and DR) would have been even greater than one stop.
By the way, I've just realised that my monitor is set at 1800x1440, not the 1600x1200 that it used to be. The crops are probably unnecessarily large.
Actually, I've just got myself confused, so I'll rephrase that point I was trying to make. At F13 the image is already marginally sharper than the F8 image because it's effectively a 26mp full frame image comprised of more pixels, so the lower noise of the F13 shot would seem to have little to do with resolution. The loss of resolution due to diffraction is largely compensated by the effectively larger sensor.
Since no sharpening or contrast enhancement has been applied to any of these crops, I don't see how increasing resolution by using F8 instead of F13 at 35mm FL would increase noise
-
Oops - several things happening at the same time. It gets confusing. I think the testing should isolate one variable at a time and see what happens.
-
Oops - several things happening at the same time. It gets confusing. I think the testing should isolate one variable at a time and see what happens.
[a href=\"index.php?act=findpost&pid=208809\"][{POST_SNAPBACK}][/a]
Okay! If the target is 2-dimensional, then DoF is not an issue. I'll repeat a similar experiment using F8 with all shots,
-
Ray, yes I know you were demonstrating differences in noise - it's all part of the same package I had in mind in the sense that more diffraction means [...] less apparent noise.
Your examples are consistent with this generality. [...]
[a href=\"index.php?act=findpost&pid=208802\"][{POST_SNAPBACK}][/a]
Diffraction has zero effect on noise. There are two classes of noise, one due to the sensor electronics, and the other is the photon noise that is intrinsic to the light hitting the sensor. The former couldn't care less whether there is diffraction or not, since it's a property of the sensor independent of whether it's receiving a light signal or not. The latter also couldn't care less whether there is diffraction; it's dependent only on how many photons are collected by a given photosite, and it doesn't matter whether or not those photons arrived at that photosite through diffraction, it just matters that they arrived at that photosite and not another one.
The ONLY thing that diffraction affects is the resolution.
-
Diffraction has zero effect on noise. There are two classes of noise, one due to the sensor electronics, and the other is the photon noise that is intrinsic to the light hitting the sensor. The former couldn't care less whether there is diffraction or not, since it's a property of the sensor independent of whether it's receiving a light signal or not. The latter also couldn't care less whether there is diffraction; it's dependent only on how many photons are collected by a given photosite, and it doesn't matter whether or not those photons arrived at that photosite through diffraction, it just matters that they arrived at that photosite and not another one.
The ONLY thing that diffraction affects is the resolution.
[a href=\"index.php?act=findpost&pid=208821\"][{POST_SNAPBACK}][/a]
Yes, but doesn't diffraction scatter light between photosites, and doesn't the reduction of resolution somehow also reduce the apparent definition of the noise eventhough it is not the primary cause? How else would you interpret the results Ray was showing, or do you not think the methodology is sufficiently fine-tuned or the information provided sufficient to come to any robust conclusions about cause and effect here?
-
Yes, but doesn't diffraction scatter light between photosites, and doesn't the reduction of resolution somehow also reduce the apparent definition of the noise eventhough it is not the primary cause? How else would you interpret the results Ray was showing, or do you not think the methodology is sufficiently fine-tuned or the information provided sufficient to come to any robust conclusions about cause and effect here?
[a href=\"index.php?act=findpost&pid=208824\"][{POST_SNAPBACK}][/a]
Well, here's a little thought experiment -- do you think there is no noise or less noise in OOF areas of images, or that the noise is somehow less 'grainy'? It is not, as you will see by looking through your images.
Perhaps you have in mind that, if the lens is defocussed, or loses resolution due to diffraction, that the noise will somehow also get defocussed and averaged out. This is not the case, because the noise is a property of the light that gets to a particular photosite, as well as the electronics of that photosite; it doesn't care what the other photosites are receiving, whether the image they are making is focussed or not, whether the optics are diffraction limited. A signal of N photons has sqrt(N) amount of photon noise, no matter where those photons came from, whether or how they were reflected/refracted/diffracted. It's true that diffraction and also defocussing will rearrange the pattern of photons hitting the sensor, but then the new distribution of photons will have a new pattern of photon noise that still only cares about how many photons are collected in each photosite. And of course the read noise is happening independent of what the photons are doing, and contributing the same no matter what.
-
Well, here's a little thought experiment -- do you think there is no noise or less noise in OOF areas of images, or that the noise is somehow less 'grainy'? It is not, as you will see by looking through your images.
Perhaps you have in mind that, if the lens is defocussed, or loses resolution due to diffraction, that the noise will somehow also get defocussed and averaged out. This is not the case, because the noise is a property of the light that gets to a particular photosite, as well as the electronics of that photosite; it doesn't care what the other photosites are receiving, whether the image they are making is focussed or not, whether the optics are diffraction limited. A signal of N photons has sqrt(N) amount of photon noise, no matter where those photons came from, whether or how they were reflected/refracted/diffracted. It's true that diffraction and also defocussing will rearrange the pattern of photons hitting the sensor, but then the new distribution of photons will have a new pattern of photon noise that still only cares about how many photons are collected in each photosite. And of course the read noise is happening independent of what the photons are doing, and contributing the same no matter what.
[a href=\"index.php?act=findpost&pid=208833\"][{POST_SNAPBACK}][/a]
Emil,
This is more or less what I would have thought. I just repeated a similar experiment this afternoon, to prove Mark wrong, but was surprised at the results.
Using F8 for all shots, the approximately 1 stop advantage I saw at F13 has almost disappeared. It's most puzzling.
What I see at F8 is a more obvious resolution advantage for the simulated larger sensor, which is evident across the board, in both highlights and shadows, but the obviously lower shadow noise I saw previously in the F13 shot, has virtually disappeared.
Consider the following 200% crops at F8 and 1/4 sec exposure.
[attachment=7496:attachment]
One might wonder why I'm shooting a back-lit print. This print has excellent shadow detail because I used fill flash when I took the shot in 1995. There's lots of detail in them thar shadows.
Below is a comparison with the original file from which the print was produced. Would you say that something has been lost in the reproduction ?
[attachment=7498:attachment] [attachment=7497:attachment]
-
Diffraction and noise are completely independent at the physical level (in terms of underlying causes and contribution to the recorded image) but not necessarily at the visual results level. This is because diffraction affects the overall frequency content of the recorded signal -- i.e., it makes the image blurrier. Noise is typically more visible, and hence objectionable, in blurry images.
A thought experiment: consider taking an image of a detailed colorful rug. Lots of fine detail and texture. You take a perfectly-focused image at ISO 1600 with, say, a Rebel XSi. There is noise in the image but it is largely masked by the actual image detail. The noise is generally not objectionable and perhaps even invisible in a print.
Retake the same image with the same capture parameters, except this time defocus the lens. So the capture noise is the same, but now the image is completely blurry. The noise suddenly becomes extremely visible and easy to pick out.
Diffraction is related to this example in that it creates a blur and hence can make noise more visible, just not to the same degree as grossly misfocusing.
Cheers,
Eric
-
Diffraction is related to this example in that it creates a blur and hence can make noise more visible, just not to the same degree as grossly misfocusing.
Cheers,
Eric
[a href=\"index.php?act=findpost&pid=208869\"][{POST_SNAPBACK}][/a]
Eric,
You seem to have got this the wrong way round. I produced an image at F13 which had clearly less noise than another image at F8.
Mark claimed that the lower noise might have been due to the blurring effect of diffraction at F13.
I retook similar shots (albeit in slightly different circumstances) and found that F8 with all shots had an effect of equalising noise, despite different pixel counts.
More test shots are required, I think.
-
I'm not convinced more tests are needed. I'm not arguing with the theory so ably explained by Emil and Bill in this thread and on websites such as Imatest/Norman Koren, Cambridgeincolour, etc. Also please note that when I speak of noise I usually qualify it with the word "apparent" specifically because I'm talking about what we see rather than what may be embedded in the image but we don't see for reasons Eric explains ( which BTW jusstifies use of selective noise reduction in image processing).
At the end of my previous post I asked Emil if he could explain what's happening in Ray's tests, or if for some reason that is not possible. Perhaps that question got overlooked by accident. Then Eric came in with a correct set of propositions which point to the "apparent" opposite of what's happening in Ray's images, as Ray pointed out.
Hence, I'm thinking that what's needed is a better explanation of an obvious disconnect between what we're all seeing in those images and what the theory says we should be seeing. There may be a missing variable or two which would clarify why Ray's results are what they are.
-
I'm not convinced more tests are needed. I'm not arguing with the theory so ably explained by Emil and Bill in this thread and on websites such as Imatest/Norman Koren, Cambridgeincolour, etc. Also please note that when I speak of noise I usually qualify it with the word "apparent" specifically because I'm talking about what we see rather than what may be embedded in the image but we don't see for reasons Eric explains ( which BTW jusstifies use of selective noise reduction in image processing).
At the end of my previous post I asked Emil if he could explain what's happening in Ray's tests, or if for some reason that is not possible. Perhaps that question got overlooked by accident. Then Eric came in with a correct set of propositions which point to the "apparent" opposite of what's happening in Ray's images, as Ray pointed out.
Hence, I'm thinking that what's needed is a better explanation of an obvious disconnect between what we're all seeing in those images and what the theory says we should be seeing. There may be a missing variable or two which would clarify why Ray's results are what they are.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=208898\")
As [a href=\"http://www.imatest.com/docs/noise.html#spectrum]Norman Koren[/url] points out, the visual appearance of noise is rather difficult to quantify and depends on more than the standard deviation. A noise spectrum plot may help to some extent. Your qualification of "apparent" is quite appropriate.
As Emil has explained, if you take two sensors with the same total sensing area and double the pixel count of one, keeping all other characteristics unchanged, the signal:noise of the higher resolution sensor will be lower. If you examine the images at 100% on screen, the noise in the higher resolution sensor will be higher. However, if you print the two images at the same print size, there may not be any difference in the perceived noise. Intuitively, one may explain this by the fact that the noise in the higher resolution sensor is finer grained (the frequency distribution is shifted to the right).
However, as the pixel size decreased there may be problems with fill factor and with the microlenses, as explained here. (http://white.stanford.edu/~brian/papers/ise/CMOSRoadmap-2005-SPIE.pdf) If you carry a thought experiment on decreasing pixel size to the point of Reductio ad absurdum, S:N falls to zero when the pixel count approaches infinity.
Imatest does supply a noise spectrum. Noiseware does also, as you so ably demonstrated in your essay. However, interpreting the plot of the spectrum is problematic.
Due to the subjective quality of noise, I think some testing such as Ray is undertaking is helpful, but as his efforts show, it is difficult to control all the variables. Results that defy the laws of physics are doubtful. It would be interesting to apply Dr. Wandell's image simulator to some real world cameras.
Bill
-
As Emil has explained, if you take two sensors with the same total sensing area and double the pixel count of one, keeping all other characteristics unchanged, the signal:noise of the higher resolution sensor will be lower. If you examine the images at 100% on screen, the noise in the higher resolution sensor will be higher. However, if you print the two images at the same print size, there may not be any difference in the perceived noise. Intuitively, one may explain this by the fact that the noise in the higher resolution sensor is finer grained (the frequency distribution is shifted to the right).
''''''''''''''''''''''''''
Due to the subjective quality of noise, I think some testing such as Ray is undertaking is helpful, but as his efforts show, it is difficult to control all the variables. Results that defy the laws of physics are doubtful. It would be interesting to apply Dr. Wandell's image simulator to some real world cameras.
Bill
[a href=\"index.php?act=findpost&pid=208920\"][{POST_SNAPBACK}][/a]
I think you are tending toward the nub of the matter. I have seen from my 1Ds3 what you are mentioning - the "character" of the noise is different compared with the original 1Ds. It is "finer-grained", lighter and tighter, so less disturbing to look at.
Looking at Ray's last set of results in post 134 (wow, this is getting up there - maybe we'll make a L-L "Book of Records" for longest thread!) both shot at f/8, the 40 result has better overall definition than the 24 result, and as he says there appears to be very little difference of visible noise between these two images. If I remember correctly, his "40" designation is the simulated higher pixel count sensor; this would seem to indicate that between the two cameras being tested, perhaps within the range of pixel sizes on test, the lower S/N which physics tells us should be doming from the 40 is having less of an effect degrading quality than is the higher resolution increasing it.
-
So much for pixel peeping!
Part of the confusion here is due to the fact I made a mistake with regard to my initial comparison. I compared F13 at 1/10th sec exposure with F8 at 1/30th sec exposure. By doing so, I've given the simulated full frame sensor approximately a 1/3rd stop advantage. I should have been comparing F13 at 1/13th sec exposure with F8 at 1/30th. There's a 1 & 1/3rd stop difference between F8 and F13.
Below are the ACR windows showing the histograms. As you can see, the histogram for the F13 shot at 1/13th indicates underexposure whereas the F8 shot looks more correctly exposed. That's as it should be because the shot at 35mm has excluded areas of sky which would have been included if the sensor had been full frame.
[attachment=7505:attachment]
I won't bore you with additional examples of degraded images. Suffice it to say, it's mostly a storm in a tea cup. Comparing F8 at 24mm and 1/30th, with F13 at 35mm and 1/13th, does change the result and diminish the perceived differences to the point where it's all largely irrelvant, in my opinion.
However, a doubling (or more) of pixel count does produce a worthwhile increase in resolution and detail at the optimal aperture of the lens, whether F8 or F5.6.
At F13, F16 and F22, I think we're getting into irrelevant pixel peeping areas of little significance, resolution-wise, for the practical photographer.
Having compared a few shots at the same focal length but different apertures, such as F13 at 1/13th with F8 at 1/30th, both at 35mm, there is a curious phenomena whereby the F8 shot really does appear very marginally noisier in the really deep shadows where greens seem to shift to magenta and there's a hint of greater banding.
But again, it's curious at the pixel peeping level and no doubt of interest to Physicists, but not of much practical concern to photographers.
-
Removed.
-
However, what I'm looking for is more detail; some smidgen or speck that is clearly defined in the 40D shot but not in the 5D shot.
On the Wallaman falls shots
Look at the coordinates 2 and 5/8" across, 3 and 3/4" down. There is a group of light green lines, possibly grasses, that sweep down to the left on the 40D. On the 5D it looks like 1 line with the rest smeared.
-
On the Wallaman falls shots
Look at the coordinates 2 and 5/8" across, 3 and 3/4" down. There is a group of light green lines, possibly grasses, that sweep down to the left on the 40D. On the 5D it looks like 1 line with the rest smeared.
[a href=\"index.php?act=findpost&pid=218812\"][{POST_SNAPBACK}][/a]
I was at a disadvantage when assessing these comparison shots initially, which I'd created on a laptop in the motel room.
On my monitor back home, there seems no doubt that the 40D crops at both F16 and F22 look better. One might have difficulty identifying a specific iota of detail that's visible on the 40D shot but not visible on the 5D shot, but the over all impression is one of finer grain and greater realism in the 40D crops.
If I were travelling with a 24mp full frame and the 5D as back-up, I can't see myself choosing the 5D for a particular shot on the grounds that at F16 or F22, using the 24mp camera would provide no additional image quality benefit.
-
If I were travelling with a 24mp full frame and the 5D as back-up, I can't see myself choosing the 5D for a particular shot on the grounds that at F16 or F22, using the 24mp camera would provide no additional image quality benefit.
[a href=\"index.php?act=findpost&pid=218884\"][{POST_SNAPBACK}][/a]
Ray:
What do you mean by this? Are you saying that regardless of the difference of resolution between 12 and 24 MP, the benefit of 24 is lost because of the diffraction created by the sub-optimal aperture? If so, why not use the lens at its optimal aperture on the 24MP camera. Seems to me you would get the best of both worlds, unless you need the DoF of the smaller aperture. Then, I would still use the 24 MP camera, if I had both, because the diffraction loss due to the aperture should be about the same on both cameras, no?
-
Ray:
What do you mean by this? Are you saying that regardless of the difference of resolution between 12 and 24 MP, the benefit of 24 is lost because of the diffraction created by the sub-optimal aperture? If so, why not use the lens at its optimal aperture on the 24MP camera. Seems to me you would get the best of both worlds, unless you need the DoF of the smaller aperture. Then, I would still use the 24 MP camera, if I had both, because the diffraction loss due to the aperture should be about the same on both cameras, no?
[a href=\"index.php?act=findpost&pid=218898\"][{POST_SNAPBACK}][/a]
Mark,
I'm saying that there's some quality beyond detail resolution that I would describe as an impression of finer grain and greater realism, with the 24mp camera at F16 & F22.
When making a 23"x35" print, I would rather start from a 24mp image than a 12mp image, even at F22.
-
OK, I agree. I've also noticed with the 1DsMk3 compared to the original 1Ds the noise, when you see it, has a finer, less intrusive grain pattern. I imagine this is the result of the higher resolution and newer vintage of DIGIC processing methods.
-
OK, I agree. I've also noticed with the 1DsMk3 compared to the original 1Ds the noise, when you see it, has a finer, less intrusive grain pattern. I imagine this is the result of the higher resolution and newer vintage of DIGIC processing methods.
[a href=\"index.php?act=findpost&pid=218934\"][{POST_SNAPBACK}][/a]
Were you showing us jpegs or raw conversions? DIGIC IV is the jpeg engine, and has little to do with what the raw data will show. What you may be seeing as finer "grain" is the pushing of Bayerinterpolation artifacts to finer scales.
-
EJ, I wasn't showing anything - maybe that comment is for someone else.
For whatever reason the grain of the noise is finer with the newer processors - I consider it an improvement of image appearance, don't you?