Luminous Landscape Forum

Site & Board Matters => About This Site => Topic started by: 32BT on August 25, 2006, 09:48:22 am

Title: Sensor & Sensibility II
Post by: 32BT on August 25, 2006, 09:48:22 am
Like some of the others here, I had a hard time following the Sensor & Sensibility thread. And I don't consider myself technically challenged in any way, nor do I shy away from purely theoretical discussions.

However, I would like to redo a similar discussion, but in a more pragmatic approach. That is, instead of purely speculating about the advantages of larger pixel-bins, or whatever the preferred term is these days (minutes?), how about discussing the real world examples that are available to us, void of marketing speak...

DPReview has an interesting image test in the Canon 5D review at the bottom of this page (http://www.dpreview.com/reviews/canoneos5d/page22.asp) and RAW example on the next page (http://www.dpreview.com/reviews/canoneos5d/page23.asp).

So the 5D obviously has large pixel-bins, if not the largest in the industry, and the 20D would be the perfect example of a camera that represents what most people here consider the best compromise between sensor-size, pixel-count, lens-performance, and printing resolution needs for most common situations.
(note that this is a premise you're obviously free to dispute if appropriate).

If I understand the various discussions properly, the question now becomes this:
Are the purported advantages of larger pixel-bins evident in the real-world performance and do they justify the expenses required?

Based on nothing but gut feeling I think that the 5D requires less processing than the 20D to yield the same performance figures which translates to more natural looking images at low iso, and a significant difference in processing latitude for high-iso images... ?
Title: Sensor & Sensibility II
Post by: Ray on August 25, 2006, 01:30:28 pm
I don't really have any doubt, Oscar, that the larger sensor with the larger pixel pitch will have certain advantages, either in terms of wider dynamic range, less noise, a shallower DoF capability, or higher resolution if it has more pixels as well as larger pixels, as the 5D has compared with the 20D.

Since I own both a 20D and 5D, I could be testing some of these issues myself. Perhaps I'll get around to it, one day.

However, I feel there's one important issue that is often ignored when comparing different format cameras, and that's the DoF equivalence at a required shutter speed. It's widely recognised that small format cameras, particularly P&S digicams, are of little use for creative photography employing shallow DoF. But the reverse is equally true. They are of great use for getting good DoF at wide apertures and fast shutter speeds.

Most comparisons between a 20D and 5D, I can almost guarantee, will be at the same aperture, same ISO and very similar shutter speeds, almost as though DoF is never a consideration. Since both DoF and shutter speed are usually amongst the most important considerations when composing a shot, it seems to me appropriate to make comparisons between the 5D at f11 and ISO 1600 and the 20D at f6.3 and ISO 600, for example (and of course same FoV). DoF and shutter speed will be the same, but the 20D might actually have less noise on the same size prints despite having smaller and fewer pixels. I'd be surprised if the 20D displayed more noise in these circumstances, but it's something I'll have to check for myself I guess.
Title: Sensor & Sensibility II
Post by: John Sheehy on August 25, 2006, 02:14:02 pm
Quote
If I understand the various discussions properly, the question now becomes this:
Are the purported advantages of larger pixel-bins evident in the real-world performance and do they justify the expenses required?
[a href=\"index.php?act=findpost&pid=74439\"][{POST_SNAPBACK}][/a]

The big pixels that capture more electrons per pixel will have higher signal-to-noise in the highlights, at the pixel level.  The blackframe noise, however, limits the dynamic range.  "Maximum signal divided by blacklevel noise" is the determiner of dynamic range.  "Maximum signal divided by the noise of *that* signal" is not DR, and the latter is what counting more photons does for you.

IOW, the bigger pixels mean a lot less noise in the highlights, and in the midtones, and even the brightest shadows (to a lesser extent), but do not help as much in the deeper shadows, and therefore with DR.
Title: Sensor & Sensibility II
Post by: 32BT on August 25, 2006, 02:51:06 pm
Quote
IOW, the bigger pixels mean a lot less noise in the highlights, and in the midtones, and even the brightest shadows (to a lesser extent), but do not help as much in the deeper shadows, and therefore with DR.

What do you mean by "a lot less"? And how does that translate to real-world images?
Page 21 of the same review seems to suggest that the real world noise difference for gray is marginal, certainly not a lot less??

But suppose it's true, then it's obvious that the reduced noise advantage can only be harvested through more bits in the A/D conversion. But that's not the case. Both camera's use 12bits AFAIK.

What does seem to be the case is a difference in processing (noise reduction amongst others) required to generate the images. This translates to more natural looking images in the 5D, and better image performance at high iso.

Just saying that "bigger pixels = higher dynamic range and/or less noise" while all of the examples or technical graphs indicate otherwise, to me means we have to redefine how we judge the net effect of the advantages IF they truly exist...

Note that I'm not trying to attack the theory, I'm just trying to find out what improvements people actually SEE in an image from a large pixel-bin sensor vs a smaller pixel-bin sensor. How do the theoretical advantages actually translate into real-world images?
Title: Sensor & Sensibility II
Post by: 32BT on August 25, 2006, 03:16:25 pm
Quote
However, I feel there's one important issue that is often ignored when comparing different format cameras, and that's the DoF equivalence at a required shutter speed. It's widely recognised that small format cameras, particularly P&S digicams, are of little use for creative photography employing shallow DoF. But the reverse is equally true. They are of great use for getting good DoF at wide apertures and fast shutter speeds.

Yes, but not if there is some theoretical limit to lens sharpness. And that is what we are alluding to in these discussions. Specifically now that we have a bunch of 10mpx APS-C cameras, and possibly a 1 series successor with anything between 22mp to 26mp.

The reason I like the comparison between a 5D sensor and a 20D sensor is that it is the closest thing to apples vs apples. Comparing a 5D sensor to a Sony chip has been less than useful in the past. Yes, the Sony chip is a lot more noisy, but that seems attributable to other aspects of design.
Title: Sensor & Sensibility II
Post by: bjanes on August 25, 2006, 03:20:49 pm
Quote
The big pixels that capture more electrons per pixel will have higher signal-to-noise in the highlights, at the pixel level.  The blackframe noise, however, limits the dynamic range.  "Maximum signal divided by blacklevel noise" is the determiner of dynamic range.  "Maximum signal divided by the noise of *that* signal" is not DR, and the latter is what counting more photons does for you.

IOW, the bigger pixels mean a lot less noise in the highlights, and in the midtones, and even the brightest shadows (to a lesser extent), but do not help as much in the deeper shadows, and therefore with DR.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=74463\")

Jobn,

From your previous posts, I know that you are quite well versed in technical matters, but I must respectfully disagree with the above assertion. As you say, dynamic range is defined as full well electrons/read noise.

[a href=\"http://www.photomet.com/library_enc_dynamic.shtml]http://www.photomet.com/library_enc_dynamic.shtml[/url]

Read noise can be determined from very short exposures with the lens cap on as Roger Clark describes on his web site, but your use of blacklevel noise could cause confusion with dark current, which becomes bothersome with much longer exposures such as in astrophotography.

http://www.clarkvision.com/imagedetail/eva...-1d2/index.html (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)

Other things being equal, a large sensor will have more dynamic range than a smaller one because the numerator of the above equation will be larger. The read noise of a large and small sensor may be similar, but the effect of n electrons of read noise will be smaller with the large pixel when the number of electrons of read noise is converted into a pixel data number, since the large pixel collects more electrons (i.e. the gain is larger with the large pixel). Roger discusses this matter further in this post:

http://www.clarkvision.com/imagedetail/doe...tter/index.html (http://www.clarkvision.com/imagedetail/does.pixel.size.matter/index.html)

With most digital cameras, highlight noise is dominated by shot noise and deep shadow noise by read noise, as you properly state. However, with large pixels, the effect of read noise on the data number will be less as explained above.
Title: Sensor & Sensibility II
Post by: 32BT on August 25, 2006, 03:39:23 pm
Quote
With most digital cameras, highlight noise is dominated by shot noise and deep shadow noise by read noise, as you properly state. However, with large pixels, the effect of read noise on the data number will be less as explained above.
[a href=\"index.php?act=findpost&pid=74473\"][{POST_SNAPBACK}][/a]

Please don't get caught up in definitions or theory. (Perhaps the original thread is useful for that). I'm interested in the real-world, visible benefits. If you personally feel that some theory is the correct theory, do you then possibly have images to share that clearly show the corresponding benefits. Where are the higher dynamic range images? Where are the less noisy images?

I mean the significant images. The ones that justify the heated discussions. The ones that clearly show that 10 mpx is a step back if you happen to believe that. The ones that clearly show that paying a premium for FF, large pixel-bins has some substantial gain.

Or in case of P&S: if you take a 10mpx image and reduce the size to 6mpx, do those 6mpx then look a lot more pleasing due to anti-aliasing of the processing artifacts than a 6mpx image from a 6mpx sensor straight?
Title: Sensor & Sensibility II
Post by: bjanes on August 25, 2006, 04:46:55 pm
Quote
Please don't get caught up in definitions or theory. (Perhaps the original thread is useful for that). I'm interested in the real-world, visible benefits. If you personally feel that some theory is the correct theory, do you then possibly have images to share that clearly show the corresponding benefits. Where are the higher dynamic range images? Where are the less noisy images?

I mean the significant images. The ones that justify the heated discussions. The ones that clearly show that 10 mpx is a step back if you happen to believe that. The ones that clearly show that paying a premium for FF, large pixel-bins has some substantial gain.

Or in case of P&S: if you take a 10mpx image and reduce the size to 6mpx, do those 6mpx then look a lot more pleasing due to anti-aliasing of the processing artifacts than a 6mpx image from a 6mpx sensor straight?
[a href=\"index.php?act=findpost&pid=74477\"][{POST_SNAPBACK}][/a]

Opgr,

Do we need to do a drag race to show that a Porsche turbo can outrun a VW? The differences  between P&S and a large pixel camera such as the 5D are obvious, especially with higher ISO. Most noise measurements are handicapped by the in camera application of NR, which reduces detail along with noise. Sharpening also accentuates noise.

Meaningful tests require raw with no NR or sharpening, and many P&S do not even offer these options. Even so, just look at some of the images in DPReview between the 5D and the P&S cameras.

The Clark website compares P&S to the EOS 1D. The differences are dramatic. Why do photographic tests to show what should be obvious? I do not really see why Michael's essay has provoked so much controversy. What he said is generally accepted.
Title: Sensor & Sensibility II
Post by: 32BT on August 25, 2006, 05:07:53 pm
Quote
Do we need to do a drag race to show that a Porsche turbo can outrun a VW? The differences  between P&S and a large pixel camera such as the 5D are obvious, especially with higher ISO.

I never referenced a comparison between DSLR and P&S.

The comparisons are between large pixel-bins a-la 5D vs smaller pixel-bins a-la 20D and whether the increase to 10mpx APS-C is sensible. The other comparison is between 6mpx and 10mpx P&S and whether that increase is sensible.

The former comparison also relates to a possible 1-series upgrade to 22+ mpx. The latter comparison may not be complete if we don't look at a reduced scale image to see if a 6mpx file from a 10mpx sensor is better than a 6mpx file from a 6mpx sensor straight...
Title: Sensor & Sensibility II
Post by: Ray on August 25, 2006, 09:04:22 pm
Quote
Yes, but not if there is some theoretical limit to lens sharpness. And that is what we are alluding to in these discussions. Specifically now that we have a bunch of 10mpx APS-C cameras, and possibly a 1 series successor with anything between 22mp to 26mp.

Oscar, with all due respect this is a red herring. Lens sharpness is always a concern whatever camera one is using and one certainly needs to know one's lenses if one is serious about these issues you've raised. For the comparison you've suggested, both cameras use the same lenses and most 35mm lenses have a range of a few stops where, for real world practical purposes, lens resolution is as close as matters. After all, we're talking about a 1 1/2 stop difference. Sometimes this difference will favour the 20D (eg f6.3 compared with f11, or f11 compared with f18) and sometimes it will favour the 5D (f2 compared with f3.5).

It should also be mentioned that system resolution does fall off to some degree at high ISOs. For example, if you want to see clear, undeniable differences between the 5D and 1Ds2, check out the dpreview images at ISO 1600 for these cameras. You'll find that the 1Ds2 is quite noticeably sharper (but also noisier) than the 5D at this ISO.

One should also bear in mind that a lens at (or close to) full aperture tends to lose performance around the edges, which doesn't concern the 20D. Of course, if you know a lens has noticeably worse resolution at a particular aperture, one would try to avoid using it for such comparisons. It would not be ideal to compare the 100-400 IS at f5.6 (and 400mm) using the 20D with the same lens at f9 using the 5D (at the same shutter speed but different ISO and different distance for same FoV). Such a comparison would tip the balance in favour of the 5D because we know this lens is noticeably soft at full aperture.

Quote
The reason I like the comparison between a 5D sensor and a 20D sensor is that it is the closest thing to apples vs apples. Comparing a 5D sensor to a Sony chip has been less than useful in the past. Yes, the Sony chip is a lot more noisy, but that seems attributable to other aspects of design.

It is, but I believe that some of the qualitative differences you refer to are due to the fact that the 5D has over 50% more pixels and the individual pixels in both cameras appear to have similar noise characteristics.
Title: Sensor & Sensibility II
Post by: 32BT on August 26, 2006, 12:34:16 am
Quote
Oscar, with all due respect this is a red herring. Lens sharpness is always a concern whatever camera one is using and one certainly needs to know one's lenses if one is serious about these issues you've raised. For the comparison you've suggested, both cameras use the same lenses and most 35mm lenses have a range of a few stops where, for real world practical purposes, lens resolution is as close as matters. After all, we're talking about a 1 1/2 stop difference. Sometimes this difference will favour the 20D (eg f6.3 compared with f11, or f11 compared with f18) and sometimes it will favour the 5D (f2 compared with f3.5).

Ray,

It certainly isn't meant to be a red herring and I completely agree with your points and indicated differences between FF and APS-C. But I think that the whole sensibility issue revolves around the following question:

is it sensible to increase the megapixels in APS-C from 8 to 10?

people in the no-camp argue that 8 mpx;

1) is already at the limit of the resolving power of most lenses even at their optimal aperture,

2) is more than enough for most common printing needs,

3) has better image characteristics because of larger pixel-bins.

The same arguments are raised for the other two "formats", FF and P&S, but for different resolutions obviously. So we are comparing equal formats. The 5D vs 20D comparison rears its ugly head for argument 3 above, not argument 1 or FF vs APS-C differences.



Quote
It is, but I believe that some of the qualitative differences you refer to are due to the fact that the 5D has over 50% more pixels and the individual pixels in both cameras appear to have similar noise characteristics.

But doesn't this completely negate the larger pixel-bin advantages?
Title: Sensor & Sensibility II
Post by: DiaAzul on August 26, 2006, 05:53:40 am
Quote
is it sensible to increase the megapixels in APS-C from 8 to 10?

people in the no-camp argue that 8 mpx;

1) is already at the limit of the resolving power of most lenses even at their optimal aperture,

2) is more than enough for most common printing needs,

3) has better image characteristics because of larger pixel-bins.

[a href=\"index.php?act=findpost&pid=74513\"][{POST_SNAPBACK}][/a]

It would make sense to go from 8 to 10 Megapixels in line with advances in technology.

Without making this technical, the basis for my this statement is observation of improvements in the quality of images I could produce in moving from a 10D to a 1DII. Even though the 10D is 6Mpix and the 1DII is 8Mpix there is a noticeable improvement in the quality of pictures that can be produced, an increase in flexibility (or latitude) to adjust the image and a lowering of noise in the image. Most of this can be attributed to a general advance in technology rather than any particular sensor size, pixel density or pixel size.

Any comparison or generalisation made for today is unlikely to stand up in a year or two years as both camera and lens technology continue to improve.
Title: Sensor & Sensibility II
Post by: John Sheehy on August 26, 2006, 08:43:23 am
Quote
Without making this technical, the basis for my this statement is observation of improvements in the quality of images I could produce in moving from a 10D to a 1DII. Even though the 10D is 6Mpix and the 1DII is 8Mpix there is a noticeable improvement in the quality of pictures that can be produced, an increase in flexibility (or latitude) to adjust the image and a lowering of noise in the image. Most of this can be attributed to a general advance in technology rather than any particular sensor size, pixel density or pixel size.[a href=\"index.php?act=findpost&pid=74522\"][{POST_SNAPBACK}][/a]

I would think that part of it would have to be the increase in pixel-to-pixel contrast with the coarser pixel-pitch, in the center of the images, as the farther-spaced pixels don't need to strain the MTF as much as the closer-spaced ones.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 26, 2006, 08:56:37 am
Quote
I would think that part of it would have to be the increase in pixel-to-pixel contrast with the coarser pixel-pitch, in the center of the images, as the farther-spaced pixels don't need to strain the MTF as much as the closer-spaced ones.
[a href=\"index.php?act=findpost&pid=74528\"][{POST_SNAPBACK}][/a]

Have you been on a special course to teach how to write bollocks dressed as techno-babble? As per your earlier comment on dynamic range which bjanes saved me the effort of commenting on, this is a complete load of drivel put in such a way as to sound intelligent.

The only thing being stained here is credibility.
Title: Sensor & Sensibility II
Post by: Ray on August 26, 2006, 09:02:45 am
Quote
But doesn't this completely negate the larger pixel-bin advantages?
[a href=\"index.php?act=findpost&pid=74513\"][{POST_SNAPBACK}][/a]

Oscar,
It would if the 20D pixels really are of similar quality, dynamic range and S/N etc.

I suppose the way to test this would be to use the same lens on both cameras, at the same aperture and using the same ISO, but move back a certain distance with the 20D so that equal FoV crops from both images will contain the same number of pixels. Would you be able to calculate for me what that precise ratio of distance to target should be. I don't have a calculator at hand.  

I took a few shots today using my 25-105 with both cameras at different focal lengths, different apertures, different ISOs but same shutter speed and DoF. Didn't get things quite right though. As far as I can tell, there's no way of knowing the precise focal length untill the images are downloaded. I ended up comparing 75mm on the 5D with 50mm on the 20D, which is a crop factor of 1.5 and gives a slight advantage to the 20D. After an appropriate amount of sharpening, giving slightly more sharpening to the 20D image after interpolation, I find there's no noticeable differences in image quality in the central area of the image, at 67% enlargement on screen, which at my screen resolution of 1280x1024 would look like a 20x30" print, if my monitor was that big.

At 200% enlargement, representing a print size of around 5ftx7.5ft, the 5D image is slightly sharper, a bit like the difference between the 5D and 1Ds2.

The real surprise was the performance in the corners. I expected the crop factor of the 20D would deliver good results from corner to corner. Not so. The 5D image was very noticeably better in the corners; the difference between text being legible and illegible.

The 20D shot was at f6.3 and ISO 400; the 5D shot at f11 and ISO 1250. Same shutter speed for both images. Also surprisingly, the 5D did not show more noise in the shadows, despite the much higher ISO. I attribute this to the greater number of pixels of the 5D. The noise of the 20D pixels at ISO 400 is actually less, but not after the image has been interpolated to the same size as the 5D image.

I also checked resolution of the same target at f6.3 and f11 on the 5D at 75mm. Essentially there's resolution no difference at all between these two apertures, except in the corners again. It seems that f6.3 is too wide on this lens for good corner performance, even on the 20D.
Title: Sensor & Sensibility II
Post by: bjanes on August 26, 2006, 09:10:37 am
Quote
It would make sense to go from 8 to 10 Megapixels in line with advances in technology.

Any comparison or generalisation made for today is unlikely to stand up in a year or two years as both camera and lens technology continue to improve.
[a href=\"index.php?act=findpost&pid=74522\"][{POST_SNAPBACK}][/a]

Unless the laws of physics change in the near future, one generalization that is likely to stand up is that large pixels will have less noise and small pixels will have better resolution. The tradeoff between these two factors depends on the situation. At base ISO, small pixels can produce excellent results as shown with the Nikon D2X, even with its inferior sensor technology. However, high ISO performance is poor.

For sports photography and photojournalism, where lighting is often poor, the large pixel is a distinct advantage and the Canon 1DMII would be a better choice. As Michael likes to say, "horses for courses".

Whatever improvements made with small pixel sensors can also be applied to those with larger pixels, and the relative merits of the different pixel size will remain unchanged.
Title: Sensor & Sensibility II
Post by: Ray on August 26, 2006, 09:18:36 am
Quote
At base ISO, small pixels can produce excellent results as shown with the Nikon D2X, even with its inferior sensor technology. However, high ISO performance is poor.
[a href=\"index.php?act=findpost&pid=74534\"][{POST_SNAPBACK}][/a]

But once again, Bill, you don't alsways have the same need for the higher ISOs with a camera like the D2X. You can usually get the same results with a 1.5 stops bigger aperture and correspondingly lower ISO.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 26, 2006, 09:28:58 am
Quote
Unless the laws of physics change in the near future, one generalization that is likely to stand up is that large pixels will have less noise and small pixels will have better resolution. The tradeoff between these two factors depends on the situation. At base ISO, small pixels can produce excellent results as shown with the Nikon D2X, even with its inferior sensor technology. However, high ISO performance is poor.

[a href=\"index.php?act=findpost&pid=74534\"][{POST_SNAPBACK}][/a]

And also as the quality of lenses improves (i.e. better MTFs) then for the APS-C sized sensor a higher pixel pitch can be supported without running into limitations due to lens quality.

What you are fixating on is specifics of a particular argumentation and technical babble. The point I am making is that things will get better and we will see gradual increases in performance year on year. The fact that bigger pixels have less noise and smaller pixels have higher resolution is a load of bunk if we cannot SEE and increase in image quality. Oscar was asking if anyone can show or demonstrate real world improvements in image quality rather than discuss whether big or little pixels are THEORETICALLY better.

It's OK that people have amateur discussion on sensor performance, but they are just that, the musings of semi informed armchair enthusiasts. If anyone can back up the techno babble with some serious evidence then the arguments might be taken more seriously.
Title: Sensor & Sensibility II
Post by: Ray on August 26, 2006, 10:27:27 am
Quote
If anyone can back up the techno babble with some serious evidence then the arguments might be taken more seriously.
[a href=\"index.php?act=findpost&pid=74537\"][{POST_SNAPBACK}][/a]

Here's my evidence to back up the techno babble, but using an ordinary lens unfortunately. At 100% enlargement, representing a print size larger than my Epson 7600 can handle, the superior resolving power of the 5D is beginning to show. One of the 100% crops below (maximum jpeg quality) is of a 20D image at 50mm, f6.3 and ISO 400. The other is the same crop of a 5D image at 75mm, f11 and ISO 1250.
 
I think it is very likely that a 400D would have narrowed this gap to irrelevancy at any enlargement, under these conditions of same DoF and shutter speed.

[attachment=919:attachment]           [attachment=920:attachment]
Title: Sensor & Sensibility II
Post by: jani on August 26, 2006, 10:46:56 am
Quote
I suppose the way to test this would be to use the same lens on both cameras, at the same aperture and using the same ISO, but move back a certain distance with the 20D so that equal FoV crops from both images will contain the same number of pixels. Would you be able to calculate for me what that precise ratio of distance to target should be. I don't have a calculator at hand. 
Yes, you do.

Just go to Google (http://www.google.com), type your mathematical expression, and search.

If you're running Windows, there's been a calculator included with the system since version 1.0 in 1985.

If you're running MacOS X, there's a calculator widget available on F12, as well as the usual calculator tools on any sane Unix system.

You've simply got no excuse, Ray.  
Title: Sensor & Sensibility II
Post by: jani on August 26, 2006, 10:49:58 am
Quote
But once again, Bill, you don't alsways have the same need for the higher ISOs with a camera like the D2X. You can usually get the same results with a 1.5 stops bigger aperture and correspondingly lower ISO.
I think you must have typed something else than what you meant, since this seems to require that Nikon provide lenses with 1.5 stops more of headroom than Canon.
Title: Sensor & Sensibility II
Post by: bjanes on August 26, 2006, 12:38:08 pm
Quote
What you are fixating on is specifics of a particular argumentation and technical babble.

If anyone can back up the techno babble with some serious evidence then the arguments might be taken more seriously.
[a href=\"index.php?act=findpost&pid=74537\"][{POST_SNAPBACK}][/a]

Babble definition (dictionary.com):

6.   inarticulate or imperfect speech.
7.   foolish, meaningless, or incoherent speech; prattle.
8.   a murmuring sound or a confusion of sounds.

If you regard scientific analysis as babble, then I truly feel sorry for you. These sceintific  principles have been verified many times over. A camera sensor can defy the laws of phycics no better than a perpetual motion machine can defy the laws of mechanics. While you are tyring to perfect such a machine, I will be engaged in more productive work.
Title: Sensor & Sensibility II
Post by: 32BT on August 26, 2006, 12:46:09 pm
Quote
I think it is very likely that a 400D would have narrowed this gap to irrelevancy at any enlargement, under these conditions of same DoF and shutter speed.

On the contrary.

This is exactly what makes it interesting. How do people judge image quality, and, more importantly, how does one judge the overal picture subjectively, especially in the wake of processing artifacts. The processing artifacts may not be clearly defined at certain sizes, but they may certainly impact the overal subjective picture "experience" if you will. These images are a good example:

The 5D shows clear signs of clipping or over-exposed areas, possibly because of reduced dynamic range due to higher iso.

The 20D shows clear blooming and aliasing artifacts. The latter especially will increase in number and become more pronounced in the 400D images if more processing is necessary to gain otherwise similar results. It is also these artifacts that IMO produce a less pleasing, subjective overall picture experience. From experience I know that, eventually, even the lay-man viewer can see this and experiences a kind of grainyness in the overal image that is unpleasant or even tiring.
Title: Sensor & Sensibility II
Post by: bjanes on August 26, 2006, 12:54:34 pm
Quote
What you are fixating on is specifics of a particular argumentation and technical babble.

If anyone can back up the techno babble with some serious evidence then the arguments might be taken more seriously.
[a href=\"index.php?act=findpost&pid=74537\"][{POST_SNAPBACK}][/a]

Babble definition (dictionary.com):

6.   inarticulate or imperfect speech.
7.   foolish, meaningless, or incoherent speech; prattle.
8.   a murmuring sound or a confusion of sounds.

If you regard scientific analysis as babble, then I truly feel sorry for you. These sceintific  principles have been verified many times over. A camera sensor can defy the laws of phycics no better than a perpetual motion machine can defy the laws of mechanics. While you are tyring to perfect such a machine, I will be engaged in more productive work.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 26, 2006, 05:25:09 pm
Quote
These sceintific  principles have been verified many times over.
[a href=\"index.php?act=findpost&pid=74551\"][{POST_SNAPBACK}][/a]

You mean scientific principles such as the earth is flat...

I don't mind a good qualified, well justified and rational discussion. But a lot of what is being regurgitated here is just a bunch of sheep recounting what they have read elsewhere on the internet - it is about as far away from rigorous scientific analyis as you can get.

The key point I continue to try and make is that technology will advance, image quality will improve and customer requirements will continue to move forward to capitalise on new equipment and what it can do for them. As far as I understand that is what Oscar was trying to postulate at the start and to try and avoid a rehashed technical discussion that is marginally relevant to most people who are interested in actually taking pictures rather than slaving over technical nuances that may, or may not, actually matter at the end of the day.

People who obsess about the size of their pixels are just a bunch of boring internet nerds with no real relevance to actual photography. Do we really have to judge a person by the size of their pixel? Is it important that you have a high dynamic range? Or is that a 1DsIII in you pocket or are you just pleased to see her?

As Cuba Gooding Jr once said to Tom Cruise - Show me the image. If you can talk the talk, the please (for heavens sake) show us that all this technology actually translates into better pictures.
Title: Sensor & Sensibility II
Post by: Ray on August 26, 2006, 10:37:00 pm
[/quote]

Quote
The 5D shows clear signs of clipping or over-exposed areas, possibly because of reduced dynamic range due to higher iso.

More likely because of a lack of a completely rigorous methodology. The illumination for this scene was daylight through the lounge window, which during the course of stuffing around, changing cameras and lenses, trying to get focussing accurate etc etc, would have changed slightly during the period of the testing process. I took a series of shots at 1/3rd stop intervals at various ISO settings on both cameras and then tried to match same exposure shots that were fully exposed to the right. I converted in ACR with shadows and contrast settings at zero and did my best to get the histograms looking the same, with appropriate amounts of negative EC. But the amounts are not identical in these comparisons. There's was a limit to the amount of time I was prepared to spend to get every aspect completely accurate, especially when the tests were flawed from the beginning by using a 75mm focal length with the 5D instead of 80mm. (Both cameras were used on the same tripod from the same position).

Quote
The 20D shows clear blooming and aliasing artifacts.

On a 20x30" print? I think I know what you are referring to. Below is a pair of 400% crops showing an enlarged specral highlight at the bottom fringed by a more pronounced, jagged edge in the 20D crop. Is this the sort of thing you mean?

This 400% crop is 5 1/2 inches wide on my screen and, by my calculation, represents a print size of 10ftx15ft. The image is in vertical orientation.

If I were to make a print this size, for any unimaginable reason, I'd be uprezzing with Genuine Fractals and no doubt introducing a lot more artifacts. I think we're into serious pixel peeping here, Oscar   .

[attachment=923:attachment]
Title: Sensor & Sensibility II
Post by: Ray on August 26, 2006, 11:00:07 pm
Quote
I think you must have typed something else than what you meant, since this seems to require that Nikon provide lenses with 1.5 stops more of headroom than Canon.
[a href=\"index.php?act=findpost&pid=74547\"][{POST_SNAPBACK}][/a]

Jani,
I should have emphasised the word always. If both the Nikon D2X user and 5D user have an f2 lens of equivalent focal lengths so they can get the same shot from the same distance, the 5D user gets the opportunity to take a shot which is less noisy but which also has less DoF. If the shallowest DoF possible is your goal, the 5D user has a clear advantage of both a less noisy image and one with shallower DoF. As far as I know, there is no 10mm or 11mm f2.8 lens for the cropped format, equivalent to the Canon 16-35/2.8, which is another disadvantage for D2X users. Is that right?
Title: Sensor & Sensibility II
Post by: John Sheehy on August 26, 2006, 11:14:00 pm
Quote
The 20D shows clear blooming and aliasing artifacts.[a href=\"index.php?act=findpost&pid=74555\"][{POST_SNAPBACK}][/a]

Can you link to a 20D image that has blooming?

I've never heard of blooming in Canon CMOS sensors.

The purple halos you get around extremely bright edges are infrared light out of focus, IMO.
Title: Sensor & Sensibility II
Post by: John Sheehy on August 26, 2006, 11:25:48 pm
Quote
Have you been on a special course to teach how to write bollocks dressed as techno-babble? As per your earlier comment on dynamic range which bjanes saved me the effort of commenting on, this is a complete load of drivel put in such a way as to sound intelligent.

The only thing being stained here is credibility.
[a href=\"index.php?act=findpost&pid=74531\"][{POST_SNAPBACK}][/a]

I've always been intrigued by the fact that moderated forums always have more rude people than usenet, where satan himself can post and say anything he wants.

... and what I said made perfect sense.  A sensor with a coarser pixel pitch doesn't require the MTF performance of one with a finer pixel pitch.  Is that so difficult to comprehend?
Title: Sensor & Sensibility II
Post by: Ray on August 27, 2006, 12:01:03 am
Quote
If you're running Windows, there's been a calculator included with the system since version 1.0 in 1985.
[a href=\"index.php?act=findpost&pid=74545\"][{POST_SNAPBACK}][/a]

Sometimes the obvious escapes me   . So, having done a few precise sqrt calculations, instead of moving backwards 1.6x the distance to the target with the 20D (to get same FoV as the 5D with same lens), I should move back just 1.265x the distance, to get an equal number of pixels in equal FOV crops. Is that right?
Title: Sensor & Sensibility II
Post by: 32BT on August 27, 2006, 02:09:30 am
Quote
More likely because of a lack of a completely rigorous methodology. The illumination for this scene was daylight through the lounge window,

way to go Ray. F**k up the primary, most fundamental step in the Photographic process. Read the sarcasm. This test wasn't flawed because of the lens issue...
Bad dog Ray, bad bad dog...


Quote
This 400% crop is 5 1/2 inches wide on my screen and, by my calculation, represents a print size of 10ftx15ft. The image is in vertical orientation.

If I were to make a print this size, for any unimaginable reason, I'd be uprezzing with Genuine Fractals and no doubt introducing a lot more artifacts. I think we're into serious pixel peeping here, Oscar   .

Nope, it doesn't work that way, and you seem to miss my point.

Let me try to give you a visual analogy. We all know the common wisdom from the printing industry that a good print requires about 2x the raster lpi. Which translates to about 300dpi resolution for a common print situation and which is now also used for inkjet printing.

But have you ever wondered what the resolution for text and line elements is supposed to be according to that same industry?

It ranges anywhere from 1200dpi to 2400dpi. Why is that?
Title: Sensor & Sensibility II
Post by: 32BT on August 27, 2006, 03:29:58 am
Quote
Can you link to a 20D image that has blooming?

I've never heard of blooming in Canon CMOS sensors.

The purple halos you get around extremely bright edges are infrared light out of focus, IMO.

That may be true. I was referring to the metal piece. But especially the left side is intriguing. The 20D image shows relatively large blobs of discolorisation. The 5D image, which appearantly has more contrasty light shining at it, doesn't seem to suffer the same issues. Could be a better IR filter, or even the lens performance of course.

(http://www.theimagingfactory.com/examples/comp20D5D.jpg)

Point is that even though we (as imaging professionals) need pixel-peeping to explain what we see, the net effect on a simple 8x5 image appearantly is a sense of graininess that is tiring to the eyes. Jagged edges are a prime example. A beautifully anti-aliased edge vs a jagged edge (you know it when you see it) makes a world of difference in final image appearance. To explain the phenomenon requires pixel-peeping of the highest degree. But to appreciate anti-aliasing requires a normal print or viewing situation.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 27, 2006, 07:35:19 am
Quote
That may be true. I was referring to the metal piece. But especially the left side is intriguing. The 20D image shows relatively large blobs of discolorisation. The 5D image, which appearantly has more contrasty light shining at it, doesn't seem to suffer the same issues. Could be a better IR filter, or even the lens performance of course.
[a href=\"index.php?act=findpost&pid=74619\"][{POST_SNAPBACK}][/a]

This has been a long standing issue around specular highlights and contrasty edges and (IMHO) is generated by a combination of clipping of the RAW data due to insufficient dynamic range in the sensor and the effect of interpolation of beyer data to generate the final image.

The clipping of the RAW data is before application of the tonal curve, so the in camera histogram is not particularly useful for assessing the impact of this problem at the time that the shot is taken and, as these are specular highlights, it is unlikely that you would see them in the histogram either - a classic case where exposing to far to the right has its draw backs.

I suspect that because the 5D has a greater dynamic range then the green channel is clipping much later than for the 20D and, therefore, it is possible during the interpolation process to recover the colour information. However, for the 20D (and areas where a colour shift is perceptible in the 5D image) the green channel looks as if it has blown out first whilst either the red and/or blue channels continue to increase in value. As the RAW converter doesn't know by how much the green channel has blown out then it has to guess at the colour for those pixels where this effect has taken place. Where there is a continuous tone then it is not unreasonable to extend the same colour information across the surface, however, at edges transitioning from one tone to another then it becomes very difficult to best guess the required colour information and the RAW convertor tends to get things wrong.

Two further points:
1/ You appear to be applying a high level of capture sharpening to your images. This will emphasise any 'jaggies' along the edges. Try easing back a bit on the sharpening if you are getting overly harsh and noisy pictures.

2/It is very unlikely that IR will affect the sensor, Canon (much to the disappointment of IR photographers) has a very tight IR and UV filter on the front of the sensor which blocks most IR radiation from ever getting near the sensor itself. However, the UV filter doesn't appear to be quite so tight which affects flower photographers who tend to get blown out details in the petals if an additional UV filter is applied or exposure compensation is used.
Title: Sensor & Sensibility II
Post by: Ray on August 27, 2006, 07:52:22 am
Quote
This test wasn't flawed because of the lens issue...

Good! A 75mm FL instead of 80mm gives a slight advantage to the 20D, but I think it's quite fair to give an advantage to the underdog, don't you?  

As I see you are a painstakingly thorough sort of guy, I'll repost the images without any sharpening, WB adjustment or tweaking of the histogram. It has now occurred to me that the exposure differences in the earlier examples could have been partly due to the discrepancy in FL equivalence. The 5D shot takes in a slightly greater expanse of lighter shades. The true ISO values might also vary to different degrees between the 2 cameras, although I get the impression they are roughly the same.

I've converted the following images in ACR, with zero shadows and contrast, no sharpening, no luminance smoothing, no post processing, no nothing except EC adjustment, cropping. conversion to srgb and maximum quality jpeg compression.

I've reselected the images, but to get the histograms looking the same, I'm comparing a 5D image with 1/5th sec exposure and minus 0.9 EC, with a 20D image with 1/4 sec exposure and minus 1 EC. (Is this called fudging the results?) I've included screen shots of the histograms.

[attachment=927:attachment]  [attachment=928:attachment]  [attachment=929:attachment]  [attachment=930:attachment]

Quote
Nope, it doesn't work that way, and you seem to miss my point.

Let me try to give you a visual analogy. We all know the common wisdom from the printing industry that a good print requires about 2x the raster lpi. Which translates to about 300dpi resolution for a common print situation and which is now also used for inkjet printing.

Oscar, I'm not suggesting I would print the 10ftx15ft image at the screen resolution of 96 dpi. Of course I wouldn't. I'd interpolate the image to 240 ppi with GF, as I mentioned. The file size would then be huge. The point I'm making, which you seem to have missed, is that the 5 1/2" wide 400% crop which I'm scrutinising on my monitor from the distance one might view an 8x10" print, is part of a 10ftx15ft image. In other words, if I wanted to see the whole image at this degree of detail and at this resolution of 96 dpi, my monitor would need to be 15ft high and 10ft wide.
Title: Sensor & Sensibility II
Post by: 32BT on August 27, 2006, 09:01:40 am
Quote
Good! A 75mm FL instead of 80mm gives a slight advantage to the 20D, but I think it's quite fair to give an advantage to the underdog, don't you? 

I'm all for the underdog. Speaking of which: could you repost DPP conversions of the same files? (or post the RAW).

It seems the 20D is outresolving the lens by a fair amount, any way we can change that?

What could be the source of the rather significant chromatic aberration in the 20D file?

Would it be interesting to change the 20D iso to its optimal value, and set the 5D to what you consider the equivalent?

Quote
Oscar, I'm not suggesting I would print the 10ftx15ft image at the screen resolution of 96 dpi. Of course I wouldn't. I'd interpolate the image to 240 ppi with GF, as I mentioned.
I know, but that was not the question. The question is: why does the industry want 1200dpi to 2400dpi for text and line-art?
Title: Sensor & Sensibility II
Post by: 32BT on August 27, 2006, 09:20:11 am
Quote
This has been a long standing issue around specular highlights and contrasty edges and (IMHO) is generated by a combination of clipping of the RAW data due to insufficient dynamic range in the sensor and the effect of interpolation of beyer data to generate the final image.

But this would thus result in a clear and obvious sample of better performance of larger-pixel-bin sensors vs smaller-pixel-bin. That is, at the current state of affairs. I completely agree with you that the performance of smaller-pixel-bins can and will increase.

As a matter of fact, eventually the larger-pixel-bin advantage may for example be countered by speed advantages of reading smaller-pixel-bins. I'm purely speculating here, but I recall research about taking multiple exposures in a single exposure. Eventually the read-out speed may make this viable for smaller-pixel-bins, but larger pixel-bins simply don't "fill up" quick enough, their "native" iso being to high?

Quote
2/It is very unlikely that IR will affect the sensor, Canon (much to the disappointment of IR photographers) has a very tight IR and UV filter on the front of the sensor which blocks most IR radiation from ever getting near the sensor itself. However, the UV filter doesn't appear to be quite so tight which affects flower photographers who tend to get blown out details in the petals if an additional UV filter is applied or exposure compensation is used.

But Canon made a relatively big deal out of the newer filters applied in the 5D. So there may well be a difference between a 20D and 5D. It also begs the question: Are these newer filters applied in the new 400D?
Title: Sensor & Sensibility II
Post by: John Sheehy on August 27, 2006, 09:30:23 am
Quote
If you're running Windows, there's been a calculator included with the system since version 1.0 in 1985.
[a href=\"index.php?act=findpost&pid=74545\"][{POST_SNAPBACK}][/a]

It has no square root function, per se.  You have to use the x^y with 0.5 as the exponent.
Title: Sensor & Sensibility II
Post by: John Sheehy on August 27, 2006, 09:45:43 am
Quote
That may be true. I was referring to the metal piece. But especially the left side is intriguing. The 20D image shows relatively large blobs of discolorisation. The 5D image, which appearantly has more contrasty light shining at it, doesn't seem to suffer the same issues. Could be a better IR filter, or even the lens performance of course.


It can also be the 5D's coarser pixel pitch.  Imagine that the pixels were even bigger, and the fringes occured only in a very small percentage of any given pixel.  The effect will be almost invisible.  This is the fact that Diazul argued against and insulted me for; the 5D pixels gather more high contrast material, and less blurred edge effects, because optical edges are a smaller percentage of any given pixel's capture.

Quote
Point is that even though we (as imaging professionals) need pixel-peeping to explain what we see, the net effect on a simple 8x5 image appearantly is a sense of graininess that is tiring to the eyes. Jagged edges are a prime example. A beautifully anti-aliased edge vs a jagged edge (you know it when you see it) makes a world of difference in final image appearance. To explain the phenomenon requires pixel-peeping of the highest degree. But to appreciate anti-aliasing requires a normal print or viewing situation.
[a href=\"index.php?act=findpost&pid=74619\"][{POST_SNAPBACK}][/a]

For that I thank Canon for not falling into the trap of creating false detail sharpness with very weak or absent AA filters, like some other MFRs have done.  I also appreciate that they give you fairly raw RAW files, complete with "negative" values (less than black).  Even though Canon itself doesn't seem to use this data in its converters for anything, I find it invaluable for subtracting line noise in extremely underexposed images.  I hope these things remain for future cameras.
Title: Sensor & Sensibility II
Post by: John Sheehy on August 27, 2006, 10:22:42 am
Quote
This has been a long standing issue around specular highlights and contrasty edges and (IMHO) is generated by a combination of clipping of the RAW data due to insufficient dynamic range in the sensor and the effect of interpolation of beyer data to generate the final image.

That certainly acounts for flat clipped areas, without the necessary intelligence to recognize clipping in the converter.

Quote
I suspect that because the 5D has a greater dynamic range then the green channel is clipping much later than for the 20D

That's not true at all ISOs. 100 has greater DR by about a half-stop (part of the reason is that 0.25 stops is missing in the 20D), and the highest ISOs have greater on the 5D, but at 200 and 400, they are roughly the same.

Quote
and, therefore, it is possible during the interpolation process to recover the colour information. However, for the 20D (and areas where a colour shift is perceptible in the 5D image) the green channel looks as if it has blown out first whilst either the red and/or blue channels continue to increase in value. As the RAW converter doesn't know by how much the green channel has blown out then it has to guess at the colour for those pixels where this effect has taken place. Where there is a continuous tone then it is not unreasonable to extend the same colour information across the surface, however, at edges transitioning from one tone to another then it becomes very difficult to best guess the required colour information and the RAW convertor tends to get things wrong.

I certainly hope that interpreting this as color never becomes a converter default.  

Quote
2/It is very unlikely that IR will affect the sensor, Canon (much to the disappointment of IR photographers) has a very tight IR and UV filter on the front of the sensor which blocks most IR radiation from ever getting near the sensor itself.

It is only a relative effect that the IR-cut filter has.  It cuts a percentage; it is not 100% effective.  When you get specular highlights that concentrate light on the sensor that is several stops stronger than 100% matte reflectance, and you are metering for 12-18%, the IR content is also much stronger than usual in that highlight.  If the lens focuses IR much differently than it does visible light, then you get bokeh where the IR should be focused, creating a halo.  This *has* to be happening at some level, even if you discard the idea as the main reason.

I have looked at other people's test results, and the purple fringing effect does not vary with exposure, and it varies quite a bit with lenses.  That is one reason that I conclude that it is OOF IR.  If you shoot IR with a 93 filter, the results are purple.

Quote
However, the UV filter doesn't appear to be quite so tight which affects flower photographers who tend to get blown out details in the petals if an additional UV filter is applied or exposure compensation is used.
[a href=\"index.php?act=findpost&pid=74629\"][{POST_SNAPBACK}][/a]

What kind of petals are you talking about?  The typical blown-out red petal is not blown out or even close to blowing out in the RAW data.  It is an artifact of converters trying to get a saturation in the color space that approaches the petal in human perception.  The blue channel is very dark in red petals, the red is strongest but green is fairly strong because of the overall higher sensitivity in the green channel.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 27, 2006, 11:47:22 am
Quote
It cuts a percentage; it is not 100% effective.  When you get specular highlights that concentrate light on the sensor that is several stops stronger than 100% matte reflectance, and you are metering for 12-18%, the IR content is also much stronger than usual in that highlight.  If the lens focuses IR much differently than it does visible light, then you get bokeh where the IR should be focused, creating a halo.  This *has* to be happening at some level, even if you discard the idea as the main reason.

[a href=\"index.php?act=findpost&pid=74646\"][{POST_SNAPBACK}][/a]

That's a fair comment. Though the cutoff on the filters is quite steep and out of focus areas are going to be spread out and difuse disipating the incident energy over multiple pixels. The purple halos are very intense and it would take a large amount of incident light to achieve the effect which is visible in the images.

For specular highlights I would tend to agree with what you have described, however, this is only so where the source of light is particularly strong. Where the sensor is only just clipped then other effects which are the result of interplay between clipping, bayer sensors and interpolation in the RAW convertor are probably more likely.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 27, 2006, 12:05:20 pm
Quote
But this would thus result in a clear and obvious sample of better performance of larger-pixel-bin sensors vs smaller-pixel-bin. That is, at the current state of affairs. I completely agree with you that the performance of smaller-pixel-bins can and will increase.

As a matter of fact, eventually the larger-pixel-bin advantage may for example be countered by speed advantages of reading smaller-pixel-bins. I'm purely speculating here, but I recall research about taking multiple exposures in a single exposure. Eventually the read-out speed may make this viable for smaller-pixel-bins, but larger pixel-bins simply don't "fill up" quick enough, their "native" iso being to high?

[a href=\"index.php?act=findpost&pid=74635\"][{POST_SNAPBACK}][/a]

This all comes back to a trade off and choice if which camera is most appropriate for your needs. Is it really necessary to have a really high dynamic range sensor if what you need is resolution and you are typcially shooting low contrast scenes? There are definite advantages to higher dynamic range sensors from a preservation of colour and tonality information prior to applicaton of a tonal curve. However, if you can tolerate some clipping of channels, then for most images it could be argued that the 5D is over engineered for the task and that pixels one half or quarter the size with higher density/resolution would be better - the Nikon route for instance.

There is another aspect that comes into play - john points out that the 5D/20D dynamic range converges from 200ISO upwards. This is indicative that the limitation on dynamic range is not generated as a result of the size of the pixel, but perhaps by other sources of noise in the line amplifiers, analogue to digital converters. We may need to see improvements in the support electronics around the sensor before we see any difference in performance between small and large pixels (i.e. pixel size is not the only limitation to image quality)

There has been prior discussion on alternative strategies for reading information from pixels - non linear response characteristics, refresh when full and reading multiple times during the exposure. All are technically feasible but generally introduce other problems and sources of noise which limit their usefulness. In the case of multiple samples over the course of one exposure, the main problem would be one of high clock speeds required to achieve enough samples (and higher clock speeds generate more on chip noise, higher power consumption and more heat) and also allowing sufficient time for the pixels to reset 3-5mSec during each sampling period (ie. you loose 3-5mSec of data each sample and you would severely limit the maximum effective shutter speed that could be used). The next most likely innovation is  an improved version of the foveon sensor with full colour read out per pixel - but we shall see what turns up over the next couple of months.
Title: Sensor & Sensibility II
Post by: 32BT on August 27, 2006, 01:04:35 pm
Quote
speed that could be used). The next most likely innovation is  an improved version of the foveon sensor with full colour read out per pixel - but we shall see what turns up over the next couple of months.

I would love to see a hybrid design:

A bayer sensor with two layer read out (instead of 3 as in foveon sensor), and a yellow-cyan checker board pattern to capture a full channel of green, and two checkerboard channels of red and blue. "Debayering" a checkerboard pattern is fairly easy and requires far less channel-cross-over data. The latter will be used anyway to accommodate other aspects of the image processing (like noise reduction), so I think that would be the perfect balance between the high bit advantage of a single layer vs the image integrity of capturing 3 colors.
Title: Sensor & Sensibility II
Post by: DiaAzul on August 27, 2006, 03:20:21 pm
Quote
I would love to see a hybrid design:

A bayer sensor with two layer read out (instead of 3 as in foveon sensor), and a yellow-cyan checker board pattern to capture a full channel of green, and two checkerboard channels of red and blue. "Debayering" a checkerboard pattern is fairly easy and requires far less channel-cross-over data. The latter will be used anyway to accommodate other aspects of the image processing (like noise reduction), so I think that would be the perfect balance between the high bit advantage of a single layer vs the image integrity of capturing 3 colors.
[a href=\"index.php?act=findpost&pid=74654\"][{POST_SNAPBACK}][/a]

Interesting theory, however, I am not sure how you are trying to get any advantage out of what you are suggesting. The foveon sensor is based upon the principle that different wavelengths of light are absorbed at different depths within the semiconductor and relies upon a 'stacked' set of sensors of different thickness to capture the light within a particular pass band. In a bayered sensor elecron-hole recombination takes place at particular depth within the semiconductor and varies by colour. There is nothing particular efficient about the bayer design when it comes to light gathering capability. Putting a yellow-cyan filter in front of a foveon sensor and then debayering the result would add complications - each alternate pixel would have to be of a different design i.e. a green-blue pixel would be different design from a green-red pixel and you would be filtering out light that could reasonably be captured and extracted for the final image.

Two things stand out with foveon type designs:

They should be more efficient long term compared with a bayer type sensor assuming that the fill factor can be increased to a comparable level. The issue here is not light capture, but creating a capacitor for each of the three layers within the same space as the one capacitor for a bayer device.

It may be possible to go more creative and have a four/five/six layer sensor e.g Infra-Red, Red, Green, Blue, UV for those that want more capability in their conversion to black and white.
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 04:21:34 am
Quote
I'm all for the underdog. Speaking of which: could you repost DPP conversions of the same files? (or post the RAW).

Oscar,
I haven't got DPP installed on this computer. I don't use it. Also, I don't know how to send you the RAW images. I'm on a dial-up 56kb connection.

Quote
It seems the 20D is outresolving the lens by a fair amount, any way we can change that?

As far as I know, all my lenses are sharpest at f8. Some are possibly equally sharp at f5.6. I took a few comparison shots today of the same target but using the TSE-90mm, probably my sharpest lens excluding the Canon 50/1.8 ll. Having taken a brief look at some of the images, some of which were taken at a different distance such that equal FoV crops from both cameras contain an equal number of pixels, I find no resolution advantage for the 20D with the better lens.

This notion of some DSLR sensors outresolving lenses seems very exaggerated to me. The only time I ever get a clear indication the 20D sensor is outresolving the lens is when I stop down to f16 or lower, or stop up to full aperture or close to, or compare corner resolution with centre resolution. The 5D has the advantage of delivering (outside of extreme pixel peeping) virtually equal resolution at f16 as at f8, with my lenses anyway.

Quote
What could be the source of the rather significant chromatic aberration in the 20D file?

No idea! But the word birefringence springs to mind   .

Quote
I know, but that was not the question. The question is: why does the industry want 1200dpi to 2400dpi for text and line-art?

I don't know, Oscar. You tell me   . I could offer a few guesses. Because the technology is now available; why not use it? Because it delivers an almost imperceptible additional crispness to the text? Because it looks very good to extremely shortsighted readers or readers who use high powered reading glasses or simply a magnifying glass?

I think you are evading the issue which started this diversion, which is, a 5.5" wide 400% crop of a 5D image on a 1280x1024 monitor is representative of a 10ftx15ft print of the whole image. It's true that there will be jagged edges visible on the monitor as a result of the uninterpolated 96 dpi monitor resolution, but these will tend to disappear when the file is uprezzed  for printing, and of course they will also disappear on the monitor when viewing the 400% file after uprezzing, as in the examples below of previously posted 400% crops which have now been uprezzed to 240 dpi with GF and displayed at 40% enlargement on the screen.

I suppose one could now say, the images below are more representative of a 10ftx15ft print of the entire image, because this is now a small part of the file I would send to the printer, although it's still not sharpened and processed in any other way. So it's still not yet fully and truly representative of the 10ftx15ft print. Agreed?

[attachment=931:attachment]
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 04:59:37 am
I should mention, for the benefit of those who own both a 5D and 20D and who might want to compare images at the pixel level without interpolation, my initial calculation of 1.265x the distance to target for the 20D proved to be slightly out. It was necessary to bring the tripod and camera back a further inch or two. Having done that, all crops contained close to exactly the same number of pixels. But I was curious as to why my calculations were not exact, so I went back to the basic information about these sensors, on dpreview. First, the 5D sensor is not 36mmx24mm but 35.8x23.9mm. Secondly, and most significantly, the 20D has 8.2 effective megapixels. I'd used a figure of 8mp in my calculations.

My initial calculation was 1.265x. A more accurate multiplier is 1.275. Just for the record   .
Title: Sensor & Sensibility II
Post by: 32BT on August 28, 2006, 05:30:50 am
Quote
As far as I know, all my lenses are sharpest at f8. Some are possibly equally sharp at f5.6. I took a few comparison shots today of the same target but using the TSE-90mm, probably my sharpest lens excluding the Canon 50/1.8 ll.
Okay, a 50/1.8 is definitely fine for the 20D, so that is not the problem. But the 20D image is far more blurry than is to be expected. Simple focussing differences?

Quote
I don't know, Oscar. You tell me   . I could offer a few guesses. Because the technology is now available; why not use it? Because it delivers an almost imperceptible additional crispness to the text? Because it looks very good to extremely shortsighted readers or readers who use high powered reading glasses or simply a magnifying glass?
You're kidding me, right?
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 05:48:36 am
Quote
You're kidding me, right?
[a href=\"index.php?act=findpost&pid=74715\"][{POST_SNAPBACK}][/a]

No I'm not. I thought commercial printers (as opposed to giclee printers) used 150 dpi for all purposes. I'm not involved in the printing profession. I'm a mere amateur.

I rely upon people like you to advise upon and explain such matters   .
Title: Sensor & Sensibility II
Post by: 32BT on August 28, 2006, 06:28:35 am
Quote
It can also be the 5D's coarser pixel pitch.

After digesting all the information I have come to the conclusion that this may be the core of the issue. Here's my thinking:

Supposed advantages of larger pixelbins:

1) more dynamic range,

2) less noise,

3) better per pixel contrast.


1) more dynamic range
From everything we know this is only marginally true. Using a middle-gray exposure it may be somewhat relevant but certainly not earth-shattering. For a correct ETTR highlights exposure, the difference obviously resides in the darktones where it becomes obscured in a discussion about how much noise is acceptable.
Apparently the difference becomes smaller at higher iso.
For an 8mpx to 10mpx jump it is completely irrelevant.

 2) less noise
While true, the benefits are not immediately obvious, and can possibly be harvested only through higher bit A/D conversions. For 5D vs other brands, I believe there are too many design difference for meaningful comparison. For 5D vs 20D comparison, the 5D high iso performance is remarkable and the size of the 20D sensor components may result in more secondary irregularities such as stuck pixels. For an 8mpx to 10mpx jump, it is irrelevant.

3) better per pixel contrast
Seems certainly a decisive factor, especially in a 5D vs 20D comparison. If it is the source of the fringing, then it would certainly account for part of the subjective, easy-on-the-eyes, butter-smooth images from a 5D vs a 20D. However, for an 8mpx to 10mpx jump, it is hardly relevant.


All-in-all, I think that the smaller components in an APS-C sensor vs a FF sensor do create some system/design specific problems that are compensated for in data processing, but for an 8mpx to 10mpx jump I think that none of the advantages is remotely significant. Far more significant seem the additional components such as the AA & IR filters and microlenses.
Obviously, even more significant are the interchangeable lenses and Photographic technique used. In that respect it remains a kind of ironic paradox for an entry-level designed camera to out-resolve the more affordable lenses.

I believe that one of the reasons so many heated discussions exist on this subject is that there may be a relevant difference in large pixel-bin and small pixel-bin sensors, but these are compensated for in data processing. So, the net result is that images do not contain the usual visible noise differences that people look for, nor do the technical graphs show any significant differences.

But I believe that people certainly can experience the differences between an image on the very limit of reasonable processing vs an image that is butter-smooth from the capture stage. Whether that difference truly translates to the expenses required, remains debatable. I personally think it doesn't, not in the least because of the decreasing quality in image reproduction of the past 15 years.

A 1-series with 26mpx will NOT produce those butter-smooth pixels within the next 3 years, but the resulting images will most certainly be accepted as the quality point of reference for the industry.

Even so, I would still consider purchasing a 5D if I can afford it. Not because my photographic skills justify it, not because my imaging needs require it. But there is something magical about butter-smooth pixels which then result in buttersmooth but extremely sharp images, which is simply irresistible.
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 08:01:39 am
Quote
Even so, I would still consider purchasing a 5D if I can afford it. Not because my photographic skills justify it, not because my imaging needs require it. But there is something magical about butter-smooth pixels which then result in buttersmooth but extremely sharp images, which is simply irresistible.
[a href=\"index.php?act=findpost&pid=74723\"][{POST_SNAPBACK}][/a]

Buttersmooth? What are you talking about, Oscar? One can always create a buttersmooth effect with a bit a gausian blur. Buttersmooth and sharp? That's an oxymoron, old chap.
Title: Sensor & Sensibility II
Post by: 32BT on August 28, 2006, 08:09:50 am
Quote
Buttersmooth? What are you talking about, Oscar? One can always create a buttersmooth effect with a bit a gausian blur. Buttersmooth and sharp? That's an oxymoron, old chap.
[a href=\"index.php?act=findpost&pid=74729\"][{POST_SNAPBACK}][/a]

LOL... but that's exactly what a 5D is; an oxymoron amongst digital cameras. It's neither a professional, weather-sealed 1-series, nor is it an affordable, middle-league APS-C.
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 12:19:19 pm
Quote
LOL... but that's exactly what a 5D is; an oxymoron amongst digital cameras. It's neither a professional, weather-sealed 1-series, nor is it an affordable, middle-league APS-C.
[a href=\"index.php?act=findpost&pid=74732\"][{POST_SNAPBACK}][/a]


Sounds like the best of both worlds to me.  

For what they are worth, below are a couple of uninterpolated 100% crops of the 5D and 20D, both using the TS-E 90mm at f8, 1/6th sec exposure and ISO 100, at adjusted distances to get an equal number of pixels in each crop.

There are slight differences, but nothing particularly significant to my eyes. The images are unsharpened and unprocessed as before with same settings in ACR regarding shadows, EC and contrast. The 5D image looks slightly contrastier.

[attachment=932:attachment]  [attachment=933:attachment]
Title: Sensor & Sensibility II
Post by: jani on August 28, 2006, 06:13:04 pm
Quote
No I'm not. I thought commercial printers (as opposed to giclee printers) used 150 dpi for all purposes. I'm not involved in the printing profession. I'm a mere amateur.

I rely upon people like you to advise upon and explain such matters   .
150 dpi for high contrast material such as black text on a white background is way too low.

300 dpi was considered acceptable twenty years ago.

600 dpi is nice.

1200 dpi is lovely, but can be hard to distinguish from 600 dpi without close examination.

2400 dpi is unnecessary for that purpose.  

Well, that's just my opinion, and of course you'll find differing opinions, but it's mostly quibbling.

2400 dpi in black/white is, however, quite necessary for creating rasters to emulate really smooth grey tones.

I think that when Oscar mentions the text and line art resolution for "the printing industry", he's also including grey scale rasters.

This plays back to previous discussions regarding the validity of DPReview's resolution tests for line pairs, and IIRC, the consensus was that high-contrast images and high apparent resolution go hand in hand.
Title: Sensor & Sensibility II
Post by: bjanes on August 28, 2006, 08:31:56 pm
Quote
150 dpi for high contrast material such as black text on a white background is way too low.

300 dpi was considered acceptable twenty years ago.

600 dpi is nice.

1200 dpi is lovely, but can be hard to distinguish from 600 dpi without close examination.

2400 dpi is unnecessary for that purpose. 

Well, that's just my opinion, and of course you'll find differing opinions, but it's mostly quibbling.

2400 dpi in black/white is, however, quite necessary for creating rasters to emulate really smooth grey tones.

I think that when Oscar mentions the text and line art resolution for "the printing industry", he's also including grey scale rasters.

This plays back to previous discussions regarding the validity of DPReview's resolution tests for line pairs, and IIRC, the consensus was that high-contrast images and high apparent resolution go hand in hand.
[{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=74782\")

In this context, it is important to distinguish between pixels and halftone dots such as in Postscript. In the latter case, a halftone dot may be represented by a 16 by 16 raster pattern so that 256 levels can be represented. In this case one pixel would contain 16*16 dots.

[a href=\"http://www.ps.missouri.edu/ps2/support/TutorialFolder/ScanningTutorial/ScanningTutorial.html]http://www.ps.missouri.edu/ps2/support/Tut...ngTutorial.html[/url]
Title: Sensor & Sensibility II
Post by: Ray on August 28, 2006, 08:46:51 pm
Quote
150 dpi for high contrast material such as black text on a white background is way too low.

[a href=\"index.php?act=findpost&pid=74782\"][{POST_SNAPBACK}][/a]

Actually, I realise now that I'm using the wrong terminology. Offset printers and the like provide 150 lines per inch (lpi not dpi) and usually require an image ppi of twice that.

I fail to see why high image resolutions of 1200ppi and 2400ppi would serve any purpose outside of this 'apparent' characteristic of commercial printers to require double the resolution in the file.

I mean, how do these high resolutions equate to the accepted wisdom that the human eye cannot distinguish finer detail than 6.7 lp/mm on the printed page? That's 167 line pairs per inch or 335 ppi. Call it 360 ppi if you like.

I was not aware that either line art or text in my images would require an image file resolution of greater than 360 ppi.
Title: Sensor & Sensibility II
Post by: 32BT on August 29, 2006, 01:21:15 am
Quote
I mean, how do these high resolutions equate to the accepted wisdom that the human eye cannot distinguish finer detail than 6.7 lp/mm on the printed page? That's 167 line pairs per inch or 335 ppi. Call it 360 ppi if you like.


Because there is a fundamental difference between pattern recognition and sharp focus, you can easily discern a hairline in print which may represent 1/1200th of an inch. Additionally to create such a pattern, you would have to overcome dot-gain which means the black lines in the line pairs have to be slightly smaller than the white spacing. How can you make slightly smaller lines in a regular grid? Only if the lines are comprised of even smaller dots, and a lot of them i might add...
Title: Sensor & Sensibility II
Post by: 32BT on August 29, 2006, 01:22:58 am
Quote
Sounds like the best of both worlds to me.   

For what they are worth, below are a couple of uninterpolated 100% crops of the 5D and 20D, both using the TS-E 90mm at f8, 1/6th sec exposure and ISO 100, at adjusted distances to get an equal number of pixels in each crop.

There are slight differences, but nothing particularly significant to my eyes. The images are unsharpened and unprocessed as before with same settings in ACR regarding shadows, EC and contrast. The 5D image looks slightly contrastier.

[attachment=932:attachment]  [attachment=933:attachment]
[a href=\"index.php?act=findpost&pid=74755\"][{POST_SNAPBACK}][/a]

Interesting, something is blurring the 20d image. It's visible at normal size. See the text in the lower-right corner. While larger in the 20d image, it's readable in the 5d image.

If this is solely due to the AA filter, then it again begs the question: does the new 400D have the new filter? Given that the 5d earned its nickname (5-Dust) purportedly because of the new filter, and given that the 400D has dust-busting technology build in, I would speculate it does.

Of course, it could also be a resolving issue again or a combination.

Additionally, the 5d image shows no highlight aberrations, while it does preserve shadow transparency really well. See the left background for example which is flattened in the 20D image, and the coin stand in the 5d image renders just beautifully. This leads me to believe that the higher dynamic range is real and visible. I suppose both images were shot at iso 100?


As for line-art and text:
If you use low-resolution and anti-aliasing (if that were possible in print), then you would get tired real soon reading a page. The text looks smudged, appears to float, and you will continuously try to refocus your eyes.
So you have to create sharp type-setting without staircasing which, depending on the printing process, requires 1200dpi to 2000dpi. Your average laser print obviously requires less resolution than a quality offset print on glossy paper.

Of course, our perception processes text and line-art differently from images. Irregularities are more easily perceived and distracting, which is why the seemingly excessive text resolution requirement is as high as it is, and which is why you can't use raster for anti-aliasing.

But in my opinion the same processes are at play in image viewing. Given enough slight aberrations, no matter how small pixel wise, it will generate the same type of viewing experiences as anti-aliased text. If you try to overcome the aberrations by sharpening, it has the tendency to create irregularities in the image, exactly as it would in text or line-art.
Title: Sensor & Sensibility II
Post by: Ray on August 29, 2006, 07:50:47 am
Quote
Because there is a fundamental difference between pattern recognition and sharp focus, you can easily discern a hairline in print which may represent 1/1200th of an inch. Additionally to create such a pattern, you would have to overcome dot-gain which means the black lines in the line pairs have to be slightly smaller than the white spacing. How can you make slightly smaller lines in a regular grid? Only if the lines are comprised of even smaller dots, and a lot of them i might add...
[a href=\"index.php?act=findpost&pid=74817\"][{POST_SNAPBACK}][/a]

Oscar,
I still don't understand. A line 1/200th of an inch wide can be represented by a resolution of 100 line pairs per inch or 4 lp/mm. To print 4 lp/mm (100 lp/i) on an 8x12" print, it seems that an image file resolution of 240 ppi is sufficient, that is, 2.4 pixels for each line pair, or 1.2 pixels for each line.

Dot gain might often be a problem, but surely less of a problem with high quality, glossy paper, which you need to use for high resolution work. I'm not sure how small a dot a picolitre of ink produces on glossy paper, after it has spread out, but I imagine a lot smaller than the eye can see. I've read that a 6 picolitre droplet results in a 45 micron dot about half the width of a human hair. Don't know if that's true. Have you got any specific and accurate information on this issue?

I understand that commercial printers are a different ball game. Different file sizes may be required for different printing processes, but on this site we're mainly concerned with desktop printing and inkjet (or bubble jet) technology where we often have 2 or more shades of gray, 2 shades each of magenta and cyan and 4 picolitre (or smaller) ink droplets, aren't we?
Title: Sensor & Sensibility II
Post by: Ray on August 29, 2006, 08:39:15 am
Quote
Additionally, the 5d image shows no highlight aberrations, while it does preserve shadow transparency really well. See the left background for example which is flattened in the 20D image, and the coin stand in the 5d image renders just beautifully. This leads me to believe that the higher dynamic range is real and visible. I suppose both images were shot at iso 100?

Oscar,
Both images were shot at ISO 100. I have to agree that in general the 5D image is superior in lots of subtle ways, but there's still a question as to how significant these differences are, after skillful processing in PS (not something I consider myself an expert on).

Getting back to the whole-image print size that these 100% crops represent, I calculate that the 20D image, at my monitor resolution of 96 dpi, which would later be interpolated to 240 dpi for an Epson printer if I wanted to print it, would produce a 37.2"x24.8" print and the 5D whole image a 45.5"x30.3" print.

Since this new thread was started in order to examine real world implications of different sensor sizes, I really believe that the significance of these differences ultimately should relate to specific print sizes, but that would take a lot of time, ink and paper and I'm now preparing for my next photographic trip to SE Asia.
Title: Sensor & Sensibility II
Post by: 32BT on August 29, 2006, 09:35:35 am
Quote
Oscar,
I still don't understand. A line 1/200th of an inch wide can be represented by a resolution of 100 line pairs per inch or 4 lp/mm.

the denominator reads 1200 (twelve hundred)...

try this with your own printer:

1) create a 1440 x 1440 pixel image.
2) make a hairline cross (1 pixel wide line, both horizontal & vertical)
3) print the file at 1x1 inch.


4) Now duplicate the layer and set blend mode to multiply
5) offset the new layer 2 pixels in both horizontal and vertical directions so that a single pixel wide gap results
6) print the file again

See if the print shows two distinct lines almost touching (you may use a loupe at this stage)
if not, nudge 1 more pixel print again until you do.

You'll then know approximately how wide your printer is printing the line, and you'll also be convinced that you can see a lot more than 1/360th of an inch... Note also that the second print shows a clearly darker plus than the first.
Title: Sensor & Sensibility II
Post by: Ray on August 29, 2006, 09:49:14 am
Quote
the denominator reads 1200 (twelve hundred)...
[a href=\"index.php?act=findpost&pid=74840\"][{POST_SNAPBACK}][/a]

Oops! Just goes to show that sometimes what we think we see is influenced by pre-conceptions, or maybe I just need better spectacles   .

Interesting! Unfortunately, I haven't got time to try these experiments right now. I'll have to leave it for another day.

Cheers!
Title: Sensor & Sensibility II
Post by: 32BT on August 29, 2006, 09:57:36 am
Quote
Both images were shot at ISO 100. I have to agree that in general the 5D image is superior in lots of subtle ways, but there's still a question as to how significant these differences are, after skillful processing in PS

And that's exactly what my original question was about. I believe we can conclude that:
1) it is significant when comparing 5D vs 20D, and likely can be extrapolated to comparisons with other designs/brands,
2) in general most people don't see the differences, because the camera or RAW conversion does most of that processing for you already,
3) for high-end needs it is certainly worth the money, but in many other cases it remains questionable,
4) a 26mpx 1 series will not make the grade in the foreseeable future. 22mpx may be stretching it already,

and, most importantly:

5) there is no reason to suspect a potential degradation in image quality when jumping from 8mpx to 10mpx in APS-C, solely based on pixel-bin size differences. Image quality was bad to begin with  



Quote
Since this new thread was started in order to examine real world implications of different sensor sizes, I really believe that the significance of these differences ultimately should relate to specific print sizes, but that would take a lot of time, ink and paper and I'm now preparing for my next photographic trip to SE Asia.

Well, happy shooting, have a save one, and bring us back some nice images, because ultimately that is what this really is about, ain't it?