The dangers of pixel peeping

[Ray]

BTW, Bruce Fraser points out that you should always check sharpening at 100% with Photoshop, or else the image will be alaised.

Of course. Getting the image right for a particular print size is important. But that does not mean I should not also check or examine effects at other degrees of enlargement. If I want to start counting lines, it's simply easier at 400% magnification. My examples above are not recommendations for finished prints of a particula size, but demonstrations of increased contrast in conjunction with increased micro detail. Whether or not such USM settings are appropriate for a particular image at a particular size, is up to the photographer. I'm striving to propduce works of art, not blindly follow a set of rules. I'll use whatever effect appeals to me.

Did you ever wonder why checking the resolution charts on DPReview is so frustrating if you use 400% in PS?

No, I never did. I've never bothered copying dpreview test charts into Photoshop. You're a greater pixel peeper than I, Bill   . However, examining the test chart images you've posted, I have no problem counting the lines at either end of either image because both images are the same size. The top image, however, looks as though it needed resampling before sharpening, to avoid the jaggies.

Having copied your test chart image into Photoshop, I find it curious that all filters are disabled. How do they do that?

[bjanes]

Having copied your test chart image into Photoshop, I find it curious that all filters are disabled. How do they do that?
[a href=\"index.php?act=findpost&pid=70979\"][{POST_SNAPBACK}][/a]

Ray,

I that is because these filters to not work on GIF. I saved the screen capture in GIF since the text comes through better. JPEG causes artifacts in text but is fine for images. Convert to grayscale or RGB and you should be in business. The two screen veiws are of the exact same image. Photoshop display is antialiased.

If you want to look at the original image, it is from the DPReview test of the Canon 1DsM2.

No, I never did. I've never bothered copying dpreview test charts into Photoshop. You're a greater pixel peeper than I, Bill biggrin.gif
[a href=\"index.php?act=findpost&pid=70979\"][{POST_SNAPBACK}][/a]

You got me on that, Ray. Actually I downloaded the chart for analysis in Imitest, but I was intrugued by the alaising at 400% until I read Burce's article.

Bill

[Ray]

Bill,
Despite all the arguments in this thread, I'm still interested in a 24mp 5D successor with a 'real' ISO 3200 option and improved noise all round, if possible.  

[bjanes]

Bill,
Despite all the arguments in this thread, I'm still interested in a 24mp 5D successor with a 'real' ISO 3200 option and improved noise all round, if possible. 
[{POST_SNAPBACK}][/a]

Ray,

According to the analysis of Dr. Roger Clark, the current Canon sensors are reaching the theoretical limits; however, if the quantum efficiency of the sensor (currently about 25%) could be improved and a larger full well could be implemented, your 5D successor would be possible. A new generation of lenses to take advantage of this resolution would probably be needed.

[a href=\"http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html]http://www.clarkvision.com/imagedetail/eva...-1d2/index.html[/url]

Bill

[Ray]

According to the analysis of Dr. Roger Clark, the current Canon sensors are reaching the theoretical limits; however, if the quantum efficiency of the sensor (currently about 25%) could be improved and a larger full well could be implemented, your 5D successor would be possible. A new generation of lenses to take advantage of this resolution would probably be needed.
[a href=\"index.php?act=findpost&pid=71109\"][{POST_SNAPBACK}][/a]

Well, I don't know what to say. I've already taken lots of test shots with my cropped version of a 22mp 5D (my 20D) and have found it produces noticeably more detail than the 12.8mp 5D with even medium quality zooms such as the 100-400 IS. When I say noticeably more, I mean significantly more noticeable than the differences between the 5D and the 1Ds2 that we saw recently in another thread.

Roger Clark appears to be very qualified, but ultimately I have to believe my own eyes.

[Jack Flesher]

Well, I don't know what to say. I've already taken lots of test shots with my cropped version of a 22mp 5D (my 20D) and have found it produces noticeably more detail than the 12.8mp 5D with even medium quality zooms such as the 100-400 IS. When I say noticeably more, I mean significantly more noticeable than the differences between the 5D and the 1Ds2 that we saw recently in another thread.

So are you telling us your 20D is sharper than your 5D?

but ultimately I have to believe my own eyes.

Most of us do -- and unfortunately that creates these endless threads on forums: Folks trying to "prove" their point with words and when those don't work, jpeg posts and crops, and when those don't work because of jpeg artifacting, they go back to words about print resolutions and the process repeats itself. Ad-nausium.

[bjanes]

Ray,

According to the analysis of Dr. Roger Clark, the current Canon sensors are reaching the theoretical limits; however, if the quantum efficiency of the sensor (currently about 25%) could be improved and a larger full well could be implemented, your 5D successor would be possible. A new generation of lenses to take advantage of this resolution would probably be needed.

Bill
[{POST_SNAPBACK}][/a]

Well, I don't know what to say. I've already taken lots of test shots with my cropped version of a 22mp 5D (my 20D) and have found it produces noticeably more detail than the 12.8mp 5D with even medium quality zooms such as the 100-400 IS. When I say noticeably more, I mean significantly more noticeable than the differences between the 5D and the 1Ds2 that we saw recently in another thread.

Roger Clark appears to be very qualified, but ultimately I have to believe my own eyes.
[a href=\"index.php?act=findpost&pid=71122\"][{POST_SNAPBACK}][/a]

Ray,

So far as I know Roger has made no statements regarding details captured by the cropped 22 mp sensor versus the 12.8 mp with medium quality zooms. However, the smaller pixel size resulting from going to 22 mp does have disadvantages--you would be going from a pixel size of 8.2 microns to 6.4 microns. Noise characteristics and dynamic range would suffer unless improvements were made in the sensor technology:

[a href=\"http://www.clarkvision.com/imagedetail/does.pixel.size.matter/index.html]http://www.clarkvision.com/imagedetail/doe...tter/index.html[/url]

BTW, he does report some tests with the 100-400 IS zoom and some other lenses with teleconverters using the Canon 10D. He noted that the teleconverter did not degrade resolution, which would indicate that these lenses do have some reserve resolution not made use of by the sensor.

http://www.clarkvision.com/imagedetail/rel...lens-sharpness/

If you have plenty of light or a tripod the smaller pixels could give excellent results as evidenced by the Nikon D2X with 5.5 micron pixels. However, with hand held candid type situations, one often gets better results with larger pixels. Under these conditions, the full resolution of the 22 mp resolution can not be realized.

My comment about a new generation of lenses comes from Michael's frequent statements that the EOS 1Ds MII sensor out resolves the lens, for example:

http://www.luminous-landscape.com/reviews/...ital-view.shtml

[Ray]

So are you telling us your 20D is sharper than your 5D?
[a href=\"index.php?act=findpost&pid=71142\"][{POST_SNAPBACK}][/a]

With the same lens from the same distance, it appears to have greater resolution using a medium quality zoom. Of course, it's understood when comparing images of equal FoV that 12.8mp produces more detailed results than 8mp just as the 16mp of the 1Ds2 produces (very slightly) more detailed results than the 12.8mp of the 5D.

I haven't got around to comparing results using a better lens, such as my TS-E 90 because, it would be obviously silly to use a 90mm lens on the 20D instead of a 140mm lens on the 5D.

[Ray]

So far as I know Roger has made no statements regarding details captured by the cropped 22 mp sensor versus the 12.8 mp with medium quality zooms. However, the smaller pixel size resulting from going to 22 mp does have disadvantages--you would be going from a pixel size of 8.2 microns to 6.4 microns. Noise characteristics and dynamic range would suffer unless improvements were made in the sensor technology:

The concept that larger pixels produce less noise and can capture greater dynamic range has been repeated many times on this forum and others, but the facts tend to be slightly different. Sensor design and in-camera processing seem to be a more significant factor. Canon's history in DSLR development actually demonstrates the reverse of this concept. On every single occasion that I can recall (and there aren't many of course), Canon has delivered lower noise each time they reduced pixel size.

Quite extraordinary, isn't it? The D60 had twice as many pixels on the same size sensor as the D30, yet still managed lower noise than the D30. The 1Ds had larger pixels than the D60 yet hardly more dynamic range and actually slightly more noise on a pixel for pixel basis. The 20D with the smallest pixels of all Canon's DSLRs, shares the feature of lowest noise of all with cameras with larger pixels such as the 5D.

As I see it, if Canon were able to design a FF sensor using the same size pixels as the 20D, and all they did was to maintain the noise characteristics of each current 20D pixel, then they would have achieved an over all reduction in noise compared with the 5D on the basis that 2 pixels are better than one. This was the case when comparing the D60 with the 1Ds. The D60 photodiodes, despite being smaller than those of the 1Ds, were slightly less noisy. However, 1Ds images, viewed at the same size, were significantly less noisy than D60 images.

[Ray]

Folks trying to "prove" their point with words and when those don't work, jpeg posts and crops, and when those don't work because of jpeg artifacting, they go back to words about print resolutions and the process repeats itself. Ad-nausium.
[a href=\"index.php?act=findpost&pid=71142\"][{POST_SNAPBACK}][/a]

Am I making you feel nauseous, Jack?    Sorry!  

Just for the record, all the jpeg images I post do not have visible artifacting viewed at the magnification I post (ie 100% of whatever crop I'm posting, whether the crop before jpeg compression is a 100% crop of a TIF or a 400% crop of a TIF.) I use the maximum quality of jpeg compression which, to my eyes, is virtually lossless, but obviously not lossless at a machine level.

The argument that subtle differences cannot be transmitted over the net because of jpeg artifacting is a bogus argument in my opinion. It's just a matter of getting the magnification and crop right before compressing at maximum quality. Of course, it's not practical to transmit a 36MB image at maximum jpeg quality because that's still something like a 5 to 7MB image which would severely tax Michael's server as well as many folks' dial-up connection.

The reasons why we have these endless debates are as follows, in the order that springs to mind.

(1) Lens variability amongst different copies of the same model. It can be solved by the issuing of sets of MTF charts with each lens that's sold.

(2) Lack of proper and rigorous shooting technique. Ie., inappropriate choice of subject material for the point that is being demonstrated; sloppy focussing, whether due the camera's inadequate autofocussing system or due to the difficulty of accurate manual focussing because of a small viewfinder and/or poor eyesight on the part of the photographer; lack of a sturdy tripod, remote cord, MLU; sneaky sources of unnoticed vibration such as wind, heavy trucks trundling by, minor earth tremors etc etc. The list could be long.

(3) Complete ignorance of the scientific method.

(4) A common psychological reluctance to solve the problem once and for all because we enjoy the mess we're in and are afraid of unfamiliar territory.

[BJL]

On every single occasion that I can recall (and there aren't many of course), Canon has delivered lower noise each time they reduced pixel size.
[a href=\"index.php?act=findpost&pid=71210\"][{POST_SNAPBACK}][/a]

True, and probably mostly due to progress in reducing read-out/pre-amplifier noise. But at some stage, this will reduce those noise sources below the level of other sources, like sensor dark-current and photon shot noise and those other sources will become dominant. So perhaps sensors are getting close to a theoretical upper limit, and further improvements will be less than in the past.

On the other hand, I see no major problems (besides cost and maybe frame rate) with putting Canon's 6.4 micron pixels into a larger sensor, or even using something like Sony and Panasonic's 5.5 micron pixel spacing:

- Noise and DR would still be fine for a wide range of shots, particularly at low to moderate ISO, as the 20D, D2X and E-330 demonstrate

- In cases where higher S/N ratio or DR is needed, downsampling back to a "mere" 16.5MP or even 12.7MP would roughly restore the S/N and DR of using sensors with those lower pixels counts. In fact, simply keeping one's print sizes within the resolution limits of those 12.7 or 16.5MP sensors would, due to the higher PPI used, given roughly the same visible noise levels: the miracles of "dithering".

The only reason that downsampling or same sized printing would given worse noise levels would be read-out/pre-amplifier noise, as dark current and photon shot noise are roughly constant per unit area of sensor, regardless of how many photosites are used to gather the light from that area. And this "read" noise seems to be getting to be a rather minor factor in overall DSLR sensor noise levels, and exactly where Canon (along with all DSLR sensor makers) is making its main progress.

[bjanes]

The concept that larger pixels produce less noise and can capture greater dynamic range has been repeated many times on this forum and others, but the facts tend to be slightly different. Sensor design and in-camera processing seem to be a more significant factor. Canon's history in DSLR development actually demonstrates the reverse of this concept. On every single occasion that I can recall (and there aren't many of course), Canon has delivered lower noise each time they reduced pixel size.

[{POST_SNAPBACK}][/a]

Ray,

I would suggest you study Dr.  Clark's analysis in more detail and learn the basics of photon noise and read noise, and how the camera gain affects read noise when the 16 bit data number is output. The effect of read noise is greatly increased with small pixels because the small pixels have a low gain factor (gain = electrons/ADU data number).

The improved noise characteristics in the newer Canon sensors is related largely to improvements in read noise. These same improvements can be applied to sensors with large pixels. For the same read noise, the effect on the noise expressed in 16 bit data numbers will be less with the large pixel because of the increased gain. Photon noise is directly related to pixel size unless the laws of physics can be changed. The large pixels have the advantage for both photon noise and read noise.

[a href=\"http://www.clarkvision.com/imagedetail/does.pixel.size.matter/index.htm]http://www.clarkvision.com/imagedetail/doe...atter/index.htm[/url]

Limits are being reached with the newer sensors and continued improvements will be more difficult than in the past. When Canon introduced the 30D they could have upped the pixel count to match the Nikon D200, but they chose to keep the same pixel size in order to maintain the noise levels at Canon standards.

Bill

[Ray]

I would suggest you study Dr.  Clark's analysis in more detail and learn the basics of photon noise and read noise, and how the camera gain affects read noise when the 16 bit data number is output. The effect of read noise is greatly increased with small pixels because the small pixels have a low gain factor (gain = electrons/ADU data number).
[a href=\"index.php?act=findpost&pid=71259\"][{POST_SNAPBACK}][/a]

Bill,
You mean I should study Roger Clark's analysis in more detail so I can understand that what already exists in reality cannot possibly exist?  

I have read the article you refer to and I also have a general understanding of the various sources of noise that sensors are subject to. The principles are interesting and it might well be true that no further S/N improvements are possible (in Canon sensors) without improvements in quantum efficiency. I'm certainly not qualified to argue whether or not that's true, but if it is true, then so be it. I'm sure we'll get sensors with improved QE eventually which will enable even lower noise levels than in current Canon DSLRs.

All I'm doing here is making a general point based upon photograph evidence from real images taken with various cameras with different sized pixels. There's no direct correlation between pixel size and noise in Canon DSLR models. That does not mean that Roger Clark is wrong. The principles he espouses might well be quite sound. What it probably means is that there are other technologies at work that Roger has not addressed.

Take for example the recent comparisons, in another thread, between 5D and 1Ds2 images, in search of resolution differences. It seems to be established that the 1Ds2 does indeed deliver very marginally more detail, and this is confirmed in the dpreview comparisons. What is interesting is what happens at ISO 1600, something that no-one addressed in that thread.

According to dpreview, the 5D images have noticeably less noise than the 1Ds2, at ISO 1600. Is this due to the larger pixel of the 5D? One might think so. On the other hand, the 1Ds2 images at this high ISO are significantly more detailed than the 5D images. If one is searching for meaningful resolution differences between the two cameras, they are to be found at ISO 1600.

The question that now arises is this. Is the lower noise of the 5D at ISO 1600 due to its larger pixel size or due to in-camera noise reduction? Perhaps the answer is, a bit of both. However, if no additional noise reduction or cancellation has been applied to the 5D image, why has resolution suffered so much in relation to the 1Ds2?

[bjanes]

Bill,
You mean I should study Roger Clark's analysis in more detail so I can understand that what already exists in reality cannot possibly exist?   
[a href=\"index.php?act=findpost&pid=71338\"][{POST_SNAPBACK}][/a]

An excellent comeback.    You are correct that thus far, the cropped frame Canon sensors have increased resolution while maintaining low noise, but has the limit mostly been reached with the 30D? Roger seems to think so, but he does state that there is still room for increased quantum efficiency

According to dpreview, the 5D images have noticeably less noise than the 1Ds2, at ISO 1600. Is this due to the larger pixel of the 5D? One might think so. On the other hand, the 1Ds2 images at this high ISO are significantly more detailed than the 5D images. If one is searching for meaningful resolution differences between the two cameras, they are to be found at ISO 1600.

The question that now arises is this. Is the lower noise of the 5D at ISO 1600 due to its larger pixel size or due to in-camera noise reduction? Perhaps the answer is, a bit of both. However, if no additional noise reduction or cancellation has been applied to the 5D image, why has resolution suffered so much in relation to the 1Ds2?
[a href=\"index.php?act=findpost&pid=71338\"][{POST_SNAPBACK}][/a]

Canon claims that their NR does not affect image detail. I do not understand their NR, but it sounds a bit like double correlated sampling, which reduces read noise. I used Roger's methods to assess the noise characteristics of my Nikon D200 and found it has a good full well capacity and photon noise characteristics for its pixel size, but that its read noise is considerably higher than Canon's. Noise is worse than with the D70 and I relate this to pixel size. Poor high ISO noise characteristics are my least favorite feature of this otherwise excellent camera and this has sensitized me to the importance of pixel size.

Without testing it is difficult to answer your question about resolution and noise between the 5D and 1Ds2, but pixel size seems a reasonable explanation to me. Since you are fond of extensive tests, why don't use use Roger's test methodology on your 5D and report the results?

Bill

[Ray]

Since you are fond of extensive tests, why don't use use Roger's test methodology on your 5D and report the results?
[a href=\"index.php?act=findpost&pid=71341\"][{POST_SNAPBACK}][/a]

Why? Because it's not going to help me take better photos. All my testing of lenses at different apertures, comparing one lens with another and one camera body with another, is for the purpose of helping me make a more informed decision in the field when I'm shooting. When I test lenses in the store before buying, it's to help me get the most bang for my buck.

Testing out of pure curiosity is something I rarely do. For example, I have thought about comparing the 5D and 20D with my sharpest lens, the Canon 50/1.8, but I have no reason to waste time doing this because I have no reason to believe I would ever choose to use a 20D with the 50/1.8 instead of a 5D with an 80mm lens. On the other hand, if I ever decide to get a Canon 50/1.4, I'd want to test that against my 50/1.8 to make sure I was not wasting my money and getting a copy that had slipped through the quality control net.

[bjanes]

Why? Because it's not going to help me take better photos.
[a href=\"index.php?act=findpost&pid=71367\"][{POST_SNAPBACK}][/a]

It might not enable you to take better photos, but it would most likely increase your understanding of the issues involved. At least that was what I found. Each time I go over the results, perhaps in preparation for posting online, I learn something new.

Bill

[Ray]

It might not enable you to take better photos, but it would most likely increase your understanding of the issues involved. At least that was what I found. Each time I go over the results, perhaps in preparation for posting online, I learn something new.
[a href=\"index.php?act=findpost&pid=71376\"][{POST_SNAPBACK}][/a]

 

That may well be true. It's really a matter of priorities. Should I be stuffing around trying to confirm that read noise reduces with increasing ISO when I have advanced Photoshop tutorials bought in Bangkok 8 months ago which I haven't yet studied,  dozens of unread books on my shelves,  thousands of digital images that need to be catalogued and thousands of slides, negatives and even odd sizes of B&W plates that need to be scanned before further deterioration takes place.

[Jack Flesher]

Ray and Bill:

Just a simple question I am curious about:

In your mind, consider the top three images you have personally ever produced.  What equipment was used for each and how were they processed and displayed?

I recently did this for myself and was surprised by my answers.

[Ray]

In your mind, consider the top three images you have personally ever produced.  What equipment was used for each and how were they processed and displayed?
[a href=\"index.php?act=findpost&pid=71390\"][{POST_SNAPBACK}][/a]

Jack,
No surprises for me. From thousands of images, it's very difficult to select just 3 top images, but 3 of the several images currently hanging on my walls could be considered the 3 top images by virtue of the fact they are on my wall.

One was taken with my 5D and 2 with my first digital camera, the D60. 2 were converted from RAW in ACR and one in RSP. Further processing was done in Photoshop and they were all printed on my Epson 7600 with K2 pigment inks.

Where's the mysterious significance?

[Jack Flesher]

2 with my first digital camera, the D60.
SNIP
Where's the mysterious significance?
[a href=\"index.php?act=findpost&pid=71397\"][{POST_SNAPBACK}][/a]

The significance is that obviously pixel peeping -- analysing the technological aspects of our gear: lens resolution, diffraction limitation, dynamic range, color accuracy, noise/grain -- seem to have little to do with producing our favorite (and ergo by inference, best) images

I should have also asked when you started doing photography seriously...

Cheers,