haijak wrote:
Do you have any other sources or documented tests to support this Hypothesis? Everyone who shares this idea, cites the exact same document.
The rest of us can go to almost every other photography site in the world to support our theory.
(Emphasis added) haijak, you may not realize this, but Richard ("brainiac") "cites" this document because... he wrote it.
To me the argument seems like one of those things that's so obvious that nobody bothers to mention it. It seems very obvious to me that with higher resolutions, noise is less noticeable when viewing the image as a whole. However if there is less noise at 100%, the full image will be even cleaner, even if you can see it. You're argument is that since you won't notice it, the idea is wrong. That just not necessarily correct.
I think it is clear that we would all appreciate less noise (that is, better signal-to-noise) at 100%, if lower noise does not come at the expense of detail. Achieving that lower noise depends on either {a} better technology, to reduce artificially added (e.g., "read" noise), and/or {b} defying the quantum nature of light (which results in "shot" noise).
Since {b} does not appear to be an option, we are left with {a}. One of the things that Richard and like-minded people are pointing out is that the higher-pixel-density camera may not be doing any worse than the lower-pixel-density camera in the read-noise department, but because of the finer pixel pitch, the quantum shot noise is more apparent. What that means is that the higher-pixel-density camera is doing a better job of recording physical reality, which at finer and finer scales is noisier (but when compared on a level playing field [equal sensor areas], the higher-pixel-density camera presents a better image). Another potential scenario is that the higher-pixel-density camera could be noisier (artificial read noise) at the pixel level, but because of the much higher number of pixels, the net effect of the read noise may be lower per sensor area, when compared with a lower-pixel-density camera.
paulfeng wrote:
(Emphasis added) haijak, you may not realize this, but Richard ("brainiac") "cites" this document because... he wrote it.
Yes - and unfortunately I felt I had to write it because confusion about this issue is rife. Correct comparison between two cameras to see which is noisier is done by comparing equal proportions of the overall picture. It's surprisingly hard to find that kind of comparison online, and it's amazing how many people have strong opinions based on false comparisons. I come across erroneous analysis so often online that I decided to write an article to cite rather than keep spelling out the problem on thread after thread. Luckily there are one or two others, like John Sheehy, who also get this. Just because nearly everyone's wrong, that doesn't mean you have to agree with them. For low light, pixel density is not the issue that most people seem to think it is.
brainiac wrote:
It's surprisingly hard to find that kind of comparison online, and it's amazing how many people have strong opinions based on false comparisons.
I had the same argument with the people from WhatDigitalCamera.com and they ended up saying "so dpreview and us are wrong and you are right"..
paulfeng wrote:
I think it is clear that we would all appreciate less noise (that is, better signal-to-noise) at 100%, if lower noise does not come at the expense of detail.
The same can be said of resolution - it’s fine to have extra megapixels if they do not come at the expense of extra noise.
It seems that some people are not fully comprehending the relationship between resolution and noise.
I guess that’s because reviewers make resolution tests on black& white charts at ISO 100 and noise tests on color patches at various ISOs.
For this and other reasons, it’s very easy to assume that resolution stays constant at any ISO and somehow just noise increases at higher ISOs. Therefore, if you clean up the extra noise – voila, you have a clean image with lots of detail in it.
But let's recap what noise is - it’s the collection of artifacts in an image created by wrong sensel readings.
A gray dot in the frame, for example, is read by the sensor (and recorded in the image) as red, and a blue dot is read as black, for example.
In other words, the sensor is replacing the actual luminance and color values in the frame with incorrect/artificial values – thus replacing the actual detail in the frame with fake color artifacts.
So, if Sensor A has more pixels but is also noisier than Sensor B, its resolution is compromised.
Sensor noise (not image noise) can be judged only at 100% magnification.
If we compare image noise, as some suggest, then it’s also fair to compare resolution on rescaled images as well.
What the proponents of the image noise theory are doing right now is that they quote the resolution figures measured at 100% magnification but then insist on comparing noise on rescaled images.
That’s double standard.
We either compare both noise & resolution at 100% magnification, or we compare both noise & resolution at rescaled images.
Thus, if we compare the 40D vs 7D for example, we either compare rescaled images and the figure 18mp is not mentioned anywhere in the comparison.
Or, we compare at 100% and take the 8mp difference into account.
You can't have it bothways - 18mp resolution and better image noise (at least not with the 7D, that is).
brainiac:
Yes - and unfortunately I felt I had to write it because confusion about this issue is rife. Correct comparison between two cameras to see which is noisier is done by comparing equal proportions of the overall picture.
I think that part of a problem when looking at the resolution and noise is that in your approach, these are two components that contribute to the overall image (preferably printed).
Some, including me, would prefer to see these two (noise and resolution) as two separated parameters. We could say that if "pixel noise" (or the noise of the same number of pixels) is the same, then camera A (with 30 Mipxels) would have higher resolution with detail proportionally higher that camera B (21 Mpixels). Otherwise the comparison has to include looking at "image equivalents"
By the extension, I can just throw away (crop) some pixels and know what to expect as far as overall noise is concerned.
I am actually having problems seeing why 30 Million of noisier pictures would give less noise than 21 Million of much "cleaner" pixels, if there is no other noise contributions.
jorkata wrote:
You can't have it bothways - 18mp resolution and better image noise (at least not with the 7D, that is).
You can have it both ways: in bright conditions set your camera to iso 100, and in dark conditions set your camera to iso 6400. Of course, the 6400 image will contain less detail, but I think most photographers know that.
mfurman wrote:
I think that part of a problem when looking at the resolution and noise is that in your approach these are two components that contribute the overall image (preferably printed).
Some, including me, would prefer to see these two (noise and resolution) as two separated parameters. We could say that if "pixel noise" (or the noise of the same number of pixels) is the same then camera A (with 30 Mipxels) would have higher resolution with detail proportionally higher that camera B (21 Mpixels). Otherwise the comparison has to include looking at "image equivalents"
Comparison of detail is no different to comparison of noise: it is best done by shooting the same shot and viewing at equal magnification.
The easiest way to view at equal magnification for comparison is to uprez the smaller file to the same number of pixels as the larger. Then 100% on screen comparisons mean something, and you can easily see which camera captures more detail, and which camera suffers more from noise.
I spent 2 hours trying to MA my 7D with 3 of my lenses. The 100 macro is sharp without any adjustment. My 17-55 is almost there with no adjustment. My 70-200 f/2.8 is way off, and my Sigma 500 was off too on the 7D. So, I'm sending it in .. I decided to keep it and not return it to Amazon. But, I want Canon to fix the AF assembly if that's indeed what's causing the missed focus. If not, no harm done. I'm also sending in my 70-200 (out of warranty).
brainiac wrote:
The easiest way to view at equal magnification for comparison is to uprez the smaller file to the same number of pixels as the larger.
Resizing is an intrinsically incorrect way to test, since you are no longer looking at the pixels as produced by the sensor.
Instead, you are looking at the pixels produced by the resizing algorithm.
I was just giving it as an example of what would be a fair comparison for those who think that noise should be evaluated using resized images.
What reviewers should start doing is test resolutions using color charts at different ISOs - in addition to their B&W resolution tests at ISO 100.
A resolution chart with gray background and red/green lines will easily show the resolution degradation caused by noise.
I predict that the results of such tests will be dramatic and a lot of people will start questioning the usefulness of cramming more and more megapixels into sensors.
saaketham wrote:
I spent 2 hours trying to MA my 7D with 3 of my lenses. The 100 macro is sharp without any adjustment. My 17-55 is almost there with no adjustment. My 70-200 f/2.8 is way off, and my Sigma 500 was off too on the 7D. So, I'm sending it in .. I decided to keep it and not return it to Amazon. But, I want Canon to fix the AF assembly if that's indeed what's causing the missed focus. If not, no harm done. I'm also sending in my 70-200 (out of warranty).
That is a good thing to do. Then you know for sure.
jorkata wrote:
Resizing is an intrinsically incorrect way to test, since you are no longer looking at the pixels as produced by the sensor.
Instead, you are looking at the pixels produced by the resizing algorithm.
I was just giving it as an example of what would be a fair comparison for those who think that noise should be evaluated using resized images.
A fair comparison would be to compare equal prints from both cameras, say, 16 X 24 inch and see which one shows the least noise and the best detail. But prints are hard to compare online. So, the next best thing it to resize one or the other to match the output of it's competitor so that you are comparing the equivalent of equal-sized prints.
Tom_W wrote:
A fair comparison would be to compare equal prints from both cameras, say, 16 X 24 inch and see which one shows the least noise and the best detail. But prints are hard to compare online. So, the next best thing it to resize one or the other to match the output of it's competitor so that you are comparing the equivalent of equal-sized prints.
Jorkata just joined this forum several months ago. So far, ALL his posts so far are related to bashing the 7D. Tells you something, don't you think?
thw2 wrote:
Jorkata just joined this forum several months ago. So far, ALL his posts so far are related to bashing the 7D. Tells you something, don't you think?
jorkata wrote:
Resizing is an intrinsically incorrect way to test, since you are no longer looking at the pixels as produced by the sensor.
Instead, you are looking at the pixels produced by the resizing algorithm.
I was just giving it as an example of what would be a fair comparison for those who think that noise should be evaluated using resized images.
What reviewers should start doing is test resolutions using color charts at different ISOs - in addition to their B&W resolution tests at ISO 100.
A resolution chart with gray background and red/green lines will easily show the resolution degradation caused by noise.
I predict that the results of such tests will be dramatic and a lot of people will start questioning the usefulness of cramming more and more megapixels into sensors. ...Show more →
viewing by resizing down won't ever give a bonus to the lower MP cam though
and sure the noise eats away signal and maybe at some point it eats away so much that even the lower MP cam can't get any signal at the grain of the sensor but so what? If the higher MP count hurts less than 1/3 stop then so what? Not that noticeable, if even, and you can get a ton more detail at ISO100 and a lot more reach.
and the lower MP never resolved those fine lines to begin with
