 |
brainiac
Offline
[X]
|
Yakim Peled wrote:
I guess we must all decide which and whom to follow….. 
I don't want anyone to "follow me", but I don't think it helps just to ask 'how can all these people be wrong?'.
They are wrong, and one difference between them and me is that I have provided detailed explanations of the problem, with examples, on many threads here. Meanwhile, those sites refuse even to acknowledge that they are making an error which misleads their readerships.
But I'll try again. Take one image. Print it at 10x15 and at 40x60. Now cut a square inch out of each print, take the two square inches, and walk up to someone on the street. Ask her which image suffers more from noise. The more enlarged crop will lose every time. Every time. The more enlarged print clearly has worse noise than the less enlarged one. How can that be, considering they were made from the same file?
Now imagine a 100x100 pixel 100% crop from a 21 megapixel file, and a 100x100 pixel 100% crop from a 12 megapixel file on your monitor. If we imagine that you monitor has 100 dots per inch, then you are looking at excerpts from a 56 inch wide image, and a 43 inch wide image. Because each crop is at 100%, and the pixels are the same size, the 21 megapixel file is considerably more enlarged. That means that you are handicapping the 21 megapixel file in the noise comparison because you are comparing PER PIXEL noise between the files, when the 21 megapixel file can actually afford to have noisier pixels in the final print, since each pixel is a smaller constituent part of the image. Noise averages out that's why it seems to go away when you downrez.
Sites like LL and DPR all conclude that the D700 is a much better camera for shooting at 12800 ISO than the 1Ds3. They suggest that while the 1Ds3 is quite good at 3200 ISO, it wouldn't compete with the D700 at higher ISOs. They make this error because they examine the noise patterns in 100% crops from both cameras. If they bothered to do a fair comparison, unlike the badly biased comparisons that they do, they would find that the 1Ds3 and D3/700 offer quite similar capabilities at ISO 12800, and that the 1Ds3 is better than the D3/700 at ISO 3200. My crops to support this are here, all 12800 iso: http://cyberphotographer.com/d700v1ds3/
What is more, if you resize the DPR noise comparison crops to equal magnification, you find that they tell the same story. DPR has good evidence right under its nose, but is not capable of interpreting the outcome correctly because it is making the basic error of comparing crops at differing magnifications, i.e. 100%. I put this to you as a matter of fact, as I have done before many times. The noise comparison pictures of a postage stamp at DPR show the Queen's head at differing magnifications on screen, as do the bottle label/oak tree crops. Here are some DPR examples resized to equal magnification, which contradict the verdict given in the DPR review:
DPR data resized to equal magnification:
ISO 3200 1Ds3, 1Ds3, D3
http://cyberphotographer.com/1ds3/dpr_D3v1Ds3_iso3200.jpg
ISO 3200 1Ds3, D3 (resized to 1Ds3 magnification), 1Ds3 (with chroma NR)
http://cyberphotographer.com/1ds3/dpr1ds3noisecrop.jpg
If you can't see why that shows that it's wrong to compare pixel noise in 100% crops at differing physical image sizes, then I don't think I can help to disillusion you. Feel free to continue enjoying the enlightenment available at LL, DPR, and RG.
Edited by brainiac on Sep 10, 2008 at 12:46 PM GMT
Edited on Sep 10, 2008 at 07:46 AM
|
| Sep 10, 2008 at 05:37 AM |
| |
|
 |
