Here is an informal rank of "Canon ISO 100 black frame noise appearance"
Based on images posted here, processed according to the "agreement" and resized.
A space in the list indicates a significant difference. Best one on top, just in case you wonder
Interesting though this test is it does not tell us that much as it's basicaly where the noise floor is but it tells us nothing about how far away this is from the clipping point/sensor saturation and this is what is realy important - how many stops between the highest value under clipping and the lowest value where this pattern noise becomes objectionable.
but with everyone using different colorspaces and who knows which downsize algorithm different people were using, etc.....
5DMkII
anyway below are ISO100, 1/60th, set DPP to aRGB, set brightness to 2, faitful profile, set it to daylight WB, set saturation to 0 (note 0 is NOT min), sharpness 0 (zero IS min here), turned off NR and all that, saved as 16bit TIFF, opened in CS4 in aRGB colorspace since that seems what most people used, did the levels thing to 25, scaled to 900 across using Resample Bicubic. Saved as jpg at level 12 and keeping it saved in aRGB (viewing will be different depending if your browser handles profiles or not).
Here are measured DR of some Canon cameras at ISO 100, by different sources. Some sources publish DR only, some Read Noise only, and some both.
The saturation values are from Clarkvision [2] except 1Ds3 (van der Wolf [7] ) and are assumed to be constant for each camera. Saturation values and read noise are in electron units. Read noise in this context means all noise from the camera electronics.
The measurements seem fairly consistent, but with some variance.
What is particularly interesting, is the saturation level of 5D2 compared to 1Ds3.
Dividing the saturation electron count by the respective system gain, we get the RAW saturation values (actual values may vary slightly due to rounding errors):
5D2: 12375
1Ds3: 14250
This means that the 5D2 should have LOWER noise floor in the RAW file to match the DR of the 1Ds3.
p.19 #10 · Are you seeing 5D2 Criss-Cross Pattern?
David Baldwin wrote:
Absolutely no way the 40D has better noise control in practice than the 5D2. I own both.
I don't question that. Midtone noise is dominated by photon shot noise, and then the larger photosites of the 5D2 win. Also at higher ISO's there is no comparison.
This is the special case of low ISO, shadow pattern noise. If you still find the 5D2 better there, I have no problem with that. It should be.
p.19 #11 · Are you seeing 5D2 Criss-Cross Pattern?
shirozina wrote:
Interesting though this test is it does not tell us that much as it's basicaly where the noise floor is but it tells us nothing about how far away this is from the clipping point/sensor saturation and this is what is realy important - how many stops between the highest value under clipping and the lowest value where this pattern noise becomes objectionable.
I agree...
With the 1Ds3, the lenscap procedure shows little noise and no pattern when pushed in PS to 25. You have to got to about 10 for something to become visible. It would be interesting to know where the treshold lies for the 5D2. After all, we kind of randomly choose the level value of 25, in order to make the testing uniform.
p.19 #12 · Are you seeing 5D2 Criss-Cross Pattern?
skibum5 wrote:
anyway below are ISO100, 1/60th, set DPP to aRGB, set brightness to 2, faitful profile, set it to daylight WB, set saturation to 0 (note 0 is NOT min), sharpness 0 (zero IS min here), turned off NR and all that, saved as 16bit TIFF, opened in CS4 in aRGB colorspace since that seems what most people used, did the levels thing to 25, scaled to 900 across using Resample Bicubic. Saved as jpg at level 12 and keeping it saved in aRGB (viewing will be different depending if your browser handles profiles or not).
What FW does your 5D2 have installed?
Apr 06, 2009 at 02:09 AM
Ludo Monchat Offline • [X]
p.19 #13 · Are you seeing 5D2 Criss-Cross Pattern?
A quick question although I'm afraid I already know the answer. Is the below shot normal, I mean, can you consider a 5D2 RAW converted with LR2.3 "normal" when it shows such pattern ?? I have it since the day Canon did fix the cam for a complete lockup. They changed the whole electronic assy. Before that day the cam was perfect, smooth transitions and no wierd patterns. Seems that othe people is having the same problem. Some ideas ??
p.19 #14 · Are you seeing 5D2 Criss-Cross Pattern?
David Baldwin wrote:
Absolutely no way the 40D has better noise control in practice than the 5D2. I own both.
It does when pushing low ISO files. For noise control alone, I would take a 40D ISO 100 file pushed 2 stops over a 5D2 ISO 100 pushed 2 stops any day
p.19 #15 · Are you seeing 5D2 Criss-Cross Pattern?
Hi, I don't know if anybody is interested, but I analyzed the ISO 100 black frames from my friend's 5D2 and my 40D using IRIS and dcraw.
Actually, the 5D2 measures a lower DR than the 40D, using standard deviation.
Also made a plot of the RAW file shown as Color Filter Array. Absolutely no processing. Dark frame offset of 1024 not subtracted.
The ADU levels for the image were set to 1000 - 1100.
This is just to show the nature of the unprocessed pattern noise. The magnitude has no meaning compared to other images.
Note that these are unscaled crops. http://i306.photobucket.com/albums/nn242/Overflate/demo/CFA_5D2_40D_1000_1100.jpg
p.19 #17 · Are you seeing 5D2 Criss-Cross Pattern?
Daan B wrote:
alundeb... Even though I don't understand much from the procedures you used, I appreciate your efforts...
Are you saying that the 5D2 has NR applied (just like Nikon's latest RAW files) to the RAW data, even if everything is set to zero?
No, the NR is applied in the RAW conversion i DPP. And other RAW converters as well. I could try to export a converted image from IRIS, but it is difficult to apply the same tone curve for comparison. Not sure if I can at all.
Daan B wrote:
And are you saying that the 40D has a larger DR than the 5D2?
Only my friend's copy of the 5D2, and probably the "worst" ones here.
There is some variance among copies of cameras, shown both in measurements and images. I would expect the "best" copies of 5D2 to have better measured DR than the 40D, this is also supported by the measurements presented in the table in a previous post.
My understandig is as follows:
The difference in measured DR from RAW files between 5D2 and 40D is smaller than the diffrence in measured pattern noise in processed images. To be honest, I expected a closer relationship before making the analysis.
It seems that the pattern noise is amplified by the RAW conversion algorithms, that are designed to look for patterns. (brainiac, ejmartin and others knew that, but I was in doubt). This effect can make cameras with pattern noise that measure good dynamic range still appear worse than cameras with less DR and also less pattern noise.
The differences in measured DR among 5D2's come on top of that.
p.19 #18 · Are you seeing 5D2 Criss-Cross Pattern?
I would add that pixel level DR is only part of the story, in comparing cameras with different pixel counts, because the pixel is sampling a different scale in the image. Comparing pixel level DR in a 21MP camera to that of a 10MP camera is comparing the DR at scales about 1.4x smaller relative to image size in the 21MP camera. Without going through the details, the 21MP camera's pixel DR should be multiplied by 1.4 (a half a stop) to compare its image DR to that of the 10MP camera. This compensation is what DxO does for instance in generating its "print" figures of merit (normalizing to an 8MP reference scale), while its "screen" figures of merit are pixel-based quantities.
p.19 #19 · Are you seeing 5D2 Criss-Cross Pattern?
I'm not sure if this has already been noted, but I don't believe the results from shooting at ISO 100 and "pushing" 4 stops in the RAW converter are the same as shooting at ISO 1600. I think if you do the former, some weirdness is to be expected.
p.19 #20 · Are you seeing 5D2 Criss-Cross Pattern?
ejmartin, are you planning any measurements on the 5D2 yourself?
It wasn't my intention to compare the 5D2 to the 40D, it was done only because I don't have a 1Ds3 available, and it turned out that the 40D has very little pattern noise, making it suitable for comparison at pixel level. The saturation levels in ADU of the 5D2 and 40D are also very close, making histogram and CFA plot comparisons easy.
Still, the resized 40D black frame looks "better" than the resized 5D2 black frame, because the pixel count effect doesn't help on pattern noise. But that visual betterness is mostly because the image is balck, and in a real image you would see more shadow detail from the 5D2.
