I'd doubt the A/D itself is the culprit here.
Looks like there is some requantization going on there. Maybe due to multiplications.
Have you done the same test with ISO 125 and 160?
There is an obvious cross hatching pattern to the 5DII sample which is almost totaly absent in the 40D sample. The random noise can be dealt with quite effectivley in post production noise reduction tools like NeatImage whereas the cross hatching pattern is impossible to remove. This is the problem - not the noise so much but the patterned structure.
MichaD wrote:
Wow, that 5DII histogram looks downright scary.
I'd doubt the A/D itself is the culprit here.
Looks like there is some requantization going on there. Maybe due to multiplications.
Have you done the same test with ISO 125 and 160?
I would be happy to get an explanation.
Haven't done any testing on other ISO's, and for now, I only have the ISO100 and ISO200 black frames.
I would be interested to see what the saturation level is at ISO160.
alundeb wrote:
I would be happy to get an explanation.
Haven't done any testing on other ISO's, and for now, I only have the ISO100 and ISO200 black frames.
I would be interested to see what the saturation level is at ISO160.
This sort of glitch in the black frame histogram seems to be common in the mark 3's as well; see
Ludo Monchat wrote:
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 ??
I think it's a flare-related issue. I've only had it show up in one or two images with very strong backlighting. My guess is that the pattern is reflections from the AA filter or some other filter in front of the sensor.
It would seem from reading more on that website about noise analysis of Canon raw dark frames that the reason for the steps is due to some manipulation of the RAW data in-camera to adjust images shot on fast lenses where peripheral illumination needs adjusting.
stanj wrote:
Pretty hard to do with a moving subject (not that the sample picture was such one), or when you're shooting PJ style and can't set up a tripod and ND grad. As said so often, just because you have the time to setup a tripod and grad filter doesn't mean everyone has the same luxury. Plus, in the above picture, please show me how you'd position an ND grad.
In PJ style cases a flash could be used with some FEC to bring the foreground and background exposures closer together. Not the same I know, but better than no pic at all.
And ND grads are pretty much unnecessary if you shoot multiple exposures and blend in post.
p.20 #10 · Are you seeing 5D2 Criss-Cross Pattern?
The histograms for separated R G B G channels in the 5D2 show the dip is almost 100% in one of them and very little in the others. I don't see why there would be requantization for only one channel other than for in-camera color sensitivity correction. That is certainly not going on.
If the A/D converter has local nonlinearities (missing codes if these are more than 1 ADU), some values will be underrepresented and the neighbours overrepresented.
The 1D3 plots show that the effect diminishes above unity gain ISO. That leads me to think that if the A/D's dont' feature sample-and-hold circuits, then conversion can then be affected by the increased noise at higher ISO's, blurring the nonlinearities.
This is just some thoughts, I don't even know how the readout channels are organized.
Good link, and funny that you mention it. The shadow pushed 5D2 ISO 100 image shows a lot LESS pattern in the noise than some of the real world images this thread started with.
Apr 09, 2009 at 02:28 AM
Guest
Guest
p.20 #13 · Are you seeing 5D2 Criss-Cross Pattern?
Earlier I thought 5D2 suffers from patterns only above 6400, but after seeing them at 100 in this thread... well..
What are you guys supposed to do? Will you just leave it as is?
p.20 #15 · Are you seeing 5D2 Criss-Cross Pattern?
Photomatt wrote:
It's just noise. My 20D did the same. If it were 'just noise' then it could be more or less easily dealt with by post production noise reduction software but it's not 'just noise' as it has a pattern which makes dealing with it very difficult. It would seem that by trying to get all the features and specifications into a very competativley priced body that corners have been cut in the signal transfer and processing chain. What has caused all this was Canon's statement that the 5DII had the best image quality in it's range and that included the 1DsIII. While Canon may have been guilty of being economical with the truth I think we can all be guilty of ignoring the golden guidance when dealing with such statements from where ever they come - 'if it sounds too good to be true then it is is too good to be true'.
p.20 #17 · Are you seeing 5D2 Criss-Cross Pattern?
ejmartin wrote:
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....Show more →
I actually did end up getting complete blackframe plus the colorchecker charts (under even sunlight striking at an angle to avoid specular reflections and with the normal and 3 stops under/over) for 20D,40D,50D,5DMkII before I sold all of the APS-C bodies off.
I may finally have time to get them together next week and post them to the server.
