5D3 Pattern Noise Improvement and High ISO Cooking

Started Apr 1, 2012 | Discussions
bclaff Veteran Member • Posts: 7,308
5D3 Pattern Noise Improvement and High ISO Cooking
10

Certain types of pattern noise and signal processing can be detected using 2D FFTs (Fast Fourier Transforms).

I performed such an analysis of dark frames (raw) from a 5D2 and a 5D3 using ImageJ.
Here's the 5D2

and the 5D3

The ISOs from left to right are ISO 50, ISO 1600, ISO 6400, ISO 25600, and ISO 102400 (5D3).
The images have been darkened to accentuate the patterns.

Sometimes I find it helpful to view the images from several feet to see subtle features.
The "ideal" FFT would be a uniform gray.
A good guide to interpreting FFTs is here:
http://qsimaging.com/ccd_noise_interpret_ffts.html

Stong vertical or horizontal lines general indicate pattern noise in the horizontal or vertical direction respectively.
These lines are stronger for the 5D2. It has more pattern noise than the 5D3.

Little "ticks" along an axis is usually a sign of banding, often due to unbalanced channels in a multi-channel readout scheme.
This is present in both cameras to some extent.

The 5D3 ISO 102400 has the classic "spherical" look of signal processing. There is no question that this ISO is "cooked".

The 5D3 also look suspicious to my practiced eye in that the FFTs change from a horizontal line to a vertical line from ISO 50 to ISO 102400. I can barely make out a diamond shape in the ISO 6400 image. Such a transition could only come from signal processing. (Obviously subtle compared to ISO 102400!)

-- hide signature --
Canon EOS 5D Mark II Canon EOS 5D Mark III
If you believe there are incorrect tags, please send us this post using our feedback form.
Peter 13 Veteran Member • Posts: 8,301
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

Good analysis. I am not so sure about the transition from horizontal to vertical - I am not very familiar with the architecture of the chip. But the last picture shows NR, without doubt. On the other hand, at this ISO - who cares.

tigertails99 Contributing Member • Posts: 544
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

Interesting, can you show same analysis for D800 ?

stephenmelvin Veteran Member • Posts: 4,547
so what does it all mean?

How does ISO 50000 look?

It seems to me that the processing means that ISO 100000 looks different than a file shot at ISO 25K and pushed two stops.

The question is whether this is a good thing or a bad thing.

Overall, the image quality of the new camera at higher ISO's is unquestionably nicer than that of the Mk II. They're usable, which isn't something you could always say about the older camera.

What I find interesting is how Canon's in-camera NR is terrible looking. This signal processing is done in such a way as to look a lot more organic.

Have you noticed such processing in other cameras?

You always post such interesting threads!

-- hide signature --
John Sheehy Forum Pro • Posts: 20,465
Re: 5D3 Pattern Noise Improvement and High ISO Cooking
1

bclaff wrote:

The 5D3 also look suspicious to my practiced eye in that the FFTs change from a horizontal line to a vertical line from ISO 50 to ISO 102400. I can barely make out a diamond shape in the ISO 6400 image. Such a transition could only come from signal processing. (Obviously subtle compared to ISO 102400!)

Nothing strange about that at all. Some forms of banding are proportional to amplification in strength, and some are post-gain, and therefore fixed across ISOs The vertical banding in the 5D3 is mainly post-gain, and becomes insignificant at high ISOs as the gain-related noise grows relatively stronger.

-- hide signature --

John

David Hull
David Hull Veteran Member • Posts: 6,117
Well, that was quick... ;-)

Is there any way you can quantify the 5DII to 5DIII improvement in terms of EV or dB or something like that. This seems to confirm what I have been guessing just fiddling with the dark frames which is that they have improved the noise between the two of them.

Also, I don't really understand the significance of the "cooking". I expect all manufacturers do whatever makes sense whether through HW or SW to optimize their system performance. I certainly do that in the stuff I design, I am not sure what the problem is with that exactly.

 David Hull's gear list:David Hull's gear list
Canon EOS 5D Mark III Canon EOS M5
Peter 13 Veteran Member • Posts: 8,301
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

I wonder if the NR is for the dark parts only. If you could take a shot with a white napkin wrapped around the front of the lens (focused at infinity), this would be an interesting experiment. Vignetting and other light variations are low frequency effects and they should not change the picture much.

schwoofi Regular Member • Posts: 222
Thank you very much - very informative! [nt]

Thank you very much - very informative! [nt]

 schwoofi's gear list:schwoofi's gear list
Sigma DP1 Merrill Sigma DP3 Merrill Canon EOS 5D Mark III Canon EF 50mm f/1.4 USM Canon EF 70-200mm f/2.8L IS USM +4 more
OP bclaff Veteran Member • Posts: 7,308
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

can you show same analysis for D800 ?

See for the D800 and D4
http://forums.dpreview.com/forums/readflat.asp?forum=1021&thread=41097253

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

OP bclaff Veteran Member • Posts: 7,308
Re: Well, that was quick... ;-)

David Hull wrote:

Is there any way you can quantify the 5DII to 5DIII improvement in terms of EV or dB or something like that?

Not with this type of visual analysis. Possibly by digging deeper. (I'm looking into it.)

Also, I don't really understand the significance of the "cooking". I expect all manufacturers do whatever makes sense whether through HW or SW to optimize their system performance.

Some people, for example astrophotographers, would prefer no raw data manipulation whatsoever. And in some cases there is a fine line between what is done in the hardware and what is done by the firmware.

There are quite a few people who think that Canon does not manipulate their raw data in any way and that Nikon thoroughly "cooks" their data. Clearly they both do some manipulation to achieve their respective design goals. I hope that someday people will be less polarized about this topic. I do my best to be even-handed and consistent in my terminology regardless of camera brand.

Best Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Mark Kanawati Contributing Member • Posts: 620
Re: Thank you very much - very informative! [nt]

Indeed many thanks for improving my knowledge.

Mark

 Mark Kanawati's gear list:Mark Kanawati's gear list
Carl Zeiss Apo Sonnar T* 2/135
stephenmelvin Veteran Member • Posts: 4,547
Re: Well, that was quick... ;-)

bclaff wrote:

David Hull wrote:

Is there any way you can quantify the 5DII to 5DIII improvement in terms of EV or dB or something like that?

Not with this type of visual analysis. Possibly by digging deeper. (I'm looking into it.)

Also, I don't really understand the significance of the "cooking". I expect all manufacturers do whatever makes sense whether through HW or SW to optimize their system performance.

Some people, for example astrophotographers, would prefer no raw data manipulation whatsoever. And in some cases there is a fine line between what is done in the hardware and what is done by the firmware.

There are quite a few people who think that Canon does not manipulate their raw data in any way and that Nikon thoroughly "cooks" their data. Clearly they both do some manipulation to achieve their respective design goals. I hope that someday people will be less polarized about this topic. I do my best to be even-handed and consistent in my terminology regardless of camera brand.

Indeed you do. Clearly ISO 100000 isn't going to work too well for astrophotography.

Best Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

-- hide signature --
Taikonaut Senior Member • Posts: 2,513
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

Your test reveal pattern noise on D800 are worse than 5DIII. I would rank them in this order from best to worst as follow
1st 5DIII
2nd D800
3rd D4
4th 5D2

bclaff wrote:

can you show same analysis for D800 ?

See for the D800 and D4
http://forums.dpreview.com/forums/readflat.asp?forum=1021&thread=41097253

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

tigertails99 Contributing Member • Posts: 544
Thanks Bill n/t

no text

tony field Veteran Member • Posts: 9,420
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

bclaff wrote:

The ISOs from left to right are ISO 50, ISO 1600, ISO 6400, ISO 25600, and ISO 102400 (5D3).

Thanks for the test and, maybe more importantly, to the "interpretive link".

Is there any reason you chose ISO 50 as a low end and not ISO 100. I understand that, at least on Canon, ISO 50 is really a pull processing from ISO 100.

-- hide signature --
David Hull
David Hull Veteran Member • Posts: 6,117
Re: Well, that was quick... ;-)

bclaff wrote:

David Hull wrote:

Is there any way you can quantify the 5DII to 5DIII improvement in terms of EV or dB or something like that?

Not with this type of visual analysis. Possibly by digging deeper. (I'm looking into it.)

Also, I don't really understand the significance of the "cooking". I expect all manufacturers do whatever makes sense whether through HW or SW to optimize their system performance.

Some people, for example astrophotographers, would prefer no raw data manipulation whatsoever. And in some cases there is a fine line between what is done in the hardware and what is done by the firmware.

There are quite a few people who think that Canon does not manipulate their raw data in any way and that Nikon thoroughly "cooks" their data. Clearly they both do some manipulation to achieve their respective design goals. I hope that someday people will be less polarized about this topic. I do my best to be even-handed and consistent in my terminology regardless of camera brand.

Best Regards

You are doing great work, keep it up. The cameras are what they are and there is not much that can be done about that other than selecting the best solution for a given purpose. The more we understand, the better we will be at applying which ever brand we decide on. It was an honest question on my part – I would not be surprised if all manufacturers “twiddled” things a bit here and there. I was just wondering what the harm was, if any. I guess “RAW”, by definition should be “twiddled” as little as possible.

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS 5D Mark III Canon EOS M5
Christoph Beck Regular Member • Posts: 242
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

But is this noticeable for normal pics? Why do such an analysis?

OP bclaff Veteran Member • Posts: 7,308
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

Is there any reason you chose ISO 50 as a low end and not ISO 100. I understand that, at least on Canon, ISO 50 is really a pull processing from ISO 100

Good question.
As it turns out scaling has no effect on patterns.

I did double-check; no visible difference between ISO 50 and ISO 100 for these FFTs.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

OP bclaff Veteran Member • Posts: 7,308
Re: 5D3 Pattern Noise Improvement and High ISO Cooking
1

But is this noticeable for normal pics?

Yes; pattern noise and banding can be quite visible (and annoying when present).

Why do such an analysis?

It shows objectively what some people have been reporting anecdotally about their post processing experiences.

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

M Mason Regular Member • Posts: 214
Re: 5D3 Pattern Noise Improvement and High ISO Cooking

bclaff wrote:

can you show same analysis for D800 ?

See for the D800 and D4
http://forums.dpreview.com/forums/readflat.asp?forum=1021&thread=41097253

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Bill,

This is a great example of proper posting. Canon results in the Canon forum, N results in the N forum. Answer a question with a link. Thank you very much, and also the subject is quite interesting.

That said, I can't resist: Do you think that when images get too large (i.e. too many megapixels) the fast fourier transforms will take longer, thus becoming slow fourier transforms?

Mark

 M Mason's gear list:M Mason's gear list
Canon EOS 7D Canon EF 17-40mm f/4.0L USM Canon EF 50mm f/1.8 II Canon EF 70-200mm F4L IS USM Canon EF 85mm f/1.8 USM +11 more
Keyboard shortcuts:
FForum MMy threads