A700 terrible red channel noise problem

[bakubo]

Since I got my A700 last year I have noticed a continuing problem that I didn't have with my Canon 30D. I have to be *very* careful when processing A700 shots with blue sky because even at 16 bits it is so easy to get ugly, mottled skies. I never had this kind of severe problem before the A700. Here are just two examples that were both taken at base ISO in raw and converted with ACR (all settings set to zero) and converted to a 16-bit tiff. To clearly illustrate the problem in the sky I then used the channel mixer set to monochrome and set the red channel to 100%, green channel to 0%, and blue channel to 0%.

A700:


30D:


Here are some 100% crops of the sky.

A700:


30D:


Does anyone know what the problem is here? Although I have not created examples from the A100, KM 7D, and Canon 300D I don't recall ever having the kinds of problems I have with the A700.

[bakubo]

Sony has not yet been able to get up to par yet in this area...it's an historic problem.

Yes, that's it. Thanks for the link.

Color Temp & WB adjustment may help to minimize the effect at times but Sony needs work in the noise & noise processing areas...they have needed work from the beginning.

I first noticed it last year, but while I was in Utah, Wyoming, California, etc. on the recent road trip I had lots of photos with blue sky and it was driving me nuts. Back in Hawaii it is again a big problem sometimes. Something is pretty seriously wrong with the A700 here, I think, since I have never seen it on 4 other DSLRs I have owned. I'll check the A100 to see if it is also a mess, but I never noticed it before so I think it is okay. I wonder if this is just a problem with Sony CMOS sensors?

[roysmith]

Same problem here, but removing the red channel didn't help with the noise so I'm not so sure that's the real problem... looks more like the green channel to me. You can try disabling one channel at a time in Photoshop or Gimp, keeping two channels enabled instead of one. The green channel is stronger than the red channel on skies, unless is a sunrise/sunset of course
Mi old nikon D50 shows some noise on the skies too, but not so much as the A700... I guess it could be also because of the resolution difference (6 vs 12MP)
Just out of curiosity, were you using a polarizer on that shot?

[bakubo]

Same problem here, but removing the red channel didn't help with the noise so I'm not so sure that's the real problem... looks more like the green channel to me. You can try disabling one channel at a time in Photoshop or Gimp, keeping two channels enabled instead of one. The green channel is stronger than the red channel on skies, unless is a sunrise/sunset of course

It is different for me. If I set the channel mixer to red channel 0%, green channel 100%, and blue channel 0% it looks clean.

Just out of curiosity, were you using a polarizer on that shot?

No polarizer.

[roysmith]

Same problem here, but removing the red channel didn't help with the noise so I'm not so sure that's the real problem... looks more like the green channel to me. You can try disabling one channel at a time in Photoshop or Gimp, keeping two channels enabled instead of one. The green channel is stronger than the red channel on skies, unless is a sunrise/sunset of course

It is different for me. If I set the channel mixer to red channel 0%, green channel 100%, and blue channel 0% it looks clean.

Ok but try setting the red channel to 0% and both green and blue channels to 100%. Then the green channel to 0% and both red and blue to 100%.
I noticed less noise without the green channel; at least on the picture I tested the red channel wasn't that important as a sum.

[bakubo]

Ok but try setting the red channel to 0% and both green and blue channels to 100%. Then the green channel to 0% and both red and blue to 100%.
I noticed less noise without the green channel; at least on the picture I tested the red channel wasn't that important as a sum.

If you set two channels to 100% you will get a very blown out result.

[roysmith]

Ok but try setting the red channel to 0% and both green and blue channels to 100%. Then the green channel to 0% and both red and blue to 100%.
I noticed less noise without the green channel; at least on the picture I tested the red channel wasn't that important as a sum.

If you set two channels to 100% you will get a very blown out result.

Yes, technically you're correct, sorry
This is what I meant:

[David Kilpatrick]

For any given saturated colour, the complementary (or closest) channel will contribute most to luminance values, and therefore to noise. The Sony blue sky is very different from the Canon blue, in the same situation, and usually has deeper colour, lower luminance and more saturation than a default Canon pic. If you take your Sony example and adjust the darkest sky grey to the same grey as the darkest in your Canon example - or just make the Sony sky a closer match - some of the extreme red noise will disappear.

If you took a red area of image and did the channel boost on the blue channel, it would be that channel creating most noise. The red channel is the worst to start with, but of course a blue sky being a smooth gradition of a nearly complementary colour will always show it worst.

Working from raw, the Sony A700 red channel is already being given a 100% extra gain boost by the default WB settings (this is what the so-called UniWB sets out to solve). To reduce sky noise, a simple solution for in-camera JPEGs is to fit an 85 filter (brown colour, daylight conditions to tungsten WB) and then set the camera to Tungsten fixed WB. This is not quite as extreme as using UniWB, a white balance preset which levels all three channels exactly. The lower gain applied to the red channel using Tungsten WB reduces blue sky noise, and the 85 filter acts almost like a 2X yellow in darkening the sky. This method also works OK for raw of course.

David

[agorabasta]

First off - a deep blue sky is noisy by nature. Just stare into the clear blue sky at midday - you can definitely see that.

Second - there are too few red photons there, and then that red is multiplied by a factor of about 2 for daylight after having it's shadow clipping been set too high by default.

Solution is simple - don't do channel shadow clipping (black level subtraction) before demosaicing, then demosaic at a WB settng that fills channels histograms the best possible way, then only apply WB and black per-channel subtraction on your 16bit/channel demosaiced image.

(A simpler procedure - jack the shadows up with the 'primary curve', develop with moderate saturation, push the saturation up in the Lab space by applying S-curves to 'a' and 'b'. If you need a 'Canon blue', admix some -cyan to the +black in CMYK)

P.S. That "Canon sky" is so far from real...

[agorabasta]

For those who need a 'Canon sky' at least effort, here goes:

in CMYK, do black'=black-0.1*cyan

You may prefer a different cyan multiplier, play with it a bit; the cleanest 'Canon skies' turn out somewhere between 0.07 and 0.25 (somewhere there the red noise nearly cancels out, the optimum multiplier value depends on your exposure preferences)

[bakubo]

Don, how did you split the R, G, and B channels in PS? I have PS CS2 and I can't find a way to do it. I can easily split the channels in PSP X and then recombine them. I was thinking that splitting them, use Neat Image on the red channel, and then recombine them might work. I will give that a try. I would like to do it in PS though so if there is a way to split and recombine channels that would be good. I just can't find it in PS.

[bakubo]

First off - a deep blue sky is noisy by nature. Just stare into the clear blue sky at midday - you can definitely see that.

Hmm, I just went out and looked at a blue sky. I didn't see noise. Are you saying the red channel should be mottled and noisy?

Second - there are too few red photons there, and then that red is multiplied by a factor of about 2 for daylight after having it's shadow clipping been set too high by default.

Okay, I understand what you are saying. But, why is the A700's sky so noisy compared to all the other DSLRs I have used?

Solution is simple - don't do channel shadow clipping (black level subtraction) before demosaicing, then demosaic at a WB settng that fills channels histograms the best possible way, then only apply WB and black per-channel subtraction on your 16bit/channel demosaiced image.

I didn't do any of those things when I did the raw conversion. See my OP. I set everything to zero in ACR. I don't know what ACR is doing under the covers though and I don't have much control over it.

(A simpler procedure - jack the shadows up with the 'primary curve', develop with moderate saturation, push the saturation up in the Lab space by applying S-curves to 'a' and 'b'. If you need a 'Canon blue', admix some -cyan to the +black in CMYK)

In ACR I set everything including saturation to zero.

P.S. That "Canon sky" is so far from real...

The Canon and other DSLRs may not be real but they are much realer than the A700!

[agorabasta]

Just a straightforward conversion in Ufraw, using my regular custom technique (raised black point, logarithmic primary curve, then restore black point automatically in the secondary curve, no camera profile), no image-specific adjustments:


Now a 100% crop:

[agorabasta]

Solution is simple - don't do channel shadow clipping (black level subtraction) before demosaicing, then demosaic at a WB settng that fills channels histograms the best possible way, then only apply WB and black per-channel subtraction on your 16bit/channel demosaiced image.

I didn't do any of those things when I did the raw conversion. See my OP. I set everything to zero in ACR. I don't know what ACR is doing under the covers though and I don't have much control over it.

The problem is that ACR does all those bad things for you... And you have near-zero control.

[agorabasta]

Here's a link to the junk album containing the full size developed image (and some other irrelevant junk) - http://picasaweb.google.ru/AgoraBasta/T ... directlink - quite open for dissection. (As it's a JPEG, use an appropriate space, like YIQ or any YCrCb)