Noise reduction of high ISO images near gamut boundaries

[Doug Gray]

When working in a large gamut space, like ProPhoto, higher ISO images tend to have a lot of pixels, due to noise, that are outside a display and printer's gamut even when the actual image is within gamut.  These are clipped or remapped to the gamut boundary. but others that have noise pushing them inside the gamut are rendered properly. Because they are so close human vision sees an average over  small distances and the apparent colors are rendered less saturated than they would be with a less noisy image. Seems to me that a selective, noise reduction process that works only near gamut boundary could produce somewhat better prints. Does anyone know of any software that has options to do this?

[Bart_van_der_Wolf]

When working in a large gamut space, like ProPhoto, higher ISO images tend to have a lot of pixels, due to noise, that are outside a display and printer's gamut even when the actual image is within gamut.  These are clipped or remapped to the gamut boundary. but others that have noise pushing them inside the gamut are rendered properly. Because they are so close human vision sees an average over  small distances and the apparent colors are rendered less saturated than they would be with a less noisy image.

Hi Doug,

Yes, a good point. But if we use noise reduction before gamut down-sampling, the remapping will include fewer oversaturated (and undersaturated) OOG overshoots (and undershoots). That should give a more average (and correct average chrominance) new gamut boundary.

Seems to me that a selective, noise reduction process that works only near gamut boundary could produce somewhat better prints. Does anyone know of any software that has options to do this?

Other than using a saturation selection mask, and specifically denoising that selection, I do not know of a commercial Denoising application that specifically targets (OOG) saturation (except for Topaz Denoise shadow color). While the better ones can make a specific noise profile for a typical camera and specific conversion settings, they do not allow a specific saturation approach (they are supposed to handle all noise equally, but with basic tweaks).

A program like NeatImage is already highly tuneable, even if based on a normal Raw conversion. The user can make pre-made profiles for different ISO settings per camera model, in case a specific image offers too few clues to make a complete image specific profile. One can boost the Chrominance noise reduction separately from the Luminance noise reduction. Also Topaz Denoise, which makes an automatic per image noise profile, can be set to modify the noise rendition in shadows to avoid overly red shadows (caused by clipped undershoot shadow noise conversions).

Cheers,
Bart

[TRANTOR]

You can try my new noise reduction plugin: http://nuclearlight.net/sightline.html
It works in Lch mode and you can control level of noise suppression for example in chroma channel only.

For now SightLine cannot evaluate noise reduction level that depends against chroma level, but this sounds interesting.

[digitaldog]

All I can state is that when I plot the gamut of images in ColorThink Pro, it's impossible to do so on full resolution data (at least from my 5DMII) and I usually resort to something like 300x300 pixels. Resampling a large image down that much reduces a lot of noise.

[Bart_van_der_Wolf]

All I can state is that when I plot the gamut of images in ColorThink Pro, it's impossible to do so on full resolution data (at least from my 5DMII) and I usually resort to something like 300x300 pixels. Resampling a large image down that much reduces a lot of noise.

Hi Andrew,

You could take a crop, instead of resizing, and measure before and after denoising.

Cheers,
Bart

[digitaldog]

You could take a crop, instead of resizing, and measure before and after denoising.

I could, IF the crop represented the all the colors in the image I wanted to plot. That's rarely the case. So yeah, if I had say a gray card, fine. If I had an image of say a Macbeth Color Checker, nope.

[Doug Gray]

All I can state is that when I plot the gamut of images in ColorThink Pro, it's impossible to do so on full resolution data (at least from my 5DMII) and I usually resort to something like 300x300 pixels. Resampling a large image down that much reduces a lot of noise.

Indeed it does.  And that indicates that there might be some advantage to software that selectively moved out of gamut colors to those on the gamut boundary and increased saturation of nearby pixels to compensate. As a practical matter this would only improve rendering images where the out of gamut pixels were mostly noise introduced artifacts. It might make things worse where the actual colors were beyond the gamut. It's almost certain that out of gamut pixels shown in the reduced size required for CTP mean those portions of the image were truly out of gamut.

Another thought is that noise induced, out of gamut pixels are much more likely in low luminance areas of an image since noise variation is perhaps around 5 times greater at L=20 than L=80.