Chroma Noise Reduction

[den]

Edit to my previous post...

Composite-Soft Light will produce a softer ambience in step(4) if a preference.

No it will not. Composite-Hard Light for step(4) is best... ...den...

[tomczak]

Den, this is great! I need to do more testing, but it seems to work well. I find the idea of flattening HSL-L=50% and thus leaving a combined 2-channel colour info remarkable!

I was experimenting trying to induce PWP to convert RGB to something similar to Y'CbCr channels (AS seems to use HSV-S/V for chrominance blurring, splitting the image into Y'CbCr should perceptually separate chroma from luma better). Then blur the Cb/Cr, and then recombine all channels. I went nowhere. One passing idea was to use Monochrome blending RGB channels to get the Y', Cb and Cr channels, which is possible, but I don't know of a way to recombine them back into an image in PWP (other than combining them as fauxRGB, followed by a complex chain of subsequent Monochromes and channel splits/joins again).

With Den's method, one possible modification could be trying different a blurring method: excessive Gaussian blur may cause colour 'leaking'/'glowing' at edges, and one needs quite a bit of a blur to reduce high ISO chroma noise - to potentially reduce colour spills across the edges, the Bilateral Sharpen with Sharpen value and threshold set to zero may be used in the blurring step.

Another blurring possibility that should limit colour 'glowing' at edges could be AS Blur All with low frequency edge threshold - although I had a problem with the AS blur radius not going high enough to actually blur the common high-ISO, large blue/yellow blotches. Maybe more than one pass would work, but I haven't tried it yet.

Thanks!

[tomczak]

I experimented with colour channels, trying to determine what could be done to improve on AS chroma noise reduction.

I split high ISO images into JPEG-like Y'CbCr channels (outside of PWP), then used AS to reduce noise in Cb and Cr only, and then re-assembled the images.

Here are the conclusions:

1) While using HSV to separate luminosity for Sharpen tab works perfectly fine, for Noise Reduction tab, perceived chroma noise is much better separated from luma noise by Cb and Cr channels, and thus much more effectively removed by blurring these two channels than by blurring H and S channels - the HSV-V channel carries chroma noise information as well.
2) If AS implemented CbCr-only noise reduction, there may be some images with high ISO blotches for which the Blur Radius of 7 will be insufficient.
3) Some images could benefit from blurring Cb and Cr separately (e.g. the example image at the beginning of this post had chroma noise mostly in Cb, not Cr).

Cheers!