Denoising analog capped YUY2

[Avisynth_challenged]

Hey all,

First off, a big thank you to everyone who has taken the time to read and respond to my various posts. I'd like to reiterate that this forum is an immense treasure trove of good information about all things video. The advanced level of knowledge here is coupled with a friendly community spirit that is uncommon in online forums.

Okay, with all my cheap flattery out of the way, I have another question

I'm evaluating some denoisers to clean up my analog capture of an LD source. My plan is to process luma and chroma separately, then merge them together to obtain a final denoised output.

For the luma part, I'm evaluating fft3dfilter and mvdegrain3 (I'm really impressed with the results of the latter... but, oh so sloooooow!). Are there other luma cleaning methods I should be considering for a filmed (23.976fps after IVTC) live action analog source?

For the chroma part, I'm not sure how to procede. For one thing, I'd like to keep the chroma intact (i.e. no colorspace conversions) in the final output. Also, the main problem with the source as far as chroma is concerned is residual rainbowing on certain areas of the video image (usually places with high frequency details) that got past my pre-ivtc derainbowing attempt. I think that these remaining rainbows (most were removed in the initial effort) were in the original LD master. So now I have to filter them out as progressive frames since I'm now past the IVTC stage. Any suggestions for chroma cleanup of rainbows for progressive YUY2 sources would be very much appreciated.

Thanks again for all your help. I can't believe how far Doom9 has taken the scientific art of video/image processing, and I know that my project will benefit as a result

[Comatose]

Usually when asking for advice like this it's a good idea to provide a short sample, otherwise all we can do is throw filter suggestions at you

[thetoof]

FFT3DFilter has a very nice feature: you can specify the filtering strength on 4 different frequency ranges.
Use fft3dfilter(bw=32,bh=32,oh=16,ow=16,bt=5) as your base and add settings depending on what you want.

Tweak sigma=a, sigma2=b, sigma3=c, sigma4=d to ajust the strength of the filter (sigma=highest frequencies & sigma4=lowest)

Change plane=x to select the processed color plane: 0 - luma(Y), 1 - chroma U, 2 - chroma V, 3 - chroma planes U and V, 4 - both luma and chroma

You can also select the plane to filter with dfttest. For other filters (like mvdegrain, temporaldegrain, mc_spuds, udegrain, anisotools, tnlmeans, etc etc), use mergechroma() to take the chroma from another clip and merge it in the clip where you processed luma.

source=x
a=source.filter for luma
b=source.filter for chroma

a.mergechroma(b)

[Avisynth_challenged]

Thanks for the responses so far

Originally Posted by Comatose

Usually when asking for advice like this it's a good idea to provide a short sample, otherwise all we can do is throw filter suggestions at you

Okay, I've got not one, but two short samples uploaded to Rapidshare. Both are 30 frames long, 720x480, 23.976fps (progressive), YUY2 Huffyuv encoded (via ffdshow), no audio.

Sample 1 (lumanoise.zip) *HERE*
This clip is a static nighttime scene that shows overall luma noise in the source.

Sample 2 (chromanoise.zip) *HERE*
This clip exhibits lots of rainbows in fine detail (high frequency) areas. Note that at the beginning of this clip, the camera is slowly panning right, and during this camera movement, you can see the rainbows most noticeably in the trees and on the trail the buggy is on. Towards the end of the clip, the camera stops panning, and the rainbows go away. (Perhaps I need a temporal chroma smoother?)

@ thetoof:
Thanks for the information about fft3dfilter. I will see if I can suppress the chroma noise (U and V) by adjusting the sigmas for higher frequencies.


Once again, thanks to all.