MC Denoise/Degrain: Prevent smearing effect

[moviefan]

Hi guys,

I have noticed a "smearing"-effect when degraining/denoising with e.g. MCTemporalDenoise(settings="low") in relatively flat areas with (in the source) a fair amount of grain. This seems to be a side-effect of MTools... Is there a way to prevent this effect while keeping the degrained clip stable? When I degrain less, the smearing doesn't appear but there is more grain left and it dances around.

Regards
moviefan

[Sagekilla]

The smearing probably isn't from MVTools but a spatial denoising routine that may be going on. AFAIK, MDegrain doesn't create a "smeary" effect like most denoisers do. It has a tendency to flatten out weak detail, but otherwise it won't exhibit a smearing type artifact unless you set the thresholds to "leave nothing behind"


If you try MDegrain1, do you get the same smearing artifacts?

src = last
super = src.MSuper(pel=2)
bvec = super.MAnalyse(isb=true, overlap=4)
fvec = super.MAnalyse(isb=false, overlap=4)
src.MDegrain1(super, bvec, fvec)

[aNToK]

That sounds pretty similar to the effect I've had with MCTemporaldenoise and several other motion-based denoisers. I played with them a bit and I'm guessing that what I was experiencing was what's referred to a 'banding'.

Have you tried to enable and crank up the "flat" parameter in MCTemporaldenoise? With some tweaking, it did a pretty good job of minimizing the effect in most of the vids I've been playing with. Just a rank amateur's thoughts...

[Sagekilla]

Note, banding isn't something that's limited to motion compensated denoisers. Any denoiser that removes too much noise can reveal banding, because the noise (or detail) was the only thing hiding it. MCTemporalDenoise can handle this internally, but otherwise you need to use GradFun2DB to dither your image and prevent any banding.

[moviefan]

I get the same effect using your MDegrain1 method. I have reconsidered my description of the problem and I am not sure, if it was clear what I mean. By smearing, I mean a temporal pixel locking, so that the pixel color and position remains the same for a couple of frames where there is e.g. horizontal moving of the camera. It looks to me like a smearing of the color of some pixels in a temporal manner. The shape of an almost flat area (although barely noticeable) doesn't change the position although the object moves. I hope this improves your imagination of my problem.

[Sagekilla]

Yes, that does help, since smearing can mean a lot of different things ;) I was thinking of the greasey painting type that normally happens. In your case, you can try lowering thSAD, which would cause MDegrain to filter less, and perhaps not smear things out. You could also try adjusting thSCD1 (http://avisynth.org.ru/mvtools/mvtools2.html#functions).

Could you upload a small (5 second) segment of your video where you get this smearing? Just cut it out, don't recompress or process it at all.

[moviefan]

I will do so at the weekend, when I have time. I have also noticed pretty visible chroma noise in various scenes. How can I properly denoise and degrain a video, by analysing it in some way and maybe processing the different planes separately? Is there a way, to only show the chroma plane, denoise this, then only show the luma plane, denoise it and after finding the right filters and settings processing the video with these filters and settings? Until now I have just used filter functions like MCTemporalDenoise etc. and chose settings according to my taste, but compressibility e.g. I guess to be improved by stabilizing the chroma plane properly although I don't easily notice that it needs heavier processing. So big question again: Is there a procedure for intelligent denoising/degraining of a video using analysis and filter adjustments according to these analysis? (analysis need not be computed but also judged by eyes)