[MEncoder-users] Noise reduction and interlacing: (hq)denoise3d vs. nr

[Ivo]

On Tuesday 03 July 2007 10:24, Pierre Catello wrote:
> 2007/7/2, Ivo <ivop at euronet.nl>:
> > IMHO tfields/tinterlace is prefered if you run a temporal denoiser. If
> > you only (de)interleave as Nico suggested, the denoiser works with
> > different data and will perform worse.
> >
> > For example: (F=frame, f=field)
> >
> > Input F1(f1,f2) F2(f3,f4) F3(f5,f6) etc...
> >
> > Fictional denoiser uses current frame, and the next two frames.
> >
> > If you use il=d and il=i:
> >     F1 is denoised by using f1<-->f3<-->f5 and f2<-->f4<-->f6
>
> IMHO, this is the right way, and I thought that
> tfields=0,...,tinterlace was equivalent
>
> > If you use tfields and tinterlace:
> >     F1 is denoised by using f1<-->f2<-->f3 and f2<-->f3<-->f4
>
> f1 and f2 or f2 and f3 do not belong to the same actual frame, so a

I suppose you mean scanline? Like what Nico said, top field/bottom field.

> spatial filter will likely mess things if there is motion between F1
> and F2.

I think that will be a problem no matter which method you use. Temporal 
denoisers seem to work better if there's little or no motion. But the human 
eye has trouble seeing noise if there's a lot of fast motion, so that is 
not really a problem I guess.

After what Nico said, I think that indeed temporal denoising is best done 
with il=d/i (to which you seem to agree) and spatial denoising is best done 
with tfields=0,tinterlace. So perhaps a filter chain like:

-vf il=d,temporal_denoiser,il=i,tfields=0,spatial_denoiser,tinterlace

will produce good results.

--Ivo