Avoiding blocking due to noisy source

[Aramgutang]

I've recently been having a lot of trouble making a good encoding from a noisy source. It's a 1h 30min PAL movie with lots of luma noise (the grainy type). The highest parameters that I've tried encoding with are 1,624 kbps with 0.393 b/pf, and I still get a ridiculous amount of blocking, even with "high noise" selected in GKnot.

Here is a screenshot from the original movie to illustrate the kind of noise I'm talking about:
http://img442.imageshack.us/img442/641/noisysourcesp0.jpg
(and it continuosly varies with every frame)

My XviD settings (v.1.1.0 from Koepi):
VHQ 4, Motion Search 6, Chroma optimizer on, QPel on, GMC on, B-VOPs on, Adaptive Quant on, Trellis on, Turbo off, Quant Matrix: MPEG (I've also experimented with EQM v3 ULR and H.263 and found MPEG to give the best results)

I was wondering if anyone has any advice on how to reduce the blocking caused by the noise. I don't actually mind the noise itself, just the blocking artefacts, and I'm OK with losing detail. Are any of the XviD settings I'm using particularly bad at handling noise? And can you recommend any noise filters that are good at handling this type of noise?

Thank you.

[CWR03]

Personally I use Gordian Knot and its preset grain filters. I recently encoded a few old and very grainy films which compressed very well with the degrain at its highest level.

XviD doesn't like grain, and I don't believe settings within XviD can do much to improve things.

[DarkZell666]

hqdn3d or Convolution3d avisynth filters usually do a good job.

I've also had very clean results with TTempSmooth and TTempSmoothF. Do try Frfun7() too, I like how it worked out on some sources of mine :) It's slow but worth it imho.

With noisy sources, I doubt GMC will even kick in, and it's a speed killer, so don't even bother turning it on in your case.

[Mug Funky]

oh my, is that some 16mm film?

try fft3dfilter(bw=16,bh=16,ow=8,oh=8,bt=0,sigma=1.)

if it's too much/not enough, then lower or raise the sigma parm.

[Aramgutang]

Thanks for the tips, I'll try them out this week, see if they help.

2 Mug Funky: It's an unrestored 1982 Soviet film, actually.

[Blue_MiSfit]

yeah thats some ugly luma noise. fft3dgpu / fft3dfilter should be able to nuke it.

[henryho_hk]

Try: VHQ 1, Motion Search 6, Chroma optimizer on, QPel off, GMC off, B-VOPs on, Adaptive Quant off, Trellis on, Turbo off, Two-pass encode

Since this is a noisy source, you need a smooth matrix, e.g.
EQM v3 LR, H.263 or even Jawor 1CD. (ULR ceases to be recommended by the author some time ago, while denoise filters is probably needed for EQM v3 LR.)

BTW, is it an interlaced footage? If so, you need to deinterlace it before xvid encoding. Otherwise, it would be another major source of blockiness.

------------------------

Thanks DarkZell666 for the reminder. It should be VHQ 1 and Motion Search 6.

[DarkZell666]

big fat warning about VHQ4 though: I can remember it has been mentioned here that high VHQ's weren't recommended for noisy sources, and that they were even counter-productive on noisy sources specifically. Rather stick with VHQ1. (or try yourself and see what suits you best ;)).

[halsboss]

Hi Aramgutang, I have a similar noisy clip. Could you please post the final AVS script you ended up using plus a comparative snap .JPG ? Thanks.