removegrain and removedirt parameters for anime

[jarthel]

for high and low quality please.

Thanks :)

[Mug Funky]

i tend to just use the defaults...

for low quality (i assume you mean noisy/jittery), i like to hit it with something harder, but i haven't settled on something i like just yet

i'm experimenting with the multiple-averaged-encodings technique discussed a while ago, where you make several offset xvid encodes, layer them together (removing the offsets) and re-encode.

[lamer_de]

I haven't played around too extensive with those filters but with removegrain(mode=3) I noticed the effect stated in the documentation:

If mode=3 or 4, then thin lines may be destroyed by RemoveGrain. However, such thin lines are rare in practice.
Seems those lines are frequent in Anime though. ^_^ So I wouldn't blindly reccomend them for anime unless you check the output before encoding. As always, depends on the source and therefore on the thickness of the lines. Some anime have thicker lines that are unaffected. Haven't used removedirton many encodings, simply because nowadays anime are digital and don't feature "dirt", so the filter always changed parts of the picture that were part of the original image.

CU,
lamer_de

[kassandro]

I ran today some compression tests with the following results (the number on the left side is the output size, when the filters on the right side are applied):

source: b&w 8185 Frames, xvid quantiser=5, no audio
46.021.987 greyscale()
28.786.402 RemoveDust(4,true).greyscale() (with smooth=true)
28.968.340 RemoveDust(4,true).greyscale() (with smooth=false)
28.387.513 RemoveDust_old(4,true).greyscale()
33.869.453 RemoveGrain(mode=4)
41.509.363 RemoveGrain(mode=8)
43.128.603 DeGrainMedian(limitY=255, limitUV=255)
33.855.276 RemoveGrain(mode=4, modeU=-1).greyscale()
41.506.929 RemoveGrain(mode=8, modeU=-1).greyscale()
33.838.375 RemoveTemporalGrain(4, true).greyscale()

with the following avisynth functions:

function RemoveDust_old(clip input, int _mode, bool _grey)
{
repmode = 2
_modeU = _grey ? -1 : repmode
clensed = Clense(input, grey=_grey)
rep=Repair(clensed, input, mode=repmode, modeU = _modeU )
return RemoveGrain(rep, mode=_mode, modeU = _modeU)
}
function RemoveDust(clip input, int _mode, bool _grey)
{
repmode = 2
_modeU = _grey ? -1 : repmode
clensed = Clense(input, grey=_grey)
rep=Repair(clensed, input, mode=repmode, modeU = _modeU )
rg = RemoveGrain(rep, mode=_mode, modeU = _modeU)
return TemporalRepair(rg, rep, grey=_grey, smooth=true)
}
function RemoveTemporalGrain(clip input, int _mode, bool _grey)
{
repmode = 2
_modeU = _grey ? -1 : repmode
rg = RemoveGrain(input, mode=_mode, modeU = _modeU)
return TemporalRepair(rg, input, grey=_grey, smooth=true)
}

The source material was a pretty grainy but only moderately dirty b&w film from the fifties. "smooth" is a new variable from the upcoming version 0.7 of the RemoveGrain package?

What is the conclusion from these tests:
Firstly, while RemoveTemporalGrain(4, false) is much more conservative than RemoveGrain(mode=4), it yields nearly the same compression. Thus RemoveTemporalGrain(4,false) is the best conservative choice, though "smooth" is not yet in the current version 0.6.
Secondly, for more aggressive compression gain use the new RemoveDust instead of the old one, because it achieves nearly the same compression gain with considerably less motion blurring.
Thirdly, there must be something wrong with DeGrainMedian(limitY=255, limitUV=255), which is a temporal extension of RemoveGrain(mode=8).
Theoretically DeGrainMedian(limitY=255, limitUV=255) should yield slightly better compression than RemoveGrain(mode=8).

Version 0.7 TemporalRepair with smooth=false is identical to the current version 0.6 TemporalRepair.
Further tests are needed to confirm the above results.

[helix]

Originally posted by Mug Funky
i tend to just use the defaults...

for low quality (i assume you mean noisy/jittery), i like to hit it with something harder, but i haven't settled on something i like just yet

i'm experimenting with the multiple-averaged-encodings technique discussed a while ago, where you make several offset xvid encodes, layer them together (removing the offsets) and re-encode.

Would you mind going into a bit more deatil about what exactly that is. I'm curious as to how thats done.