DearS - Weird Problem

[Xicor]

If you notice the guy's face you can see different "shades" of color. I tried numerous codecs, such as XviD, x264, and Nero AVC.

*Here is the Clip* (http://rapidshare.de/files/16589485/dears.avi.html)

Here is the script:

hqdn3d(2)
convolution3d(preset="animehq")
limitedsharpen(3)

Thanks in Advance
~Xicor

[Lil' Jer]

That's called color banding which is one of the inherent problems with YV12's lower color depth. Switching between those codecs will obviously have no effect due to the fact that they all work in the YV12 colorspace. I suggest you try looking in this thread as what they are working on may help you: http://forum.doom9.org/showthread.php?t=107910

[Xicor]

I don't think color banding is the problem. I've tried some filters, including gradfun2db, but the results remained the same.

[foxyshadis]

You still haven't posted a source clip. Is it in the source? It appears to either be banding so strong mild debanding won't eliminate it, or temporal scene bleed. Perhaps both.

[DeathWolf]

I don't think color banding is the problem. I've tried some filters, including gradfun2db, but the results remained the same.

gradfun2db isnt meant for image filtering before encode.
All current yv12 codecs will wash away it's effects

[Mug Funky]

it looks like the result of either a temporal filter (that works in a cumulative way) leaving junk from a previous scene there, or an xvid encode with cartoon mode gone horribly wrong (i've never seen it do that though - but i never let it get to quant 14 before either. looks surprisingly good at that level though).

we'll have to know encoder settings and filters used and stuff like that to be of any use. that gunk is not removable - it's too correlated to kill with a temporal filter, and too sharp and varied to kill with a spatial filter if you want to keep details.

as far as any filter is concerned, that stuff is perfectly valid signal.

[Didée]

as far as any filter is concerned, that stuff is perfectly valid signal.
Keep this quote! It is, was, and will be worth referencing to in lots of situations ...