Banding

[Benus]

Hello, I have problem with debanding BD anime. I tried to repair this but results was bad. I wanna ask for yours help. That's what I tried:

My options in DGAVCIndex: Video / YUV -> RGB -> PC Scale

First:
AVCSource("D:\K-On vol.1\vol.1\BDMV\STREAM\00000.dga")
Fielddeinterlace()
LanczosResize(1280,720)
Trim(1600,2000)
Sample
After that I tried to remove banding:
AVCSource("D:\K-On vol.1\vol.1\BDMV\STREAM\00000.dga")
Fielddeinterlace()
gradfun2db(3)
LanczosResize(1280,720)
Trim(1600,2000)
Sample
Megui profile:
program --profile high --pass 2 --bitrate 1777 --stats ".stats" --ref 12 --no-fast-pskip --b-adapt 2 --b-pyramid --direct auto --deblock 1:1 --subme 5 --trellis 2 --partitions all --me umh --merange 32 --thread-input --aq-strength 0.5 --no-dct-decimate --output "output" "input"

It's look bad so I used ColorMatrix(). Result that same. I'm not professional encoder, I'm really desperate right now.

[poisondeathray]

Have you tried a higher bitrate?

Can you upload a corresponding sample of the source?

[Benus]

I even true at constans quality=1 but it look exacly that same.
Orginal Source

[jagabo]

What do you mean by "banding"? I didn't see much in the way of posterization.

[poisondeathray]

And why fielddeinterlace() ? Source you uploaded is progressive

[Benus]

Ok, I removed Fielddeinterlace() from script. jagabo, I though that it's banding, so what it is?

[poisondeathray]

Describe what you mean by "it" or "banding"

The main problem I see is edge artifacts, and is a side effect of psy-rd. I would disable psy-rd on most types of anime, and lower AQ strength (I usually disable it).

Also this sequence has a lot of scene changes, and I-frames. So you may need more bitrate for a cleaner encode/less artifacts

[Benus]

I mean that by banding:



Psy-RD=disable
AQ=disable
Higher bitrate=doesn't work

[poisondeathray]

This type of situation is tough to deal with in anime, and it's not specfic to x264. It happens with all 8-bit compression,and you will see it happen with other h.264 encoders, xvid, mpeg2 etc...

You can increase the dithering and grain in the filter, but it will begin to look noisy. You can limit the damage to scenes where banding occurs (it normally occurs in darker areas with gradients), by using Trim(). This has to be done in the script manually, but I strongly suggest applying to sections. If you want to get more precise you could use separate masks. It's difficult to find the exact values or threshold to use where you minimize grain and prevent banding. It takes some trial and error. For example I used GradFun2DBMod(thr=1.0,str=3,strc=3) , but you might be able to get away with something lower or perhaps you want to specify something higher to your personal preferences. You're definitely going to have to make some tradeoff's in your decision

avcsource("Orginal.dga")
LanczosResize(1280,720)
Trim(24,0)
a=last

b=a.Trim(0,10)
c=a.Trim(11,46).GradFun2DBMod(thr=1.0,str=3,strc=3)
d=a.Trim(47,0)

b++c++d

gradfun2dbmod.mkv

[lordsmurf]

This is referred to as "posterization" or "posterizing", a compression of the color palette. Indeed, an issue with 8-bit colorspace encoding. However, more often, it's just a crappy encoding job -- inadequate bitrate.

[jagabo]

Ah, I was in a very brightly lit room on an unfamiliar computer earlier and couldn't see the posterization and macroblocking.

[Benus]

After that I should to try remove noise?

[poisondeathray]

After that I should to try remove noise?

Nope.

Dithering and noise are techniques to "tricks" prevent banding. If you remove the noise, you will get banding again. Pick one or the other. Trade offs. Note this will increase the required bitrate for the same quality, because of the added "noise," that's why you may want to limit filters to small sections predisposed to banding instead of the whole thing