Blocky artifact clean up

[killerteengohan]

This particular Blu-Ray source I have seems to have lots of blocky artifacts on it. They are not moving at all, its 100% still as the video plays, so I don't think its grain or compression artifacts.

Anyways, typical grain removal filters don't really have a huge effect unless I set them to overkill, and smear out everything badly.

Any ideas on targeting these artifacts while minimizing smearing or damage to everything else? If that's even possible.

https://forum.videohelp.com/images/imgfiles/eg1nfoU.png
https://forum.videohelp.com/images/imgfiles/nqgzMqQ.png

[Selur]

mClean might help, but without a small video sample that is just a wild guess,...

[killerteengohan]

I cant get that mClean filter to work for me for whatever reason.

I tried covering it up with dithering but it didn't really come out so nice.

[poisondeathray]

I tried covering it up with dithering but it didn't really come out so nice.

You would have to denoise at least some amount before "covering up" with dithering

vapoursynth vs-dpir work ok for the denoising step; minimal degradation around strength 8-10. You can add a small amount of f3dkb and use weaker denoising strengths

[Selur]

I cant get that mClean filter to work for me for whatever reason.

Works fine for me in Avisynth and Vapoursynth without a problem.
fft3d (6+ sigma) followed by some adaptive sharpening and/or line darkening should also work for anime, but is more likely to kill some wanted details along the way.

I tried covering it up with dithering but it didn't really come out so nice.

Dithering wouldn't help. Debanding could help, but probably would also destroy since you would need to increase the threshold/strength ridiculously high.

Cu Selur

Ps.: you could also protect the lines by inverting a line/edge mask and thus only filter the rest,...

[killerteengohan]

I cant get that mClean filter to work for me for whatever reason.

Works fine for me in Avisynth and Vapoursynth without a problem.

Obviously it works fine for you, or you wouldn't have suggested it, but that doesn't change that I cant get it working. It's okay though. It's just some stupid technical reason that I'm no longer in the mood to mess with.

I can clean it out like you did in those examples with other filters, but it destroys more detail than I would like. I even tried something Jagabo taught me quite a while ago about filtering only the chroma or luma planes separately, and I still ended up destroying too much. I could use a mask to protect lines, but it would still smear out backgrounds and things.

I found out these are in the DVD's from many years ago, and tv recordings as well, so it's apparently a source issue or intended to be there. I guess I will just minimally cover it up or reduce it some, and get used to it.

Thanks anyways for the ideas I was asking for.

[jagabo]

I didn't see any blocks. Just (simulated?) brush strokes and noise. A light KNLMeansCL() can reduce the noise without blurring the lines and obliterating too much detail.

ConvertBits(10) # optional
KNLMeansCL(d=0, a=2, h=1.0)
ConvertBits(8) # optional

Can't say about the temporal component since you didn't provide a video sample.

[Selur]

I didn't see any blocks. Just (simulated?) brush strokes and noise.

boost the luminance on your display and zoom in.

A light KNLMeansCL() can reduce the noise without blurring the lines and obliterating too much detail.

you are right KNLMeansCL() does also help.
unfiltered:

vs-dpir:

ff3dfilter:

knlmeanscl:

(on actual video all of these might behave differently, RealESRGAN, Anime4kCPP and similar also can do the job)

[killerteengohan]

Can't say about the temporal component since you didn't provide a video sample.

I don't mind giving one, but I only remember how to losslessly demux video from DVD VOB files for giving DVD samples.

Do you know a way to cut out only 30 seconds to 1 minute of .m2ts file without needing to encode it? The only way I know how to get you a sample from an .m2ts file is select 30 seconds to 1 minute worth of frames in avisynth script and encode with a high bitrate to help keep out compression. It should be adequate, but I'm sure you would prefer non encoded sample.

[Selur]

DGIndex allows to easily extract a video part from a vob file.

• start DGIndex
• select the section you want to extract using the buttons in the lower right cornet
• press "File->Save Project and Demux Video" and specify where the files shoud be created

this will extract audio&video streams of the selected section and also creates a d2v file.
If you then share the video stream (.m2v) you are finished.

Cu Selur

[killerteengohan]

DGIndex allows to easily extract a video part from a vob file.

• start DGIndex
• select the section you want to extract using the buttons in the lower right cornet
• press "File->Save Project and Demux Video" and specify where the files shoud be created

this will extract audio&video streams of the selected section and also creates a d2v file.
If you then share the video stream (.m2v) you are finished.

Cu Selur

I just said I know how to do DVD's and asked about .m2ts, not .VOB. DGIndex wont even open or recognize .m2ts files for me.

[Selur]

Sorry, I misread (DGIndex does work for m2ts files here assuming the contain MPEG-2 Video)
For most formats you can also use MKVToolnix GUI, open the file, tell mkvtoolnix to chunk the clip in 30 seconds chunks, then upload one of those chunks.

[killerteengohan]

Sorry, I misread (DGIndex does work for m2ts files here assuming the contain MPEG-2 Video)
For most formats you can also use MKVToolnix GUI, open the file, tell mkvtoolnix to chunk the clip in 30 seconds chunks, then upload one of those chunks.

All DGIndex says to me is "no video header sequence found" no matter what .m2ts file I try to open.

I can get the file into mkvtoolnix, but dont see how to split it yet.

EDIT: I figured it out in mkvtoolnix. Thanks!

I will upload soon.

[Selur]

mkvmerge is only key frame accurate, also it only takes times not frames as input.
https://mkvtoolnix.download/doc/mkvmerge.html#mkvmerge.description.split
DGIndex isn't frame accurate either. To cut mpeg based stuff frame accurate you usually need to reencode (ignoring special cases like cutting on key frames).

[killerteengohan]

Here is the sample
https://mega.nz/file/Fg5hlAxT#dmZfpMiJH5QfUWA46wpMAJlQooPiluVU3x6Myk-9w40

[killerteengohan]

mkvmerge is only key frame accurate, also it only takes times not frames as input.
https://mkvtoolnix.download/doc/mkvmerge.html#mkvmerge.description.split
DGIndex isn't frame accurate either. To cut mpeg based stuff frame accurate you usually need to reencode (ignoring special cases like cutting on key frames).

mkvmerge took the frame numbers as inputs and came out just fine. Was frame accurate enough for me, so it is capable of frame numbers as inputs. I cant say if it was 100% frame perfect without examining it, but it started and ended exactly where I wanted it to.


[Attachment 63688 - Click to enlarge]

[Sharc]

tsMuxerGUI
Tab Split & cut

[Selur]

mkvmerge took the frame numbers as inputs and came out just fine. Was frame accurate enough for me, so it is capable of frame numbers as inputs. I cant say if it was frame perfect but it started and ended exactly where I wanted it to.

happy that worked for you

----
vs-dpir + line darkening (+deband when going to 8bit encoding) looks the best to me, after that I would go with line darkening + fft3dfilter(sigma=10) (+deband when going to 8bit encoding).
But you will always potentially loose some fine detail (see the wall).

Problem with removing such compression artifacts is that they are basically indistinguishable from fine details and without knowing the original it's hard to tell whether some things are details or noise.

[killerteengohan]

Problem with removing such compression artifacts is that they are basically indistinguishable from fine details and without knowing the original it's hard to tell whether some things are details or noise.

I don't think they are compression artifacts for this series. They are the same in the DVD's, TV Recordings, and Blu-Rays. All of those use different bitrates and compression methods, yet all of those still have them identically shaped and positioned, while the other real compression differences are obvious. Normally I would agree they look like compression artifacts, but they are in literally every legit sold by company source of this I can find and they don't move around at all while everything else does. I think they were intended to be in there and put over the animation on purpose. Possibly the simulated brush strokes or noise Jagabo was talking about.

[Selur]

Looks to me like what could happen if you compress dithered content, but even if those are not compression artifacts the are still partially y indistinguishable from fine details.