An unusual kind of LQ-video artifacting. Need advice.

[Zaxooz]

Hi all,

I am a newb when it comes to all the smoothing/denoising filter varieties there are, but basically, this is what I am seeing:

http://i.picpar.com/4Wp.jpg

The image already has MSU Smart Deblocking 0.8 and MSU Denoiser 2.5.1 with the slider pulled all the way to "Harder" applied (though I think the denoiser isn't doing much good and slows the encode quite a bit).

Any advice on how to get rid of that patterned noise?

[paradoxical]

If you want help, you'll need to post more than just a recompressed JPEG. Post a short stream sample showing off the problem.

[LoRd_MuldeR]

Also please explain what exactly is your source, where you got it from and what you are trying to do with it.

[Zaxooz]

The video is a cutscene from a late 90s pc game - C&C tiberian sun. The resolution is 640x400 and it's running at 15fps. The framerate I don't care about that much, but the compression artifacts I'd like to get rid of. I've extracted the uncompressed video stream, cut it, ran it through the QTGMC InputType=2 (that by itself helped smoothen it out a bit). Now I intend to process it with VideoEnhancer bringing the resolution up to 1024x640 and applying the MSU Smart Deblocking plugin in the process. All of this still leaves me with quite a bit of compression artifacting and I didn't really like median filter's results. Any suggestions? Thanks.

Sample (x264 encoded from uncompressed, prior to any operations): 1a (1)-001.mkv (http://wikisend.com/download/132030/1a (1)-001.mkv)

[ChiDragon]

Download link isn't working for me.

The artifacts in your screenshot are in fact caused by Video Enhancer, based on my experience with that software. Do you see the problem with posting a processed image instead of a source sample?

[Zaxooz]

Those artifacts are native to source and the image is 99% representative of what you see in the actual video.

[ChiDragon]

Download worked with IE.

Post #1 screenshot (http://i.picpar.com/4Wp.jpg)
x264 encode sample + Spline36Resize (http://s21.postimg.org/92hg5tjif/Tiberian_Spline36.png)

For all I know the difference was only induced by the JPEG compression of your original image, but I vehemently disagree that it represents the nature of the artifacts in the video. They are typical of CD-ROM video from the era. Good luck; perhaps someone knows some good tricks for this.

[LoRd_MuldeR]

I don't think a deinterlacer like QTGMC will help here, as the video is not interlaced. Nor does it show the type of artifacts that result from a "bad" deinterlace.

As there is some apparent banding, I would try something like gradfun2db(). And to get rid of those vertical lines, you could try something like:
TurnLeft()
NNEDI3(field=2)
Merge(Last.SelectEven(), Last.SelectOdd())
TurnRight()

See:
http://screenshotcomparison.com/comparison/28141

[Zaxooz]

Thanks a lot Mulder. One more noobish question: is this what banding in a progressive source is or you were referring to something else?

http://i.picpar.com/Beq.png

[LoRd_MuldeR]

That last screenshot looks like the typical artifact you see when displaying interlaced video on a progressive screen ("combing").

However your screenshot also looks like it already has been resized without being deinterlaced first! If that's the case and you can't go back to the original, then that's a lost case :(

See also: http://100fps.com/

[Zaxooz]

This is the original in my case. QTGMC Inputtype=2 does help temper it down a bit though, especially fast movement scenes where the residual artifacting is not as easy to notice

http://i.picpar.com/tfq.png

http://i.picpar.com/ufq.png

(shifts gamma for some reason?)

edit

Could you please explain what that NNEDI script of yours does? It does help with those lines somewhat, while at the same time making it look a lot like VHS (which isn't bad in this case and considering the material :-)). Thanks.

gradfun2db, strangely, hardly makes any difference, the size goes up by 0.2mb on that small sample clip, quality wise however it looks even slightly worse to my eye.

[LoRd_MuldeR]

Could you please explain what that NNEDI script of yours does? It does help with those lines somewhat, while at the same time making it look a lot like VHS (which isn't bad in this case and considering the material :-)). Thanks.

It interpolates a full frame from the odd columns of the frame and a full frame from the even columns of the frame, then blends both together.

It's like a sledgehammer method, but definitely helps with the vertical lines. Do the same without the TurnLeft() and TurnRight() to get the same effect in horizontal direction.

Maybe one of the filter gurus could come up with something more sophisticated, e.g. based on the GeneralConvolution (http://avisynth.org/mediawiki/GeneralConvolution) filter...

gradfun2db, strangely, hardly makes any difference, the size goes up by 0.2mb on that small sample clip, quality wise however it looks even slightly worse to my eye.

Be sure you apply that last, so the dither isn't filtered away in the next steps. Also try to play with the strength.

Note: gradfun2db() has a tendency not to survive lossy compression :(