Avisynth: How to Limit Posterization?

[FredThompson]

I'm testing the optimal script on a B&W film source. Under magnification, the processed result shows some posterization and is a little harsh.

Granted, the source is B&W so the human eye is more discerning then it would be for color.

Can anyone tell me which part of these lines should be investigated to soften the transitions within the frame?

asharp(2, 4)
STMedianFilter(8, 32, 0, 0 )
MergeChroma(blur(MaxTreshold))
MergeLuma(blur(0.2))

ScriptClip("nf = YDifferenceToNext()"+chr(13)+ "nf >= 2 ? unfilter( -(fmin(round(nf)*2, 100)), -(fmin(round(nf)*2, 100)) ) : TemporalSoften(4, round(1/nf) , round(3/nf) ,0, 2) ")

There is a sample image at the bottom of this page: http://www.geocities.com/fredthompson6

[ovg64]

Hi Fred, from the look of the pics. your are using a low Resolution, the MA Optimal Script is optimise for Resolutions 528x and above; you best ask the expert Kwag as for what chainges to make on the MAS line.

[FredThompson]

All the screen grabs are 720x480 png from VHS source. I do all my captures at that resolution.

I hadn't thought about the resolution of VHS tape. That's a good point, there's only about 1/4 the true information you'd expect from a proper D1 source.

[kwag]

That's not posterization. I can clearly see interlaced artifacts

-kwag

[jorel]

me too!


another observation:
deinterlace comparision: smartsmoother vs DG decomb.
is my impression or decomb loose the dimention and focus?

[FredThompson]

yes, I know it's interlaced. Guess I should have been more explicit. IMHO, filtering should be done before decombing/deinterlacing for a number of technical reasons. Forget the fact it's interlaced for right now, look at the transition of shade across Jose's face and the curls of his hair.

Interestingly, the dropout near his hairline has become more predominant. Kind of bolsters my thoughts about scratch/dropout removal being first, then general cleaning.

wrt decomb, the latest beta is processing this capture properly, the 4.x release isn't. If you look at DG's journal you'll see how the sensing algorithm is developing. There was a bug in a release a few days ago but it was caught. Don't know what version you're using.

Page is updated now with progressive screengrabs.

[FredThompson]

I've just updated the site to include the script and some new samples. When you look at the color sample, you'll see something isn't working properly in the KVCD-based part. It seems to be confused by the white edge.

And...I'm too tired to stay up and mess with it any more tonight. Sample DV clips are there for the screen grabs.

[kwag]

Hi Fred,

When you can, try commenting the lines asharp(1, 4) and MergeLuma(blur(0.2))
Se how that affects your encode.

-kwag

[FredThompson]

I will. Saw a note you posted to somebody else about over-blurring a little while ago. Just had an impulse and tried the filter on some of my torture clips. WWE Smackdown! has an incredible amount of motion. The sample I used has a huge mix of changes on every field. Well, I'm truly stunned. You should see what this is like on a per-frame basis. Hopefully, I can get an FTP server up tomorrow and put some samples up. I've been waiting because there's a bunch of stuff to clean and encode to free up space. The filter appears to be removing some macro blocks, be they in the digital satellite source or DV. It's not perfect but, maybe, 2/3 are gone, even in super-high motion mixed CG and live action stuff. I've NOT had time to play with Tom's DCTFilter other than to determine the default settings are too strong. Same with MSharpen and MSmooth. You'll get an idea of the torture clips when the page comes back up.

It's really weird. I ran some stuff through and didn't even realize how much noise there is with S-Video to DV. I'm stunned, literally.

Regarding changing things in the filter, yeah, some of the constants will be changed to variables so it's easier to modify. I envision lines at the top where Joe Sixpack uncomments, say, B&W VHS and the script then executes with a fairly tuned group of settings to support that. If he's doing MPEG-2 from digital satellite, there's another line, and so on.

Macro block removal will never be perfect but there might be room for improvement. I had a good scene with light-colored walls but it's gone. The satellite source had that blockiness from low bandwidth. None of my test files have that so I couldn't test.

The noise removal for DV camcorder was quite impressive. There's an incense burner in the test file and some of that smoke is filtered out. I don't think there's a way around it because smoke is always going to have some aspects that are on the border between noise and signal.

[FredThompson]

Your recent changes to the optimal script have removed this posterization of B&W source. Thanks.

[kwag]

Quote:

Originally Posted by FredThompson

Your recent changes to the optimal script have removed this posterization of B&W source. Thanks.

-kwag