Here is a short sample. It's from 1993 from a VHS-to-DVD conversion and I don't have the original tape. Normally for stuff from 1996 and onwards I just use the following script which I think is enough do remove all the noise and deblock (it looks good enough to me):
Code:Mpeg2Source("M:\New\z = Encode\Dec 13 test.d2v", CPU=6) RemoveSpots() McTemporalDenoise(settings="medium", interlaced=true) AssumeTFF() QTGMC(Preset="Super Fast") SeparateFields() SelectEvery(4,0,3) Weave()
However this 1993 footage has more problems that the above script does not fix. I don't know how to describe it but it's got white flutter around the edge of the wrestlers and other problems.
I have an 8 core PC so whatever slow filters you feel I need to use, I should be able to use them now.
+ Reply to Thread
Results 1 to 11 of 11
Sanlyn you said this in an older thread about this video but nobody replied, I guess I should have written a better thread title.
Your standard script does lower the bar for interlaced VHS captures, but the clip submitted needs even more than MCTemporalDenoise. The "flutter" is lots of things, none of which your script would address: severe aliasing (to the level I've heard referred to as "sparking"), interlace/combing artifacts, and other horrors caused by poor VCR playback. But I know what you mean, this is what you have to work and there's not much choice: you can relieve those jaggies and the foggy clouds of chroma noise and macroblocks with repeated runs of fast temporal smoothers. By the time most of those disturbances are cleaned, you won't have much video left. Most of the visual data is just noise.
Yes, sorry, it's Dehalo_alpha(). You're gonna have to play around with the settings a lot.
Start with DeHalo_Alpha's default settings by simply typing "Dehalo_alpha()". The defaults work much of the time, but look over their documentation. There aren't many parameters anyway.
The video you reference has worse than halo on edges. Again, the bitrate is too low to control the extreme fast-action scenes. The original is also a copy of a copy; at some point the tape was played with heavy over-sharpening, which causes almost all the noise and artifacts you see, as well as making demands on the stingy bitrate that the encoder couldn't keep up with it. There is also a nasty banding problem in the dark background caused by serious dark crush at some point in earlier processing.
I recall this video from a while back. Here's a script I tried on it some time ago (warning: it's slow as hell):
Mpeg2Source(videopath+"edges.d2v",CPU=4) ConvertToYUY2(interlaced=true) AssumeTFF().SeparateFields() Checkmate(24) FixVHSOversharpL(30,12,8) FixVHSOversharp(30,14,10) TemporalSoften(2,4,8,15,2) TemporalSoften(3,4,8,15,2) Weave() ConvertToYV12(interlaced=true) MCTemporalDenoise(settings="medium",sigma=4,AA=true,edgeclean=true,stabilize=true,\ TTstr=2,enhance=true,interlaced=true) gradfun2dbmod(thr=3,str=0.8,mask=false) LSFMod(strength=100) Crop(6,0,-16,-10).AddBorders(10,4,12,6)
Last edited by sanlyn; 23rd Mar 2014 at 11:33.
Thanks that helps a lot. I also used DeHalo_alpha() to get rid of the white outlines around people but I can still see some. I read the documentation but I'm still not sure what setting I should be using to get rid of it all.
it's preferable to use lossless settings with qtgmc when you deal with old analog recordings in my experience, the difference is huge with a capital h
QTGMC( Preset="medium", SourceMatch=3, Lossless=2, Sharpness=0.4, TR2=2)
You don't have to use assumetff/bff for mpeg sources you can actually change the field order doing so (= bad)
Last edited by themaster1; 18th Oct 2012 at 12:50.
What lossless settings are you referring to?
Some of that halo effect is due to arena lighting. Bright spotlights and floods create edge-lighting effects. Oversharpening exaggerates those effects. It's also somewhat of an optical illusion: the oversharpening produced coarse, grainy shadows that are in sharp contrast to brighter portions, and soft gradations between edges are wiped out.
The area of halo should be covered OK with deHalo_Alpha at default settings. The width of the halo is treated by the "rx" setting. I believe the default is 2.0. You can try 3.0 ("DeHalo_Alpha(rx=3.0)"), but it will have little effect and the script slowdown will be painful. The statement could be used in the sample script by inserting it just before running GradFun2DBmod.
But the noise around some of those edges is also due to faulty pre-processing and the low bitrate. The only fix is to oversmooth, but the kind of edge-enhancement used to encode the mpg precludes that. Think about it: high-contrast edges look "sharp" but you can already see inner features such as noses and lips disappearing. More smoothers would soon blur motion (it's already blurred in several frames) and people would look as if they're wearing stockings over their heads during a 7/11 robbery. People think they gain something by oversharpening VHS, but it inflicts irreparable damage once it gets to lossy encoding. As the word "lossy" implies, data that once defined smooth, fine features is discarded. The result is broken, sawtooth edges. The data that once described finer details of smooth edges, gradations and motion is gone. That's one reason why shadows in the clip look ugly and rather stark.
In the trial script you see two instances of TemporalSoften, Replace the statement TemporalSoften(2,4,8,15) with SpatialSoften(3,4,8).
Last edited by sanlyn; 23rd Mar 2014 at 11:33.
In QTGMC, the "lossless" settings more or less means "keep the noise". I already tried QTGMC when I first saw this clip. It did little for the halo or the "sparks". Stronger smoothing with stronger settings of MCTemporalDenoise had a small effect, but faces started to disappear. That hard edge-enhancement coupled with noisy tape playback and low bitrate is just murder.
Last edited by sanlyn; 23rd Mar 2014 at 11:34.