Dot crawl! Which filter to remove??

[lamer_de]

Thing is, i want to re-encode "The Simpsons" SVCD TV caps to XVID to fit 4 Eps per CD instead of 2. But unfortunately the source clips show annoying dot crawl artifacts. These 2 pics should show it, but aren't the best as they are screenshots and can't demonstrate the effect of pixels waving forth and back.

http://members.tripod.co.uk/forumpic/chair.png [notice the chairs]
http://members.tripod.co.uk/forumpic/marge.png [here it crawls between marges hair and her bandeau]

[Side note: Why is the img tag disabled?]

I think some people already encountered this problem before, cause i found some threads about it in the forum. Unfortunateley, i wasn't able to find a satisfying solution for this problem :p

Some people suggested to use a CNR filter, but cnr2 had no visible effect. May be a noob question, but how can i display the croma noise it removes (you know, these grey pictures like here: http://engel.otaku.ru/lab/dvd/dvdq_3.html)?

Ermac suggested here (http://forum.doom9.org/showthread.php?s=&threadid=33277&highlight=dot+crawl) to use Warpsharp, but if the crawling was gone, the picture was over sharpened and introduced sharpening artifacts. He said something about a Wavel Noise reduction, but i couldn't find a filter which does it. Any hints on this?

Maybe i mixed things up, i don't have the theoretical background (c++ knowledge, mathematics in general etc) to understand how these filters work, so maybe i'm totally wrong. In this case, please point me to the right direction ;)

Any help is appreciated,
lamer_de

[zettai]

Actually... I'd use warpsharp to fix this.

Alternatively you might want to try the Wavelet Noise Reducer.

[High Speed Dubb]

That’s clearly in need of comb filtering. I’ll be porting the DScaler one over reasonably soon.

[zettai]

Doh.. I'm so sorry... I must have just accidentally skipped the hwole paragraph on warpsharp and wnr.. doh!

The dscaler comb filter would probably work wonders. The wavelet noise filter is linked on avisytnh.org - or try searching on this forum for it.

[lamer_de]

OK, i found a combination of warpsharp and wnr rather effective, still not perfect. Everytime i removed the crawling, i had also removed a quite amount of details :(

That’s clearly in need of comb filtering. I’ll be porting the DScaler one over reasonably soon.
Looking forward to this :) The faster, the better ;) Seriously, even most anime ripping groups have these problems, have a look at the P-A/AA encodes of Soultaker or AA's X TV. Pretty annoying effect.

As written before, i'm not stuying computer science, but i'm nonetheless interested in how this effect comes into existence, so if sb could try to explain it to me, i would be thankful.

I'll post some screenshots/clips tomorrow.

Thanks for your attention,
lamer_de

[High Speed Dubb]

Originally posted by lamer_de
...i'm not stuying computer science, but i'm nonetheless interested in how this effect comes into existence, so if sb could try to explain it to me, i would be thankful.

Computer science stuff isn’t too relevant to it. This is more an issue of signal processing.

It’s a side effect of the way colors are encoded in PAL and NTSC broadcasts. Originally in black and white TV, the only signal was the Y (luminance) information. To add colors, a “carrier frequency” was added to the signal. This is a constant frequency (3.579545 mHz for NTSC) wave. The color (U and V) information was encoded as changes to the amplitude of this carrier frequency.

So to get the separate Y and UV data, you just remove the 3.57... mHz wave to get the Y, and use the amplitude to get U and V.

But... What happens if the Y happens to be about that same frequency? Then it gets misinterpreted as color information, leading to “rainbows.”

And, what happens if the color change is so great that some of the information from the carrier wave is misinterpreted as Y information? Then it appears as a luminance pattern which corresponds to the carrier frequency, and appears as “dot crawl.”