Early South Park is driving me crazy

[Katie Boundary]

Seriously, what's with all the checkerboard noise? Why does it exist? Why is the image apparently composed of four fields instead of two? THE ABYSS IS GAZING INTO ME AND I AM BECOMING A MONSTER

https://scontent-sjc2-1.xx.fbcdn.net/v/t1.0-9/19248010_1879337885663998_5614890898892981729_n.jpg?oh=fe94a42e7de6a820ca676aabd8c1da17&oe=59DA67DB

[Sharc]

Is this from a tape capture? "Composite" analog signal? Is the pattern static or does it crawl when you play the video?

[Katie Boundary]

It's from the DVD. This artifact appears in the place of ghosting, in frames that would otherwise be ghosted. In this particular case, the yellow checkerboarding is from the school bus in the next frame...

https://scontent.fsnc1-2.fna.fbcdn.net/v/t1.0-9/19424334_1879776758953444_8194143594897999912_n.jpg?oh=7ba09d0060563f6b9833bf35f73142cf&oe=59C6469D

...and here, you can see red checkerboarding from where the guy's shirt used to be.

I kind of expected that everyone here would already have the South Park DVDs and would have discussed this problem a long time ago.

[johnmeyer]

I'm 95% certain that the problem is dot crawl.

If you are playing this DVD on your TV set, and the DVD player is connected to the TV set via an analog composite cable (the RCA, round yellow connector), and you then captured this image, then this is almost certainly dot crawl. It certainly has the usual dot crawl attributes where the severity of the checkerboard changes depending on the hue and intensity of the color. Here is a simple explanation of dot crawl:

Dot Crawl (https://en.wikipedia.org/wiki/Dot_crawl)

On the other hand, if you are playing the DVD on your computer (which is what I suspect is the case), then no such mechanism exists, and this image is, I presume, captured directly from your computer's media player. If that is the case, then the problem is "baked into the DVD."

Is this a commercial DVD, or did this come from some online capture, or from some other source? A lot of "shady" DVDs from China, Thailand, and other countries without any copyright laws are often created in the dumbest ways imaginable, as are many online videos posted on YouTube or on various torrent sites. It is possible that someone had a videotape of this episode, connected their capture equipment via a composite connection, encoded the result, and ended up with this sorry mess.

I had a similar problem and started a thread in this forum where I asked for help. Here is that thread:

Remove cross hatch pattern from Kinescope (https://forum.doom9.org/showthread.php?t=173030)

The solutions posted should work for your video as well. One of them uses convolution kernels, something I know you are familiar with. I got the best results using one of the convolution matrices.

Since the scripts in that thread are nothing more than customized blurring functions, the fact that your video still retains its colors (mine had been converted to B&W) shouldn't matter, and the scripts should function just as well for you as they did for me.

[johnmeyer]

A quick search at AVS Forum and the web in general does seem to suggest that the early South Park episodes -- even the official DVDs -- have this problem. Wow, that stinks. I guess the fixup route is the only way to go.

[Katie Boundary]

It is possible that someone had a videotape of this episode, connected their capture equipment via a composite connection, encoded the result, and ended up with this sorry mess.

I suspect that this is how the official DVDs were created in the first place. Evangelion got the same treatment, but suffered capture static at the bottom of the screen instead of checkerboarding.

The first and second seasons both got re-releases in 2004. I sent Warner (the distributor) an e-mail asking if the reissues cleaned up this noise. No response yet.

EDIT: for future reference, all of my workflows are 100% digital for the simple reason that I do not own any analog equipment other than a Sega Dreamcast, and although I own a few VHS tapes, the last time I watched any of them was back when "Friends" was still on the air, *NSYNC was still together, and invading Iraq seemed like a good idea.

[Sharc]

DotCrawl was my first thought as well, but can it be THAT bad??
After a second thought I am afraid 'yes', for example when the monkey who captured the video tape cross-wired the Composite signal to the S-Video input of the capturing device, hence bypassing the comb filter which would suppress the dot crawl effect to some degree (depending on the quality of the comb filter). Secondly, I suspect that -- after seeing the damage -- the same monkey applied some strong temporal smoothing which eventually produced the ghosts.....
It can be difficult to fix dot crawl properly. There are some avisynth tools available, some are good for NTSC or PAL only (you have to know how it was captured), and are more or less effective.
I would recommend to try the brute force method first for this ugly case: Downscale to half width and upscale to original width. It will smooth the picture a bit but the loss of details may be acceptable (depending on the true resolution of the original).

[Katie Boundary]

I would recommend to try the brute force method first for this ugly case: Downscale to half width and upscale to original width. It will smooth the picture a bit but the loss of details may be acceptable (depending on the true resolution of the original).

I tried something similar involving separatefields() and blur(1.0), and ended up with scanline-like artifacts similar to the ones in Babylon 5. It was still an improvement.

[johnmeyer]

DotCrawl was my first thought as well, but can it be THAT bad??
After a second thought I am afraid 'yes', for example when the monkey who captured the video tape cross-wired the Composite signal to the S-Video input of the capturing device, hence bypassing the comb filter which would suppress the dot crawl effect to some degree (depending on the quality of the comb filter). Secondly, I suspect that -- after seeing the damage -- the same monkey applied some strong temporal smoothing which eventually produced the ghosts.....
It can be difficult to fix dot crawl properly. There are some avisynth tools available, some are good for NTSC or PAL only (you have to know how it was captured), and are more or less effective.
I would recommend to try the brute force method first for this ugly case: Downscale to half width and upscale to original width. It will smooth the picture a bit but the loss of details may be acceptable (depending on the true resolution of the original).I think it may be worse because of something Katie mentioned in her first post, namely that she is seeing four fields instead of two. I don't know exactly what they means, and I don't know if she meant that is true all the time, or whether it comes and goes. However, since it is animation, it likely has a really weird cadence with lots of duplicated fields, but not following a consistent pattern. I have no idea how that would affect the mechanism which produces dot crawl, but I suspect it might produce additional artifacts beyond the usual cross-hatching.

Without an actual clip, I don't know what the ghosting actually looks like and whether it might be reduced or eliminated by clever selection of "good" fields (or frames) which are duplicates of bad fields (or frames). Once the best of each duplicate is kept and all other duplicates eliminated, you could then try to re-do some sort of pulldown in order to keep the speed correct. Since it is animation, the audio can be allowed to go off by a few frames during this process, and as long as the final video is the right length, the audio will be close enough that no one will notice.

[raffriff42]

Your might benefit from Spectral adaptive denoising with dfttest (https://forum.doom9.org/showthread.php?t=173038). Maybe related, see Remove cross hatch pattern from Kinescope (https://forum.doom9.org/showthread.php?t=173030).

Please disregard if these suggestions are insufficiently ghetto :)

[Katie Boundary]

Your might benefit from Spectral adaptive denoising with dfttest (https://forum.doom9.org/showthread.php?t=173038). Maybe related, see Remove cross hatch pattern from Kinescope (https://forum.doom9.org/showthread.php?t=173030).

Please disregard if these suggestions are insufficiently ghetto :)

Cool suggestions but I'm mostly interested in what's causing the noise, not how to fix it. Most forms of noise disappear when you scrunch the resolution down to sufficiently ghetto VCD standards :)

[johnmeyer]

Cool suggestions but I'm mostly interested in what's causing the noise, not how to fix it.Why do you keep asking when I already gave you the answer (and the correct answer, I might add), weeks ago? To repeat myself: it is caused by dot crawl when the people who transferred the video to DVD used composite video as the source. Whether they did this because they were stupid, lazy, or just couldn't get the original source material, I don't know.

I should also note that this problem with the first few episodes is well-documented on the web, so you can easily find more information with a simple Google search. It took me less than thirty seconds ...

Here is one discussion you may find interesting:

Hideous dot crawl (on early South Park DVDs) (https://forum.videohelp.com/threads/342027-Hideous-dot-crawl)

[edit]Here's another:

The trademark South Park dot crawl artefact was present at times on this DVD (http://www.michaeldvd.com.au/Reviews/Reviews.asp?ReviewID=83)

and another

South Park Dot Crawl (http://www.avsforum.com/forum/37-video-processors/664487-bob-weave-motion-adaptive-deinterlacing-one-has-best-quality.html#post7403217)

[hello_hello]

Why do you keep asking when I already gave you the answer (and the correct answer, I might add), weeks ago?

Aren't you on the ignore list due to your inability to comprehend plain English? ;)

[johnmeyer]

Aren't you on the ignore list due to your inability to comprehend plain English? ;)Maybe that is it. We'll see. However, I'm working on my Russian so I can converse in her native language.

[Katie Boundary]

I was under the impression that "dot crawl" would resemble, you know, little dots crawling around, like I've seen on Buffy the Vampire Slayer. But maybe you're right and "dot crawl" is just some weird euphemism for "checkerboard ghosting", and the artifact that resembles little dots crawling around on the screen is actually called something else.

I'm working on my Russian so I can converse in her native language.

If you can find me a LEGITIMATE copy of the Пружинки album by Маффин, and not one of the millions of bootlegs, I will personally come to your house and service you sexually.

[johnmeyer]

You are absolutely correct: "dot crawl" makes no sense when you look at a freeze frame and see the checkerboard pattern. However, on live video (and perhaps animation as well), when you get strong transitions between colors, you do in fact see dots or "ants" crawling along those boundaries (pun not intended), depending on the movement. I think that is where the term originated.

This video shows some pathological case of dot crawl:

Dot Crawl Example Video (https://www.youtube.com/watch?v=QN4KY6QCFak)

You said you are not interested at this time in fixing it, but if you do eventually want to do something about it, there are several approaches that work pretty well. I'll post the links if & when there is interest.

[Katie Boundary]

Interesting.

So what's the name for the artifact that occurs when something that should be moving appears semi-stationary instead, but with weird crawly dots, because the colors are too similar and the source was low-quality to begin with so MPEG's motion-search gets all screwed with? For example, the walls of Sunnydale High School in early Buffy DVDs?

[johnmeyer]

Interesting.

So what's the name for the artifact that occurs when something that should be moving appears semi-stationary instead, but with weird crawly dots, because the colors are too similar and the source was low-quality to begin with so MPEG's motion-search gets all screwed with? For example, the walls of Sunnydale High School in early Buffy DVDs?I'm not sure about that one. Someone else will have to chime in.

[jmac698]

The 'crawl' motion is caused by two checkerboard patterns alternating, and our eyes tune into this as motion, either up or down. If you look at dot crawl on colour bars, you can amuse yourself by trying to see the dots at the edges of the bars moving up or down.

When dots are stationary, it's called hanging dots or Hannover bars, however that effect is usually seen on PAL material.

There is a special pattern to your sample, it's clear that there's dot crawl only where the fields were different. That would point to two fields mixing somehow, like some kind of field blending due to a rate conversion, a resize, or temporal smoothing. Maybe if I stepped through a sample and thoughtt about it enough, I could guess what kind of likely process lead to it.

And yes, I still call it dot crawl because there's no more precise term for it. It's not inaccurate, it's just a frozen dot crawl.. however the term implies the source of it, which is accurate.

As far as fixing it, brute force would be way too much. I think maybe you could subtract it out perfectly by blending the colour from the last field maybe..

[Katie Boundary]

If you look at dot crawl on colour bars, you can amuse yourself by trying to see the dots at the edges of the bars moving up or down.

Is this at all related to the moving checkerboard patterns that sometimes appear at the edges of the black overscan bars on fullscreen material?

[jmac698]

I'm not aware of any other source of a checkboard pattern, than dot crawl.

I think you're referring to a letterboxed video. I have one theory, that the comb filtering has nothing to compare the line above it and so can't filter the top line. Does it appear on the bottom edge as well?

I'll also mention a little experiment I did. You can perfectly cancel dot crawl if you have two fields of the same content. The two fields come from points in time where the colour carrier is 180 out of phase. That means if you average them, you get perfect dot crawl removal. This happens naturally in a movie with the pulldown pattern. So every (1/3? I forget these details now) frame can be perfect. Based on that, I had the idea that it could somehow be a guide to remove the dot crawl in the remaining frame, as being some kind of temporal anchor to interpolate against.

Did you read the explanation of dot crawl? I think I'll summarize it here for the benefit of anyone who runs across this thread.

TV was originally monochrome. They wanted to add colour in a backwards-compatible way. They did this by adding a reference short "beep" (called the colour-burst) at the left border. This beep has a frequency of 3.58MHz (NTSC) (PAL came later and used a different method). Then the line is mixed with that frequency of varying phase along the visible line. This shows up as a dotted line pattern in the line. It gets turned into colour, by comparing the "beep" reference phase with the phase along the line. Where they are in phase, it's red, and as they get out of phase, it goes through the colours of the rainbow.

Further, each frame, the entire phase of this process inverts, same with each line. Therefore, if you average two frames together, assuming the image didn't change, you would exactly cancel the checkboard pattern. And if two lines are exactly the same (such as a simple-resized 2x vertically picture) you can average them to remove the dot crawl.

The checkboard pattern is actually a sinewave on one line and a sinewave 180 out of phase on the next line... you could say the edges of the checkboard "squares" smoothly blur along the edges.

They are most visible at the edges of strong colour boundaries because of the inability of techniques to filter out the pattern while interpreting it as colour.

To finish my story in history, they added colour by adding a slight screendoor pattern to the whole picture on older TV's. Newer ones could just filter out that frequency and the picture would look perfect again (for newer monochrome TV's).

[wonkey_monkey]

Further, each frame, the entire phase of this process inverts, same with each line.

Isn't that only the case with PAL? Phase Alternating Line... or is that something else.

[Katie Boundary]

I think you're referring to a letterboxed video.

I'm not. I'm referring to the extremely narrow black bars at the right and left sides of the screen in some fullscreen material, which exist due to a phenomenon called overscan (https://en.wikipedia.org/wiki/Overscan).

[wonkey_monkey]

It's not overscan, it's nominal analogue blanking. That they are never usually seen is because of overscan, but that's not why they exist. But then I'm blocked, so you'll never learn this.

[hello_hello]

I'm not aware of any other source of a checkboard pattern, than dot crawl.

I think you're referring to a letterboxed video. I have one theory, that the comb filtering has nothing to compare the line above it and so can't filter the top line. Does it appear on the bottom edge as well?

I pretty sure I know what she means. It's a vertical pattern one side (or both sides) of the image, at the very edge of the picture, and my memory seems to be insisting it's more often on the right side (at least for PAL).
The "pattern" can vaguely resemble a tiny strip of film at the edge of the picture, several pixels wide, where one "film frame" is picture and the next is black, or something along those lines, and it can appear to strobe or scroll when the video is playing.

That's the best description I can manage, but I've been curious as to what causes it myself. I think I've only ever seen it when the original source was video, and therefore on 4:3 DVDs. I might try to dig out a sample later and post a screenshot.

[Blue_MiSfit]

South Park was mastered on D3 tapes for a long time which definitely suffered from severe dot crawl. They had no money at first :)

[Katie Boundary]

South Park was mastered on D3 tapes for a long time which definitely suffered from severe dot crawl. They had no money at first :)

D3 tapes? Is that like Betamax but worse?

[nevcairiel]

D3 tapes are actually digital tapes, compared to the analog Betamax or VHS tapes its still quite good quality, but still a "cheap" solution for mastering.

[Blue_MiSfit]

Yep :)

Uncompressed digital composite. Totally crazy, but it filled a niche for awhile and was cheaper than D1 / D2.

https://en.wikipedia.org/wiki/D-3_(video)

It was entirely deprecated by Digital Betacam (DigiBeta) which is still in wide use today for SD tape masters.

DigiBeta introduced a perceptually lossless ~2:1 DCT compression which enabled full 4:2:2 10 bit at ~90 Mbps which was totally doable with inexpensive 1/2" tape and transports.

D1 needed 3/4" tape to store totally uncompressed component video (tho only in 8 bits per sample)

D5 later introduced uncompressed 4:2:2 10 bit on 1/2" tape, and was eventually upgraded to support 4:1 perceptually lossless HD :)

[raffriff42]

It was D-2 (https://en.wikipedia.org/wiki/D-2_(video)#Acceptance) actually. Quality-wise, D-2 was the same as D-3 (https://en.wikipedia.org/wiki/D-3_(video)): as good as PAL/NTSC ever gets, which by today's standards, is not saying much. The Making of South Park (http://wayback.archive.org/web/20140110104734/http://www.spscriptorium.com:80/SPinfo/MakingOfSouthPark.htm)
The animated shots are transferred to a digital disk recorder and from there to an Avid video editing station where Matt & Trey work with that show's editor to work out the details (the fastest way to get the images into the Avid is as a video signal, so the DDR is used to generate video from the rendered image files). If a shot needs to be changed, it gets sent back to an animator or TD for a 're-take' and then re-rendered. At the same time the video editing is done, the final audio is worked out on the audio Avids and merged into the video Avid's content. Once everything meets the stringent South Park quality standards (i.e. we run out of time), it gets written to a D2 digital video tape and sent off to Comedy Central to be aired.

EDIT Digital Betacam was component, which does not have dot crawl obviously - but required PAL/NTSC video facilities to completely rewire, buy new video peripherals etc, which cost millions. Hence D-2 and D-3: digital recording (no generation loss) while keeping your old video gear.