VCR composite output, 3D comb filter, remove baked-in artifact?
 

Responding to my friend's bold claim from another thread. :cool:

Attachment 14045
Attachment 14046

Attachment 14047
Attachment 14048

Workflow:
Mitsubishi HV-BS890 (-DNR, +629TBC, Std, Cshift Std) <YC> EVAL-ADV7842-7511 (7.5IRE) <HDMI> splitter <HDMI> AVerMedia C027
Mitsubishi HV-BS890 (-DNR, +629TBC, Std, Cshift Std) <CVBS> EVAL-ADV7842-7511 (7.5IRE) <HDMI> splitter <HDMI> AVerMedia C027

The EVAL board is the most configurable thing I've ever touched, so it's safe to say I'm not using optimal settings. Even so, the 3D comb filter successfully removed dot crawl and rainbowing recorded into this VHS tape.

The VCR menu settings (Japanese) are attached, along with the equivalent vhs-decode screenshots to further prove that these artifacts are part of the tape.

Thanks man. Notice the obvious difference? the Y/C version is sharper than the composite, That VCR could use a line TBC, there is a significant line shift between the two fields commonly known as combing or mice teeth, Less noticeable in the blured composite version. Dot crawl in Y/C? Why? Possibe faulty Y/C board.

A single frame only tells a sentence of a story, but ostensibly here I'm inclined to agree with you.

There is a lot of, erm, how shall I put this politely, dogma on here. You either agree with a certain person(s), or you're wrong, there's no room for experimentation...

It's always interesting to see experiments with different techniques, they may be pleasing or no, but it's always interesting to see.

I guess I always assumed that a comb filter only operates on information present in the analog signal that's lost upon conversion to digital, but seeing it remove the baked-in dot crawl suggests otherwise.

Is it possible to make that EVAL board comb-filter the Y/C image as it does the CVBS image? That would be an apples and apples comparison. Otherwise, remove the board from the chain and do it in software.

And I agree that a good TBC would be useful also.

The ADV7842 chips is "flawed" -- not really flawed, but near-impossible to peg down settings that don't make video worse or better+worse -- so I'd want to see this comparison with it removed from the equation.

I can do this too, I have cheap crappy s-video cables that add dot crawl. And then you get use known-good devices to process the composite cleaner-than-normal (normal = dot crawl, image invaded by fleas).

@Rubust: There is no "dogma" (at least not with us), but there are general rules, and exceptions to the rules. (The problem with exceptions is that too many people think an exception exists when it does not.) You're only wrong when you give information that is counter to facts. (Too often, disagreement in video are from limited experiences with one party, usually those working with limited sources.)

Yes, msgohan has time to do some interesting comparisons, from time to time. Always appreciated.

We'll have to disagree on that.

What gets more lost is the chroma part of the signal, as a video decoder chip will decode the two (quadrature amplitude) modulated color channels to digital component. I.e what you get out is an Y value, and two chroma values for each pixel. In theory one could modulate that back to the form it's in in a composite or Y/C, though in practice it will probably differ a bit from the original input.

A big thing Y/C filtering for NTSC relies on is that the modulated color signal is inverted every line (for NTSC, PAL is more complicated). You can see this on the dot crawl in the sample too, it's why the crosstalk looks like dots rather than vertical lines. This means that if you average two adjacent lines the color signal of the two lines cancels each other out to an extent. In practice it's a bit more complicated than that of course as the colors change, tape output won't be stable, sharp details can look like color etc. The ADV chip has a 3D Y/C filter so can look at multiple frames to improve filtering, while many devices only has a 2D one which is limited to analyzing a few lines at a time.

Still it's possible to do some of this filtering even on an already digitized signal. For PAL in particular we've seen some interesting results with ld-decode and vhs-decode with the BBC Transform decoder being able to get rid of color artifacts normal decoders can't. (The principle behind it doesn't work for NTSC.)

With Y/C filtering there will always be some trade-off between detail vs cross-color artifacts, a good filter will do a better job at retaining detail while removing artifacts, with VHS there isn't much detail in the area where the signals overlap in the first place (VHS is a bit above 3 mhz, NTSC color is centered at 3.54 Mhz), so one may be able to filter it a bit more without loosing much. Less so with a higher-bandwidth signal like from off air or SVHS/Hi8 (which is why those formats added S-Video output). It's hard to tell with this sample whether there is any loss of sharpness/detail or not, but it does at least show that it's possible to reduce baked-in Y/C artifacts by running it through composite and an advanced Y/C filter.

TLDR: yes you can do some comb-filtering even on a digitized signal.

Pretty sure the VCR has a line tbc, msgohan notes that the TBC in the deck is enabled. The TBC may help Y/C filter a bit as it makes sure the lines line up, and the color signals will line up too if they did when it was recorded.

This time I'll use a different VCR, Mitsubishi HS-HD2000U, to address the possibility that the other VCR was somehow faulty. (The vhs-decode images already demonstrated that dot crawl is in the source, but let's continue to cover our bases, I guess.) As with the previous comparison, I've enabled the VCR's line TBC. Unlike the other model, this one couples the TBC and DNR features together.

Let's establish that my S-Video cables are not introducing checkerboarding. I've attached images showing DVD player loopthrough.

Since Lordsmurf wanted to see different equipment, I'm including the Panasonic DMR-ES15 as 3D comb filter, captured using ATI 600 USB.

Note for the image comparisons: right-click, Open in New Tab, flip back and forth to compare (use CTRL and + to zoom). If you only look at them on this page, there are a lot of subtle details to miss.

EVAL: The cleanup of his shoelaces and the Oscar is great. But while the rainbowing is removed from "Steven", the letters look deformed.
Attachment 14061
Attachment 14062

DMR-ES15: The text is rendered clearly, but there is residual rainbowing in all three areas.
Attachment 14063
Attachment 14064

EVAL
Attachment 14065
Attachment 14066

DMR-ES15
Attachment 14067
Attachment 14068

AVIs attached, compressed with Lagarith. The original filenames pictured below show the workflows. I had to shorten the filenames for the forum. No screenshots of the "Director of Photography" video, but I've attached the script if you want to step through frame-by-frame yourself.
Attachment 14071

Maybe you could also upload the original png images to https://imgsli.com and paste here the link to have an additional comparison with a cursor. Sometimes it helps.

Thanks for the details on this topic, very interesting!

While your VCR could have problems with the Y/C stage due to possible bad capacitors or you are using a S-Video cable with no shielded core, The Y/C stills show great amount of details, The appearance of those artifacts being removed is not what it actually is, the lack of details in the CVBS version just blurs them out, They are still there when you look in the video. Stills don't tell the whole story.

A proper external TBC may fix that rainbowing.

I doubt an external TBC would have any effect on this unless it featured some noise reduction function - the datavideo and cypress ones do not to that. (Especially if fed via composite as they have much less advanced Y/C filters than the ADV EVAL board and ES15) The chroma signal will have already been "stabilized" by the TBC in the VCR.

There results without a vcr tbc involved could be quite different as then more factors come into play, instability will often make A/D chips fall back to a simpler Y/C separation filter. (Ironically a very dumb notch filter that just scoops out anything in the color frequency area could actually give less artifacts due to the low bandwidth of VHS not really allowing much detail in the frequency band the color data info lives in as opposed to super VHS or output from a game console or something.)

The EVAL seems to do a better job than the panasonic on getting rid of the dot crawl, especially noticeable on Myer's suit, can't really tell to what extent there is detail loss if any.

What msgohan refers to as "rainbowing" looks to me like the same alternating red/green pattern seen on Pooh's tummy in this post that was cleared up by using a Cypress CMD-1500 TBC from LordSmurf. The VCR is a Panasonic AG-1980P with its own TBC turned on.

1. This is a different VCR, so they'd both have to be broken in the same way.
2. I specifically covered how the S-Video cables don't introduce this, with DVD test patterns run through this VCR. Additionally, the cable used for the EVAL captures is Blue Jeans Cable YC-2 which is touted on the forums as top-of-the-line. (It's also used as part of the chain for the DMR-ES15, but isn't the weakest link.)

Where are you seeing "great amount of details" lost? How do you blur things like the rainbowed shoelaces to reveal individual diagonal lines?

Attachment 14082 Attachment 14081

The BV10 has a comb filter, but remembering that you don't want any random BV10 *as it has flawed chipsets as well, just like Cypress). I'll test it in the near future.

I also have something else I can test it with, interested to see what it does. :hmm:

I see a post where LordSmurf said "Cypress CDM-1500 (green AVT-8710 clone without proc amp)".

I was going to hook up my green AVT-8710 today anyway to demonstrate a separate issue, so here you go:
Attachment 14083
Attachment 14084

https://imgsli.com/Njk1Njg
https://imgsli.com/Njk1Njk

The chroma streaks in your AG-1980 images look like my faulty AG-1980 units. Rainbows and dot crawl are different. They wouldn't appear on large areas of flat color like the bear's belly.

The CDM-1500 is not a 1:1 clone, but more like a "clone" because it is slightly more advanced. Sans proc amp. So some aspects cannot be compared directly.

I can add it to me comp tests, when I get some time.

Thanks for testing with your AVT-8710. It's disappointing that it didn't clear up the rainbowing.

At the time I posted my images, my AG-1980P had just been refurbished by T.Grant, so I'd like to think that it wasn't faulty, but anything is possible.

Also, maybe the rainbowing in both cases was caused by the tape itself being old, and a faulty VCR makes it worse.

Ah, I suspect that has more to do with using composite than it being a TBC as such. The Y/C filter will average out the chroma to a degree depending between adjacent depending on how it's set up, so maybe the behaviour differs a bit with whatever capture card you were using. There is normally an internal comb filter adding in some portion of the previous line's chroma to the current one in VCRs as well (though in this deck there will be some further chrona NR), but maybe the tape had a lot of chroma errors or there was a VCR issue or something.

Regardless of opinions, just having such clear A/B playback comparisons of the same recording is very helpful. It allows people to make up their own minds on the advantage or not of the two different types of playback. I would appreciate such comparison photos at this resolution being included in the Digital Video Guide.