Decomb's Telecide() is causing strange flashing around edges

[Daner]

I am new to working with the Telecide() function of the Decomb package, but I think I have a general understanding of IVTC. My problem is that I'm getting a strange pulsating flash occurring around the edges of things, as a result of using Telecide().

My source is an NTSC DVD. It is interlaced with top-field first. Since it is a TV show, I assumed that I would want to keep the framerate at 29.97fps, rather than decimating it to 23.976fps. Is this a correct assumption?

Anyway, here is a clip (http://www.netdepot.org/personal/sample-interlaced.avi) of the interlaced source video. This was produced with the single line:

mpeg2source("FILE.d2v")

Now, here is what happens (http://www.netdepot.org/personal/sample-telecide-30fps.avi) when using the Telecide() function without changing framerate, as done with the following code:

mpeg2source("FILE.d2v")
Telecide(order=1,guide=1)

And finally, here is a clip (http://www.netdepot.org/personal/sample-telecide-24fps.avi) of using the Telecide() function in conjunction with Decimate(), to cut it down to 23.976fps. You can see that the exact same issue happens.

mpeg2source("FILE.d2v")
Telecide(order=1,guide=1).Decimate()

What is causing this flashing around the white text? How can I fix this?

Thanks for your help!

[scharfis_brain]

The flashing is caused by the postprocessing.

You said, it is a TV-Show?
then telecide/decimate are useless here.

you need a deinterlacer.

try tdeint.

[Daner]

Thanks for your reply!

Yeah, I kind of figured IVTC was inapplicable for TV source, but what I liked about it was the fact that it didn't seem to be blending fields together, an ugly effect that I have noticed most other deinterlacing methods tend to create.

[scharfis_brain]

which MOST (?!?) other deinterlace methode are you talking about?

I only know about fielddeinterlace() being a blender by default.

all others

tomsmocomp
(leak)kerneldeint
tdeint

aren't doing any blending

[Daner]

OK, here's what I mean. Take a look at this screenshot. Everytime there is motion, there is this ugly checkerboard ghost.

http://www.netdepot.org/personal/tdeint-artifacts.jpg

Note that this was done using the TDeint filter you just recommended, with the interpolation type set to "modified ELA" which is supposed to reduce "jaggies".

[KaiserS]

Originally posted by scharfis_brain
You said, it is a TV-Show?
then telecide/decimate are useless here.

No its not. There are plenty of TV shows that are telecined from film so doing an IVTC is perfectly fine and this show is one of them. If this show were shot pure interlaced then I'd agree but its not.

Originally posted by Daner
What is causing this flashing around the white text? How can I fix this?

In the bottom two clips I see combing happening but that is due to the bad matching not post-processing. Try turning guide to 0 in telecide to see if it gets rid of the combing and if that doesn't work try something like TIVTC (http://bengal.missouri.edu/~kes25c/TIVTCv098.zip) and see if you get better results (at least on the clip you posted I got no combing at all).

[Daner]

OK, I have tried out TIVTC, with this code:
mpeg2source("FILE.d2v")
TFM(d2v="FILE.d2v")
TDecimate(mode=1)
It fixed the combing issue around the white text, but not the checkerboard ghosting issue shown in the screenshot I posted above.

Upon further investigation, I discovered that the DVD source itself is interlaced like this! Certain frames appear checkerboard-interlaced rather than line-interlaced. Does that make any sense? Is that typical? You can take a copy of this short sample clip (http://www.netdepot.org/personal/untouched-source.avi) from the untouched interlaced source to see what I mean. If you are able to de-interlace this clip without any combing or ghosting happening, please let me know how you did it!

[neuron2]

I'd help you if you gave me the source file in something other than WMV3! How about HUFYUV?

[KaiserS]

Another problem is there is alot of dot crawl in the source which I believe is what is causing the problems you are seeing. The source of these DVDs is very low quality which makes doing good quality encodes of them quite hard.

Originally posted by neuron2
I'd help you if you gave me the source file in something other than WMV3! How about HUFYUV?

Or a snippet of the vob would be even better. :)

[Daner]

Originally posted by neuron2
I'd help you if you gave me the source file in something other than WMV3! How about HUFYUV?

The sample clip I just gave is HuffYUV ;)
Here's the link again:
http://www.netdepot.org/personal/untouched-source.avi
*EDIT* And here it is as a VOB:
http://www.netdepot.org/personal/untouched-source.vob

I've been reading up on Dot Crawl, and it looks like this may very well be the problem at hand. The question is why there would be dot crawl on a DVD...I mean this isn't an amateur TV capture, so why would there be chroma crosstalk happening?

But if this is indeed the problem, I guess I'm looking at some sort of de-interlacing & de-dotcrawling procedure. Would I use an IVTC followed by a GuavaComb to do that?

[KaiserS]

Originally posted by Daner
I've been reading up on Dot Crawl, and it looks like this may very well be the problem at hand. The question is why there would be dot crawl on a DVD...I mean this isn't an amateur TV capture, so why would there be chroma crosstalk happening?

It means they used the composite TV masters rather then properly sourcing it from the original materials. If they had done a proper sourcing for the DVDs there wouldn't be this problem. Basically they cheaped out and the quality of the DVDs suffered.

Originally posted by Daner
But if this is indeed the problem, I guess I'm looking at some sort of de-interlacing & de-dotcrawling procedure. Would I use an IVTC followed by a GuavaComb to do that?

Yep. You could also give DeDot a shot as I think it works better on Dot crawl then GuavaComb and is less prone to artifacts.

[wolfsoft]

Originally posted by scharfis_brain
which MOST (?!?) other deinterlace methode are you talking about?

I only know about fielddeinterlace() being a blender by default.

all others

tomsmocomp
(leak)kerneldeint
tdeint

aren't doing any blending

I think the effect of blend in AviUtl is more better. It's a pity that there aren't any good filter which likes aviutl's blend for avisynth.

[scharfis_brain]

I think the effect of blend in AviUtl is more better. It's a pity that there aren't any good filter which likes aviutl's blend for avisynth.

You talk about this setting:

Setting -> Deinterlace -> Blend

don't you?

this is exactly the same like doing

blur(0,1)

with the only difference, that it is faster ;)

[neuron2]

The VOB is in fact 3:2 pulled down (hard telecine). The artifacts are in the source and are not caused by Telecide. That does not look like dot crawl.

[Daner]

Originally posted by neuron2
The VOB is in fact 3:2 pulled down (hard telecine). The artifacts are in the source and are not caused by Telecide. That does not look like dot crawl.

If it's not dot crawl, then what is it? How do you get rid of it?

[neuron2]

It looks like the ghosting you get from standard field blending except that the ghosts are checkerboarded:

http://neuron2.net/misc/ghost.jpg

How they managed to make checkerboards, I have no idea. But it is likely that it would render the unblending approaches ineffective. Some things are just too screwed up to fix.

[wolfsoft]

Originally posted by scharfis_brain

this is exactly the same like doing

blur(0,1)

with the only difference, that it is faster ;)

Thanks. But in my watching it still more sharp than aviutl. Blur and FieldDeinterlace would be better for movie, but not good for anime.

[scharfis_brain]

uhm.

blur(0,1).sharpen(0,1)

[Leak]

Originally posted by neuron2
How they managed to make checkerboards, I have no idea.

Maybe they captured the images off a TFT monitor - mine only updates every other pixel in such a checkerboard pattern every frame... :)

[Daner]

Originally posted by Leak
Maybe they captured the images off a TFT monitor - mine only updates every other pixel in such a checkerboard pattern every frame... :)

Very strange. I'm considering writing an AviSynth plugin to fix this very weird and very specific issue.

I want to clarify something though: Would such a filter be applied AFTER an IVTC or BEFORE? I'm guessing after...

[neuron2]

If you're referring to the checkerboard ghosts, they are present as field blends. Therefore, you should apply the filter on the field stream.

[Daner]

Originally posted by neuron2
If you're referring to the checkerboard ghosts, they are present as field blends. Therefore, you should apply the filter on the field stream.

I'm a little bit confused by what you said...why would there be field blends happening during IVTC?

[manono]

They're not being created during, or as a result of, IVTC. They're in the source. Do a SeparateFields or some sort of a Smart Bob to confirm. Your filter should be applied to the fields.