Decomb Show Option Chaning the Output

[blazerqb11]

I certainly could be wrong, but reading the guide it seemed to me that the show option was there to help you set other parameters and didn't itself affect the actual de-interlacing. If that is the case then I am experiencing something pretty odd.

The following two pictures show the same frame with the given options:

Telecide(guide=0,post=2,vthresh=15,show=true)
Decimate()
http://img223.imageshack.us/img223/469/showtrue.png

Telecide(guide=0,post=2,vthresh=15)
Decimate()
http://img16.imageshack.us/img16/4315/showfalse.png

In the first picture his mouth is looks good, where as in the second picture his mouth shows evidence of problems, but what really confuses me is the fact that they look different at all.

Here is a sample video, for anyone interested.

http://www.megaupload.com/?d=R76H2JIU

This same issue can be found on frame 538(after Telecide(), Decimate()) of the sample.)

[neuron2]

It can change what frames are deleted by Decimate() because the video is altered going into Decimate().

Let me look at the clip and report back.

[Leak]

Well, you're giving Decimate() two different images, and it happens to handle the one with the progressive text stamped on it differently than the combed one...

No surprises there - in your first case it's also matching the (uncombed) text and thus comes to a different decision than in your second image...

np: Robert Henke - Metropol (Atom/Document)

[neuron2]

That's one of those "mouth torture" clips. I don't know of any way to handle it rather than manual overrides.

Also, you should crop the black bars and closed captioning noise before field matching.

[blazerqb11]

Also, you should crop the black bars and closed captioning noise before field matching.

I had wondered about that. How much do I want to crop, just everything that is completely black?

As for the closed captioning noise, I assume I should always crop that, even with a 4:3 source? For example, I should crop out two lines from the following source:

http://img402.imageshack.us/img402/9079/closedcaptioning.png

How exactly does that work? On a channel with 528X480 resolution, is the original video scaled to 528X478, or is the captioning data layered on top of the first two lines?

[TheRyuu]

I don't know of any way to handle it rather than manual overrides.

tfm is generally more resilient when it comes to mouth combing so you can maybe give that a try.

I would generally always try tfm().tdecimate() first before decomb but obviously both are made to accomplish the same task. I'm not saying decomb doesn't have its uses just TIVTC will normally provide the better matches the majority of the time.

Edit:
Upon looking at the sample, it looks like it may be some weird R1 vfr (i.e. section 2886)

[blazerqb11]

I tried TIVTC and it worked great. Can anyone answer my previous question about cropping?

[thewebchat]

Crop out anything that's not important. However, for YV12, you can only crop in multiples of 4 vertically or you will break the field order. I also don't see what relation aspect ratio has to cropping.

[blazerqb11]

I also don't see what relation aspect ratio has to cropping.

I was curious about how only 2 lines of closed captioning would make a difference.

What I'm really curious about is my other question in that post about how the closed captioning data is mixed with the video.

[neuron2]

When you have very small mouth movements, the CC data, if present in the video seen by the field matcher, can contribute more than the mouth movement to the matching algorithm and thus cause false matches.

In the analog signal the CC data is contained in line 21, which is normally in the overscan and so not visible. How did you capture this video?

[blazerqb11]

How did you capture this video?

I captured the transport stream of a clear digital cable source with a USB QAM tuner.