Analyzing interlaced frames

[rjc7394]

What's the best method to determine how many frames are interlaced? I mapped frames using avisynth decomb filters and imported into Vdub. I found 1/5 frames that I think are interlaced. Is this a good method or is there a better way?

[Guest]

What do you mean by "you mapped frames"?

And why would you only "think" a frame is interlaced?

What exactly are you trying to accomplish?

[mrslacker]

I'll let neuron2 field this with his expertise, but I'm curious... why do you think a certain proportion of frames would be interlaced for your source. The answer might be an important piece of info.

To answer the question blindly, I would tell you to use any decent deint filter with show=true, etc. and an appropriate detection threshold. Watching combed frames be detected in various material was (is) a rather educational exercise for me. Hope I'm not talking below your level of understanding.

[rjc7394]

I ran it in DGIndex and it said the video (NTSC) was interlaced.
So then I imported into Vdub with Decomb filter and chose map=true (fieldinterlace) so I could look at the combed areas: 1/5 frames were combed. Then I tried it with kerneldeint and it looked like all the frames were combed. (I used default values for threshold and dthreshold for each filter). I wanted to know if there was a more precise method.

[Guest]

OK, you tried three methods, none of which is fully reliable! Don't worry, I'll tell you the reliable way in a minute, but first a little explanation about the three:

1. DGIndex: It just reports the progressive_frame flag in the MPEG2 syntax. It is often not encoded correctly and so cannot be relied upon.

2. FieldDeinterlace: This one looks for vertical detail at the spatial frequency of interlacing (combing). That means real image detail can look like interlacing. But if you get the threshold right and look for the combing only in moving areas, this can be reasonably reliable.

3. KernelDeint: This uses differences between successive frames. For technical reasons which I can't go into here, this too is not fully reliable.

Now, here is the best way: Open the video with field operation NONE and no other filtering, and serve it into VirtualDub. Single step through the video and look for combing artifacts in non-static scenes. Yes, your own eyes are the best instrument for analyzing interlaced frames!

BTW, telecining generally produces at least two combed frames per cycle. If you get 1 in 5, that may be field blended.

[rjc7394]

Call me stupid but what's the syntax for this? I can't find where the "field operation" field is located in the script. Couldn't locate it on your site.
I just imported the .d2v file straight into Vdubmod and looked at it and saw that 1/5 frames are definitely combed.
This would benefit from decombing? I've deinterlaced (fielddeint, kerndeint, smartdeint) about a dozen times (w/different parameters) and results are marginal w/some resulting "jaggies" around edges. Thanks.
Also wonder if global motion compensation and deinterlacing at the same time poses a problem? Two processes analyzing/correcting motion at the same time seems like the wrong road to take: more problems could be created.

[Guest]

Field operation is a DGIndex menu option accessible via the Video pulldown.

There are many sources of combing: interlacing, telecining, field blending, field phase shift, chroma offset, field distortions, etc. You need to determine the cause of the combing to decide how to treat it. You'll need to study this area for some time before being able to decide how a specific clip needs to be treated. I wish it were easy, but it just isn't. :(