should I de-interlace 480x480 vidcaps destined for SVCD?

[miamicanes]

I know that SVCD is 29.97 frames per second progressive-scan and gets turned into 59.94 fields per second by the DVD player at display time (presumably, showing the odd lines of frame 1, then the even lines of frame 1, then the odd lines of frame 2, and so on...).

What I'm not entirely sure about is how I should best capture the signal from videotape prior to mastering and burning the SVCD.

To give a concrete example, if I use VirtualDub with the Morgan MJPEG codec to do the initial capture at 480 x 480 (with the intention of letting TMPGEnc chew on the captured .avi file overnight and turn it into an ultra-efficiently compressed MPEG that makes maximum use of B and P frames to squeeze the most detail out of the bit bandwidth available to it), should I have VirtualDub de-interlace it at capture time?

In theory, it seems as though the ideal capture scheme for SVCD would take the 59.94 fields per second and combine fields 1 and 2 (odd and even scanlines) to make frame 1, fields 3 and 4 (the next round of odd and even scanlines) to make frame 2, and so on.

The resulting videocapture would look like absolute shit when viewed using Windows Media Player or on a progressive-scan DVD (because the even scanlines would be temporally offset by 1/60th of a second from the odd scanlines), but would be exactly right when displayed on a normal, interlaced NTSC TV. The DVD player would render the frame's odd scanlines as a field, then render the frame's even scanlines as another field 1/60th of a second later. Because the even scanlines were, in fact, captured 1/60th of a second after the odd scanlines and simply rolled into the odd scanlines as a pseudo-progressive 29.97fps frame, everything would work out properly in the end.

On the other hand, it seems as though intentionally de-interlacing the 59.94 fields per second to create a "true" 29.94 frame per second progressive source would ultimately render WORSE when viewed on a normal TV from a normal VCR (but look better viewed on a computer or progressive video display), because the act of de-interlacing would introduce new artifacts (mainly ghostly blur) that would look right in progressive, but basically be the worst of both worlds when re-interlaced because information that had originally been "cleanly" associated with one scanline or another would now be mushed and mixed together.

Now, the big question is, "Is that right?"

The bigger question is, "How, exactly, do I tell VirtualDub to create a 29.97 frame per second pseudo-progressive capture using sequential odd and even scanlines, making NO effort to deinterlace or otherwise make sense of them. Just take the odd, the even, mesh them together, and save them as one field. Don't double the odd or even, or try averaging them, or anything?

Once again, the video is being captured from videotape for burning to a SVCD; how it looks on a computer or other progressive display is irrelevant -- only its appearance on a normal TV matters.

[Bradskey]

I'm not an interlace guru either, but I think you've answered your own question a bit. I never bother de-interlacing, because I eventually author everything to some kind of disk format that will be played on a DVD player to a standard TV set. If, on the other hand, you were making say DivX or some other format to store on your PC or swap on the Internet you might want to deinterlace it. Not sure about de-interlace in Vdub (if it can be done I'm sure its simple to find), but you should be able to have your MPEG encoder handle the de-interlacing.

Go to http://www.lukesvideo.com and read what he has to say about intelacing.

[fortension]

I think you're right, and I read something about not deinterlacing for dvd (and also therefore for svcd). Capture at 480 vertical resolution and you'll get an interlaced capture, and deinterlacing will result in some of the deinterlacing artifacting - jumpiness and/or ghosting. So leave it alone... Go Bucks.

[miamicanes]

Well, I'm trying to run an experiment at the moment (encoding the same 15 minute video from 640 x 480 interlaced MJPEG source to VCD, 1800bit/sec VBR SVCD, and normal SVCD to compare...

... except TMPGenc keeps randomly freezing my computer at different points while encoding (no discernible pattern or spot. It seems to be entirely random). Grrrrrrrrr. If it could at least continue where it left off I wouldn't mind so much, but I just had freeze #4 and it's getting REALLY old...

[Sulik]

There ar epros and copns about deinterlacing, and depending if the source is FILM or not (FILM is already progressive).

Keeping interlaced:
Pros: smoother motion (50/60Hz vs 25/30Hz). Shouldn't produce interlacing artifacts even on a progressive display, as the display should deinterlace on the fly if properly encoded (progressive sequence=0)

Cons: Interlaced material requires more bitrate for the same quality, so you'll have more compression artifacts than if the sequence was deinterlaced.

Deinterlace:
Pros: Requires less bitrate (higher quality), but not as smooth motion. If the source is FILM, it may be better to deinterlace than capturing interlaced (assuming you have a decent adaptive deinterlacer), or even better using inverse telecine for NTSC.

Cons: For video content, you now have only 25/30Hz video, so motion will not be very smooth (FILM look). Especially visible on stock tickers and smooth camera pans.