Optimal codec settings for SVCD from VHS capture?

[miamicanes]

This afternoon, it occurred to me that seems to be a gross mismatch between BOTH VCD and SVCD's default parameters when it comes to burning one from video captured from a VHS videotape.

As I understand it, VHS videotape has normal NTSC/PAL vertical resolution (480/520 lines), but only 120-160 horizontal "pixels" per line(for a worn out/old tape, maybe as high as 240 for a pristine tape recorded under ideal conditions from a flawless video source on quality tape), with half again as much real color resolution -- suggesting that the "ideal" capture res would probably be something like 160 x 480 or 240 x 480, depending upon how good the recording is to begin with.

The reason why I suspect that 480 x 480 isn't optimal is because it seems like at THAT resolution, it's just using up its "bit budget" to accurately store noise, possibly at the expense of sacrificing color information or overall resolution during a transition where it would actually make a difference. Put another way, it seems like it's too busy rendering this frame's noise to worry about conveying the next scene's info via P frames, and ultimately causes the first few frames of a new scene to be more pixellated than they really need to be had it put its bit budget to better use a few moments earlier. Or, perhaps if the program rendering the MPEG *knew* that "detail" higher than 160 pixels horizontally was just noise, it could factor out the noise better and do an equally effective job with 1.6kbit/sec as it could with 2.2 or 2.3kbit/sec -- leaving room on the SVCD for more video content.

Ultimately, it seems like making a VCD from a VHS capture means throwing away half of the detail information that IS present (the vertical detail), while squandering up to half the bit budget rendering horizontal "detail" that's just noise. On the other hand, making a SVCD from a VHS source retains the vertical detail, but squanders up to half the disc's potential recording time on detail that's just noise.

Am I totally off base, or is there a major mismatch between both VCD and SVCD and VHS-sourced video? If I'm somewhat on the right track about SVCD defaults burning bitwidth on noise, *is* there a way to tell the MPEG-2 codec, "be aggressive with respect to the horizontal detail, because most of it's just an illusion anyway, but do your best to preserve the vertical detail"?

[Zubr]

Well, you're not entirely right - i suggest reading this

http://forum.doom9.org/showthread.php?s=&threadid=34122.

So for capture needed resolution of SVCD/DVD , and for storage either use SVCD or CVD.
Cheers!

[SatStorm]

The same story all over again...
And from the same people also...

Anyway, with no techical talking, the best analogue capture from VHS/SVHS/TV is 384 X 576/480. Because 384 horizontal lines ain't close enough anything digital frormat, capturing direct to 352 X 576/480 helps.
Some users strongly believe that capturing @ 480 X 576/480 gives you something. I don't see it, but who am I to judge their results? So I mention them. For me this is overkill.

The convertion now, is the hard thing: For VHS the target resolution meant to be about VCD's one (352 X 288/240) but there is one problem: If you use VCD, you don't have interlace output. VHS is interlace. So, you loose something. The only solution is go -x- and use the same resolution with vcd but with mpeg 2. That in theory makes identical VHS files. BUT:
For many reasons (including encoder limitations, hardware limitations, Cable limitations, technology limitations in generall, etc) you don't have 100% of VHS with this resolution (352 X 288/240). Personally, if I rise the bitrate (about 1500 average) and use some filters, I have results identical the source, but you can see that only with good mpeg 2 decoders (hardware/software). But this is me, most people won't bother or test 2 years for the subject, they want easy and fast solutions. So, what you can do?
1 - Live with it. Use 352 X 288/240 as target and that's it. Try to make it look better over the years...
2 - Overkill so to end up with something equal. You waste filesize and discs but it is way better and the target file gonna look virtual identical the source.
Most users choose solution 2

The easiest way to convert VHS to digital mpeg, is using mpeg2, with an average of 3000Kb/s, using multipass VBR. There are times even 3500Kb/s are neccessary. That way, you have perfect picture.
The hard way is to convert VHS to 352 X 288/240. With an average bitrate about 1600Kb/s and using interlace mpeg 2 as a target (going -x- for CDs that is, but still compatible witn DVD), using filters/technics and wise the encoder, you may have the same results (as the theory points). I have made it. But I spent years trying to. Why to bother? Go xCVD/xSVCD/half D1 DVD and that's it...

[miamicanes]

Well, one thing I discovered -- ATI's VCD capture might be realtime, but it's far from optimal.

I did an experiment where I created a VCD and a SVCD from the same VHS source material with different approaches to creating the mpeg files.

The source:

sA1 = VCD preset realtime capture using ATI Media Center.

sA2 = MPEG-2 480 x 480 @ 29.97fps capture using ATI Media Center

sMJPEG = MJPEG (Morgan Multimedia) 480 x 480 @ 29.97 using VirtualDub at 95% quality, saved as AVI.

The VCD mpegs:

vA1 = sA1, converted from stream encoding by Nero but otherwise unchanged

vA2 = sA2, converted to VCD-compliant MPEG-1 (352 x 240, 29.97fps, CBR) using TMPGenc

vMJPEG = sMJPEG, converted to VCD-compliant MPEG-1 using TMPGenc

The SVCD mpegs:

svA2 = sA2, converted from stream encoding by Nero, but otherwise unchanged

svMJPEGcbr = sMJPEG, converted to SVCD-compliant CBR MPEG-2 by TMPGenc

svMJPEGvbr1800 = sMJPEG, converted to SVCD-compliant VBR MPEG-2 with 1800 bit/second average and 2500 (whatever the defalt was) max

svMJPEGvbr2200 = s3, same as previous, but 2200 bit/second average bitrate

The Observations

vA1 looked like shit.

svA2 didn't look a whole lot better.

vA2 looked a tiny bit better than v1, and a tiny bit worse than svA2.

svMJPEGvbr1800 generally looked worse than svA2, and my DVD player had a hell of a time trying to fast-forward and rewind through scenes (probably due to the relative lack of I-frames). It pixellated badly during scene changes or movement, but *did* sharpen up rapidly when the motion calmed down and occasionally looked better than svA2 did at comparable spots.

svMJPEGvbr2200 looked a lot better than svA2

vMJPEG looked almost as good as svMJPEGvbr2200

The Verdict

VCDs made by capturing the video using VirtualDub and the Morgan MJPEG codec (approx. 10-12 gigs/hour @ 95% quality), then rendering to VCD-compliant MPEG-1 using TMPGenc (with deinterlace: odd/even field, adaptation) looked ALMOST as good as VBR SVCDs with average bitrates slightly below the 2500-something max, and looked several orders of magnitude better than BOTH mpeg-1 AND mpeg-2 captures straight from ATI media center.

On the other hand, those same captures took approximately 4-5 times as long to achieve -- the time to digitize (realtime), the times to re-digitize (because ***** VirtualDub won't stop you from accidentally anihilating your previous capture file, and apparently has no preferences setting to autoincrement EVERY time you press f6. I think I accidentally nuked about half my captures that way. Grrrr...), the time to render to MPEG-1 from MJPEG/AVI (1.25 to 2 times the playing time on an Athlon 1800XP w/512 megs). However, in my case it was far, far worse, because for some reason TMPGenc would freeze the system hard about 50% of the time within 20-30 minutes of rendering, and about 70% of the time within an hour (so much for leaving it overnight), and VirtualDub froze a few times during captures too (the overlay video still worked, but the rest of Windows was frozen hard and no further capturing was taking place).

Vented Rage

I'm reinstalling Windows XP over Christmas, so I'm hoping that will fix the problem. If it doesn't, I'm probably going to break it (ATI All in Wonder 1280 pro) in half in a rage, because historically, the ONLY time it's EVER worked reliably was the 3 month window of time in early 2001 when ATI finally released stable Win2k drivers, but before I wiped the drive and installed Windows XP (under which the AIW 128 pro has never worked reliably... doing anything related to MPEG or DirectX while working on other (unsaved) documents is roughly akin to playing russian roulette, with slightly worse odds

If I do end up getting a new card, it sure as hell won't be an ATI. I've lost just about all respect I've ever had for them after getting burned. Their attitude seems to be, "we'll pick one specific version of Windows to support, and make half-assed drivers for everything else". At least with the BT-8xx cards, there seems to be enough independent driver support to make the commitment of the capture card's actual manufacturer irrelevant in the long run (2-3 years after purchase).