More video per VCD ?

[MITSUI_1]

A modest proposal:

I was thinking of ways to put a lot more minutes of decent VCD audio and video onto a single VIDEO-CD. I came up with the following analysis, which I post here for scrutiny and comment. It seems perfect for archiving old VHS tapes and for other “meh” quality projects.

Before I begin, note that I am nowhere promising that you’re going to get super results. I think this is just as good as regular VCD, but it’s certainly not perfect. I have had the approach work but your milage may vary. Because a fundamental assumption is that your going to IVTC your video, please be aware that it only works with “film” sources and then, only on DVD players that can handle a 23.97 fps stream. (Not all of them do so gracefully).

First things first. On any VCD conversion project (even a 2 or 3 disk set), you must ask yourself: Is it a movie where high quality video and superb stereo sound really matters? If so, spend $15 to buy the real damn DVD disc and you’ll be happy. Really. The movie studios are offering a very high quality product for good prices nowadays. You won’t have pixel crawl. You won't have macro blocks. You won't have compression artifacts. You’ll have full DTS sound, and it’ll last damn near foever. Compatibility is guaranteed with every DVD player.

So if you REALLY want DVD quality and features, go buy a DVD! Really!

But if you’re just archiving old VHS tapes, why not use some video bandwidth conservation tricks and a more conservative audio stream to fit almost 120 minutes on one xVCD?

Lets start with regular VCD basics: The format itself uses 352 x 240 resolution, which equals 84480 pixels per frame. That means your encoder needs to figure out what to do with 2,531,865 pixels for each second of video at 29.97 FPS NTSC.

Lets trim some of that bandwidth. First, inverse telecine to 23.97 FPS. When your done, your still left with 84480 pixels per frame, but multipled by 23.97 gives us only 2,024,985 pixels to encode per second of video which is about 20% less.

Now, since standard VCD bandwidth is allocated on a bits per-second basis, you actually have better quality at the same bandwidth, or a situation where you need 20% less bandwidth to achieve the same quality as before. We’ll go for the latter because that means we can put 20% more video on one xVCD.

But we're not done yet. Next, lets look at putting a border on the screen. Yup. A border. What I’m talking about here is a black border 16 pixels wide on all four sides of the screen, making the actual viewable frame about 320 x 208. In TMPGEnc, you do this by setting the encoder to encode at 352x240 but in the layout, choosing “Center: Custom” and having your source placed on the 352x240 frame at 320x208. I'm not explaining this very well, but you'll figure it out.

What's that? I know, I know, you’re bitching and complaining about the borders. I might as well hock warez on my web page and hang out with Shizzzzonn... (meh). Really, though, this is a huge help, and a significant proportion of the black border will actually be outside of the viewable area of your TV in any event. When I play a clip set up in this manner on my 27" TV, I can see about a 1/4 inch black strip on the top and bottom, but the picture goes right flush to the left and right sides. That’s hardly a big problem, and its not noticeable unless you’re looking. Think about this: Why spend good bits to encode what you cannot see?

Importantly, this trick is not worth doing if you’re not willing to use at least 16 pixels worth of border all around. As I understand it, MPEG 1 compression operates on 16 x 16 macroblocks, and you only wind up with ultra high compression for all of the black border area if it corresponds with the macroblock boderers. (Otherwise, the MPEG encoder has to waste bandwidth encoding a sharp edge in the middle of a macro block, which MPEG does not do so well and which will actually pull bits from the rest of your frame to handle). Anyway, when you put 16 pixel borders all around, that gives about 65,660 pixels per screen that basically don’t have to be encoded (they’re super compressible and take almost no space in the stream). Thats about 429,542 pixels less per second to handle, which we can allocate to increasing the length of our xVCD.

Obviously different rules apply to anamorphic and widescreen films, which already have a large chunk of black on the top and the bottom. You still need to resize and configure TMPGenc in such a way as to put the top and bottom borders in line with a MPEG macroblock encoding unit boundary. Anyway, suffice it to say that if you’re encoding wide screen anamorphic stuff, you already basically have the benefit of this trick (and then some!).

Well, where are we? After applying both tricks, we’ve now got got 1,595,443 pixels of actual picture data to encode for each second of video as opposed to the original 2,531,868. That’s a savings of about 37%, but not really. We're actually saving something slightly less than that because, as suggested above, even the data to do a fully black 16x16 block takes some tiny part of an MPEG 1 stream.

From a pure mathematical standpoint, if you get this right you save almost a million pixels per second of data, which you can then use to fit more on a single sVCD.

Now, we need to do something about the audio.

The system stream on a “by-the-book” VCD is 1150 video and 224 audio for a total of 1374 kbps of data. That means about 16.3 percent of the data on a standard VCD MPEG stream is audio. So, if we cut that we’ll be able to free up even more cd space. How?

How about going to an audio stream of 128 kbps? That’s a 50% savings on encoding data per second, which gives us another 8 % total CD savings. Or even better, how about 80kbps? That’s a 65% percent savings on sound data per second, which gives us 10% more space on the CD!

Again, if you want theatre quality surround sound, don’t bother with this. Just go purchase the damn DVD at BestBuy and be done with it. I'm going on the premise that you need to archive some old VHS tapes and the audio quality really isn’t so important there.

At lower bitrates, TMPGenc’s lower quality audio encoding will be more apparent. Therefore, use use toolame.exe to convert your audio data to MPEG audio at our new bitrate, and MUX it with your video stream later.

Where are we.. We’ve now reduced our encoding requirements by 20 percent by doing IVTC, another 17 percent by doing the borders, and saved another another 10 percent of disk space by switching audio to 80kbs That means we can encode with roughly the same visual quality, but putting 47% more footage on one cd. Take an 80 minute cd, add 47% and you’re talking another 37 minutes. For a total of 117 minutes. Which gets pretty damn close to the 120 minute holy grail.

One final trick. Since we’re now going to fit all of the data onto one xVcd, encode the credits as a separate track! Encode them with even less bps sound and with only 200 or so kbps of data. (Do a test disk first to see what your DVD player can handle).

To implement all of this, you need to set your MPEG video encoder to use less data per frame to account for the video savings. Theoretically, that means set it for 37% lower bitrate than the 1150 default. I'm still working on optimum settings and trying different approaches with TMPGenc (i.e., CBR, 2Pass VBR, CQ_VBR, etc).

What are your thoughts?

[MITSUI_1]

Sorry about the weird characters in the post. I copied and pasted my notes from Wordpad and the apostrophe and quotation marks wound up getting trashed.

I'd be really interested in seeing what you guys think about the issues I describe in the post, though. Thanks.

[House de Kris]

My thoughts are that you did a good job of justifying and explaining all of your tradeoffs to reach your goal of greater time on a VCD. Thank you for staying in reality and not claiming this will give DVD quality.

[JHebert]

You better watch out or a certain Spanish guy is going to come back here claiming somebody stole his "CVCD" format again!!!

The ideas of using lower than standard audio bitrate, borders, and vbr mpeg-1 encoding to fit more minutes on one XVCD have been covered before, but I think your explanations were very good.

[Cornucopia]

Coupla more suggestions:

1. VBR. Yea, yea. Been there, done that. Not trying to "copy" somebody's idea. It's just helpful lowering the average bitrate when you can-therefore more minutes. Ah, but read on...

2. Make your source/cap AVI (or whatever) use 1/2 framerate, i.e. 15 fps instead of 30(29.97) or 12 instead of 24(23.93?)-sorry 'bout 25fps PAL guys. THEN, encode at the usual rate. This will duplicate every other frame making the interframe differences much less, and so the redundancy and compressability much more. The stutter of 15fps/12fps will be noticeable on action scenes, but only barely so on slow drama scenes. Don't know how much of a savings this is tho'.
*Note: this only works for players that can support VBR for XVCD/MPEG1.

3. Make your audio MONO. Then make the audio bitrate lower still-maybe 64kbps. I wouldn't go any lower-too many players won't support it.

4. Take #2 and that previous split-track idea to a rediculous extreme...Replace credits/titles and slowww drama scenes with an MPG Still, and just author them into a combined playlist. 8)

Happy encoding and tell the site how it went,

Scott

[House de Kris]

Wow Cornucopia, those are great ideas on saving bandwidth. Hopefully this thread will evolve into a collection of video and audio 'dumbing-down" in order to maximize xVCD time. I've got a couple to contribute, but I don't know if they will really help or not.

1. Limit Color Depth. Use a 'cartoon' effect to minimize color information and make huge areas of the scene the same color. This should be available in something like After-Effects.

2. Reduce Video Bandwidth. Use an extreme setting of the Blurr filter to take all high frequency content out of the picture. This is where MPEG compression spends a lot of bits.

Let's hear more suggestions to reduce a beautiful DVD to something you'd expect from 8mm FILM (not video tape).

[SatStorm]

Very good job MITSUI_1. All things you mention collected and justified on one post. I hope one day to be able to write something like that!
On thing more: I suggest start determine "PAL" or "NTSC" on our posts...

[Robbins1940]

Your post was just too long. I stopped reading it half way through.

David