Hi-Res Slide Show Math implies x(S)VCD not up to speed.

[dk0]

Hi everyone,

First, I would like to say that this is a wonderful forum with excellent posts. It has helped me quite a bit with my video efforts.

Many people are interested in developing MPEG Slide Shows from still images with the best resolution and appearance. Furthermore, many posts have made recommendations for bitrates and other TMPEnc settings, indicating that they are able to view all of the detail in their original images. However, I have done some basic math and do not understand how this is possible.

UN-encoded bitrate requirement:
-------------------------------------
NTSC720 x 480)(pixels/frame) x (30 frame/sec) x
(24 bit/pixel) / (1024 bit/Kbit) = 243,000 Kbps

PAL requires the same bitrate since (720x480x30) = (720x576x25).

Maximum practical bitrate for x(S)VCD is roughly 3500 Kpbs.
So, MPEG(1 or 2) would require an encoding compression ratio of (243,000 / 3500) = 69.4

Typical achievable bitrates are closer to 2600 Kbps.
So, MPEG would require a ratio of (243,000 / 2600) = 93.5

At bitrates of 2000 (many recommendations), this would impose an encoding compression of 121.5 on MPEG!!

These compression ratios are not physically possible. If you read some of the technical material on MPEG (e.g., http://http://bmrc.berkeley.edu/frame/research/mpeg/mpeg2faq.html)
the actual compression ratios that MPEG can achieve are 30 or less. Therefore, one can conclude that x(S)VCD is not providing you with the full detail of the orginal still image.

On the other hand, DVD is capable of bitrates up to 9800 Kbps. So, the required MPEG encoding compression ratio would be (243,000 / 9800) = 24.8 which lies within the realm of what MPEG can deliver.

In fact, the minimum required bitrate for encoding still images with MPEG (assuming full MPEG compression ratio of 30) would be (243,000 / 30) = 8100 Kbps.

My conclusion is that only DVD can provide you with full resolution detail contained in the original still images at 720 x 480, and that any attempt with x(S)VCD will fall short by a factor in the neighborhood of 300%.

Furthermore, I am stumped as to why Ulead DVD Picture show would claim that picture quality is the same whether it is burned on CD-R(W) or DVD-R(RW). This could only be true if "same" means that the DVD is burned at a lower bitrate in order to produce an image that is identical to the one burned on the CD-R(W), but inferior to what could be achieved at a higher bitrate.

[zx80]

I believe since they are STILL images, bit rate per second is not an issue... after all you can longer than 1 second to load the image. It's within the VCD standard to have stills at full D1 resolution as I understand it.
I have displayed such images using the same tool on my DVD player (which loads them instantly) and my stand alone VCD player (VCD 2.0) which takes longer to load them (you can see the image build up) but really does display them to the same resolution/quality.
Hope that helps.
Moving images is a completely different story

[tacosalad]

Aha! Notice that you're comparing MPEG "Slide Shows" to MPEG files created with TMPEG. That seems intuitive enough, but TMPEG doesn't create true MPEG stills or slide shows. In the posts you mentioned, people are recommending ways to tweak TMPEG to get short motion video clips that can be used as substitutes for actual MPEG stills.

An actual MPEG still or slide show permits up to two seconds of bandwidth for a single I-picture. There are no B-pictures or P-pictures. Hence, that one I-picture gets as much space as 60 pictures would ordinarily use (at 30 fps).

If you don't have a program that can encode true MPEG stills or slide shows, you can tweak TMPEG to make a short motion MPEG of a still. It has to be longer than just one frame, though, or it'll look terrible due to the impossible compression ratio you just mentioned. (Anyone who's tried it will agree!) But with the right settings, it'll have enough space to encode a mediocre I-picture followed by enough P-pictures to clean it up. That's the effect zx80 described when he said, "you can see the image build up".

If you're lucky enough to have a program that generates true MPEG stills or slide shows (like VideoPack, WinOnCD, or the Philips VCD Toolkit) they'll look perfect as soon as they appear on the screen. I assume that's the format used in Ulead DVD Picture Show, and that's why they claim the images are just as good on CDR or DVD. If you've got that program, can you tell us if it looks as good as they say?

-tacosalad

[Anonymous65]

High resolution still images are stored as jpeg, not I-frame mpeg. On a VCD you can save 704x480 JPEG still images, with as little compression as you want. This is the only way to get high resolution images on a VCD. If you want to make a SVCD, then 480x480 is the maximum you can go. When playing stills, one image is loaded into memory, and repeated 30 times per second. If you want to make a mpeg file, you will have to record 30 frames per second, and the frames will be played back at that speed even if it's only one image. So, if you want to make a high resolution slide show, then JPEG stills is the way yo go. Mpeg files are a waste of space and lowers the quality.

[vitualis]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-12-19 17:35:33, skittelsen wrote:
High resolution still images are stored as jpeg, not I-frame mpeg. On a VCD you can save 704x480 JPEG still images, with as little compression as you want.</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

Actually, real high resolution still images are encoded as I-frame MPEG-1 (VCD) or MPEG-2 (SVCD). They are not encoded in JPEG. The stills consist of only one frame (on VCD, there is one high res still frame multiplexed with one normal res still frame).

What tacosalad stated before is correct.

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE> If you want to make a SVCD, then 480x480 is the maximum you can go.</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

I'm not entire sure what you mean, but the still on a SVCD is at 704x480/576 (same as VCD).

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>So, if you want to make a high resolution slide show, then JPEG stills is the way yo go.</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

Again, S/VCD do NOT use JPEG compression for still images. It uses MPEG-1/2.

Regards.

_________________
Michael Tam

<font size=-1>[ This Message was edited by: vitualis on 2001-12-20 02:25:56 ]</font>

[tacosalad]

Thanks Michael!

I was quite puzzled by his comments, so I didn't reply at first. Upon looking at his web page, I think skittleson is used to luxurious, professional VCD/SVCD mastering software that permits jpegs to be imported directly into a project without pestering him with the details of conversion.

Lucky fellow, really.

But did we really answer the original question?

I think we did establish that proper VCD/SVCD stills (whatever format they're in) are allocated more space than an ordinary video frame, and hence they don't really need 100:1 compression. In the mpeg 2 FAQ that dc0 mentions, it doesn't actually say 100:1 compression is "physically impossible" but rather that it's impossible to "achieve high quality video with compression ratios over 100:1". The key word here is "quality".

-tacosalad

[dk0]

Hi,

Sorry for the delay, but I wanted to verify some of the excellant information that you posted (thanks to all).

It turns out that tacosalad and vitualis are correct. High resolution still images are encoded in a special manner, namely consisting of MPEG I-frames only. There is no mention of jpeg in any of the VCD standard documents, nor any of the SVCD material that I could find. Jpeg files must be encodeed to MPEG in order to be accommodated by the S/VCD format.

Now on to the quality issue. Tacosalad is correct in that MPEG stills allocate up to 2 seconds of bandwidth to the image. This changes my initial calculations tremendously as follows:

Un-encoded bitrate requirements: (see my initial post)
--------------------------------
(243,000 Kbps) x (1 frame/2 seconds) x (1 second/30 frames) = 4050 Kbps

I simply corrected the frame rate number from 30 to 0.5 in my original equation. This is now the required bitrate that the TV signal must receive in order to yield the complete resolution of the image at either NTSC or PAL rates.

To be able to use an encoding bitrate of 2600 Kbps, the compression of MPEG I-frames must be at least (4050/2600) = 1.56. This is not at all an unreasonable compression factor. However, the problem that I now have is that it is not clear whether I-frames can be compressed by themselves. In other words, from my (admittedly limited) understanding of MPEG encoding, you need the P and/or B frames in order to achieve ANY compression.

So, in summary, if I-frames cannot be compressed alone (ie, without P, B frames), then the required encoding bitrate of 4050 Kbps is not possible with (S)VCD. And, if I-frames are indeed compressed, is the compression factor at least 1.56? If someone can answer this, then we have resolved the entire quality issue with regard to MPEG still image quality on S/VCD.

However, my original calculation of required bitrate (243,000 Kbps)does have meaning. It is the required bitrate to encode high resolution (720x480/576) motion video. And, my resultant encoding rate of at least 8100 Kbps would indeed be required to saturate the quality of current TV displays (NTSC or PAL). Basically, it is a quantitative measure of DVD quality compared to SVCD, simply showing how DVD can produce better video, and approximately by how much (300%).

Anyway, thanks for the posts, and if anyone can definitively answer the I-frame issue, then we can close this out.

[vitualis]

I-frames are still compressed... I'm not entirely sure how it works, but the method of compression is actually quite similar to JPEG compression. P and B frames additionally include temporal compression which allows for the massive compression ratios seen in an MPEG video clip.

The high resolution still images on a S/VCD are NOT lossless compared to the original image. Instead, you should consider them similar to quality as a reasonable/good JPEG image.

Regards.

[tacosalad]

I can't answer everything off the top of my head, but I've got two MPEG textbooks at home so I can look up the rest later. (I once wanted to write an encoder.)

Yes, I-pictures can be compressed, and it is indeed very much like jpeg. The picture is broken into 8x8 squares, and each square is analyzed to find a combination of cosine frequencies that would represent it. (Cosines are used instead of sines because it allows some of the formulas to be simpler.) These values are cosine coefficients.

For every 8x8 matrix of pixels, there is an 8x8 matrix of values. That doesn't sound like a savings, does it? Luckily, most of the coefficients are empty, so they don't have to be included in the output--an immediate savings! For an area of solid color, for example, only one cell would contain a value, so there'd be a 64:1 compression.

I-pictures can be further compressed, when needed, by quantizing them. This is the step that introduces the "halos" and "blocks" into the image. The compressor saves bits in the coefficients by reducing their precision, basically shaving off bits, but introducing small errors in the process. Additionally, the compressor might discard entire cosine coefficients, hoping the viewer won't notice the difference. Unfortunately, the viewer usually DOES notice the difference, so MPEG provides B-pictures and P-pictures as a better way to compress.

If you crank up the compression on a JPEG, usually by reducing the "quality" setting, you can get the same result. If you get up to really high compression settings, the picture decays into a mass of blocks and halos.

Sorry, that got a bit wordy. In answer to your question, yes I-pictures can usually compress about 3:1 without loss, which is well within the 1.56 you mentioned. As Michael pointed out, a VCD usually encodes two pictures, but the low-res one needs only a quarter as many bits, and there's plenty of slack for most images.

-tacosalad