RE: MPEG-1 Codecs

Chris O'Leary (coleary@heuris.com)
Mon, 5 Oct 1998 18:42:38 -0500

> If you have a static scene, with someone alking, it will
> have data rate A.
> If you add a cutaway to a product shot and cut back to
> the talking head, I
> would think the data rate would jump at those two edit
> points, and thusly, the
> overall data rate would increase a little bit.
> If you go back and edit a dissolve to a product shot and
> dissolve back to
> the person talking into that MPEG stream, I would believe
> that the data rate
> would go up by a large extent- because the entire image is changing.

This is precisely the problem with doing edits. When the clip is first
encoded, the encoder is watching the fill factor of the decoder buffer
(that called Video Buffer Verification or VBV) to ensure that there are
as many bits as possible in the buffer, in order to ensure quality,
while making sure that you do not overflow the buffer and render the
stream illegal. If you start adding bits to a stream, then you risk
mucking with all of these careful calculations and you can't really be
certain if you won't overflow the video buffer. It's hard to know what
your buffer fill factor is from that point forward. The safest thing to
do is to encode the new stuff at really low bit rates. The problem is
that this means that the new scenes will be of very low quality. The
only way to ensure both quality, legality, and small file size is to
re-encode from this point forward.

> The only way you could maintain a given data rate while
> adding changes
> that require additional data, would be to reduce the quality
> of the video
> overall. IS that right?

Either you cut the quality or you run the risk of generating an illegal
stream - and you may get an illegal stream even if you do drop the
quality.

Chris O'Leary
V.P. of Marketing
Heuris