home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.mac.misc
- Path: sparky!uunet!pageworks.com!world!ivanski
- From: ivanski@world.std.com (Ivan M CaveroBelaunde)
- Subject: Compact Video/Vector Quantization (was re: semi-technical QT info...)
- Message-ID: <BxGHon.85A@world.std.com>
- Organization: The World Public Access UNIX, Brookline, MA
- References: <BxF8nt.5zw@world.std.com> <BxG1LA.7tG.2@cs.cmu.edu>
- Date: Mon, 9 Nov 1992 15:57:58 GMT
- Lines: 41
-
- rwb@kurn.VI.RI.CMU.EDU (Robert Berger) writes:
-
- >In article <BxF8nt.5zw@world.std.com> ivanski@world.std.com (Ivan M
- >CaveroBelaunde) writes:
- >> It's a VQ (Vector Quantization) scheme, which is similar to finding an
- >> optimum CLUT for an image. You basically have a color map (called a
- >> codebook) and you map each pixel to the closest color in the codebook.
- >Not exactly. In VQ, you map blocks of MULTIPLE pixels to codebook entries.
- >For example, each codebook entry might be a 3x2 group of pixels.
-
- This is true. I was trying to simplify the concept by using the analogy
- to an optimum CLUT. Of course, what really happens is that you take
- each block of pixels that you will be quantizing and you consider it
- an n-dimensional vector (where n could be, for example, the # of components
- [RGB, YUV, or whatever] per pixel times the number of pixels in the block).
- Then the codebook is nothing but a list of these "vectors", in essence
- coordinates in n-dimensional space, and quantizing involves doing
- an n-dimesional distance calculation, thus determining which of the "vectors"
- the pixel block is closest to. Thus the name "vector quantization."
- As you mention, calculating and updating this codebook is a bear.
-
- I thought the CLUT analogy made the concept easier to digest than the
- ugly details (this is csm.misc, not csm.programmer, after all).
-
- >VQ has the property that the compression is many times slower than the
- >decompression. Finding a codebook that works well for a set of images
- >requires lots of computation.
-
- Absolutely. VQ is EXTREMELY asymmetric, and thus very nicely fitting for
- "publishing" endeavors, where the ability to do decoding a lot more
- cheaply than encoding makes a lot of sense.
-
- The VQ codec in QT 1.5 (Compact Video) also optionally performs data
- rate limiting, so it becomes even easier to make a movie CD-playable,
- for example.
-
- -Ivan
-
- ----
- Ivan Cavero Belaunde
- DiVA Corporation
-