Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!aero-c!gumby.dsd.trw.com!deneva!news
From: thomsen@spf.trw.com (Mark R. Thomsen)
Newsgroups: comp.sys.next
Subject: Re: DCT(JPECG), WAVELET on NeXTDimension
Message-ID: <284849F5.6745@deneva.sdd.trw.com>
Date: 2 Jun 91 01:29:24 GMT
References: <1991May30.233044.7890@spool.cs.wisc.edu>
Sender: news@deneva.sdd.trw.com
Organization: TRW Inc., Redondo Beach, CA
Lines: 30

Soo Lee writes
  Hi all,
  
  I am afraid that C-CUBE is changing their mind on JPEG chip. As I learned
  that DTC(Discrete Cosine Transformation) is implemented on JPEG chip, I am
  puzzled why they didn't switch to WAVELET algorithm which appears better,
  cheaper and faster than DTC algorithm. <more discussion to support this>

It is not clear from our work here that a Wavelet (nee Gabor logon)
compression approach is better, cheaper, or faster. While decompression
is almost definitely faster, compression has proven to be slow. The main
problem we see in JPEG has been energy loss when dropping DCT coefficients.
To preserve energy would require more computations and decision branches
in the algorithm - making chips more expensive and forbidding designs.
The lossy Wavelet approach has an energy drop too, though the artifacts
are different. The differences is better or worse depending on the image
used and the viewer ... too subjective.

I think there is so much work to be done in image compression that it
is premature to fixate on a single approach. If a C-Cube chip is low
enough price and can support 20:1 ratios to match NTSC:disk drive then
it is fine, until the dust settles a bit more. With the JPEG push there
will be low commercial support for Wavelet chips, software, and file
exchange for a while. Use the first standard to enhance exchange and
functions until there truely is a better algorithmic chip/circuit.

Of course, it would be lovely if the daughter board approach could
support such a switch in the future.

Mark R. Thomsen