Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.csd.uwm.edu!cs.utexas.edu!husc6!mit-eddie!mit-amt!peter
From: peter@mit-amt.MEDIA.MIT.EDU (Peter Schroeder)
Newsgroups: comp.graphics
Subject: Re: SigGraph Fractal Compression
Message-ID: <537@mit-amt.MEDIA.MIT.EDU>
Date: 19 Aug 89 18:59:40 GMT
References: <444@mit-amt.MEDIA.MIT.EDU> <20400001@inmet> <5455@ttidca.TTI.COM> <8770@saturn.ucsc.edu>
Reply-To: peter@media-lab.media.mit.edu (Peter Schroeder)
Organization: MIT Media Lab, Cambridge
Lines: 35

In article <8770@saturn.ucsc.edu> chucko@saturn.ucsc.edu (Chuck Stein) writes:
>
>	I have heard no talk of fidelity of the reconstructed fractally
>compressed images.  Has Barnsly or anyone else compared reconstructed
>images with those produced using other techniques such as DCT.  I haven't
>seen the latest pictures, but the old ones look quite impressionistic.
>MSE-based signal-to-noise ratio fidelity metrics are poor when comparing
>across different compression techniques, but subjective rating-scale
>tests or psychophysical tests can be performed to compare between methods.
>What kind of fidelity measures do they talk about (if they do)?
>	
>	Chuck Stein	{chucko@saturn.ucsc.edu}


I saw some of the latest images at the conference Computers and Mathematics
and must say I am very impressed! I also thought the earlier ones looked
just like painters' impressions  of the original photograph. This time he
showed among other things a photo of himself and the fractally
reconstructed version of it. The only difference that I could detect, was
an ever so slight blurring, which they might have added after the
reconstruction itself. Supposedly it was a compression ratio of 1:128 which
is as good as the best compression techniques for movies (!) that I know
about. Since the fractal compression is still in it's infancy I suspect
they might do even better, while the other techniques have to push very
hard now to reach anything above 1:100...

The main question that remains for me at this point is, whether it will
work with sequences of images, that is over a 2D x time domain. It turns
out that it is nontrivial to make a technique, which works for a single
picture, work for a sequence of pictures without artifacts which only
become visible in the sequence.

Peter

peter@media-lab.media.mit.edu