Path: utzoo!attcan!uunet!kddlab!ccut!tansei!b39756 From: b39756@tansei.cc.u-tokyo.JUNET (Martin J. Duerst) Newsgroups: comp.graphics Subject: Re: Fractal Compression Summary: Approximative and Error Free Compression Message-ID: <1976@tansei.cc.u-tokyo.JUNET> Date: 23 May 88 05:58:39 GMT References: <686@thalia.rice.edu> <5546@cup.portal.com> <1947@tansei.cc.u-tokyo.JUNET> <522@etn-rad.UUCP> Reply-To: b39756@tansei.cc.u-tokyo.JUNET (Martin J. Duerst) Organization: Computer Center, University of Tokyo, Japan. Lines: 61 In article <522@etn-rad.UUCP> jru@etn-rad.UUCP (John Unekis) writes: >.... >I think I must be missing something here. I understand that it is possible to >come up with representations for graphically generated images that are very >compact. What I get frustrated with is the imprecision of terminology used >when talking about image compression. The filled and shaded polygon images >used in cartoons and advertisements are easy to parameterize and can be stored >in forms that require very little data to be recorded. SO WHAT? > There are basically two forms of image compression, approximative and error free (exact). See, e.g., A. Rosenfeld and A.C. Kak, Digital Picture Processing, Second Edition, Volume 1, Chapter 5 (Academic Press, New York, 1982). If speaking about exactness, etc., it is clear that this has to be approximate compression, so nobody mentiones this explicitly. The reason that approximate picture processing is so widely used (in contrast to the fact that there is nothing such as approximate text compression) is that the digital picture (X*Y pixels of Z bits depth) is already an approximation of an image that is, unless we descend to the level of photons, analog. How much 'compression' happens during digitalization and how much during actual compression is not so important, as long as the final (decoded) image is good enough for the specific application. >Suppose I give you an image from the real world. An example might be a >digitized X-ray image. An exact representation of this image is absolutely ^^^^^^^^^^^^^^^^^^^^ An exact representation of the original image is impossible, digitalization allways includes approximations. Instead of searching error-free compression algorithms, why not digitize at twice the resolution and use an approximate compression. This might give better compression factors with the same ability to detect cancer (or better cancer detection with the same compression rate). >critical since such problems as lumps of cancerous cells may initialy show up >as one or two dark pixels on the slide. No compression algorithm which might >either remove such pixels or allow them to be created by error is acceptable. There are many other components in your imageing system that can introduce errors and inaccuracies, so absolute precision in one step might only hide the fact that you deal basically with inprecise information. A cancerous cell could initially be smaller than a pixel and not visible, but well worth to see to further the progress of medicine. >It is simply not good enough if the compression/decompression merely preserves >major details in recognizable form. Most hospitals will not allow compression >and decompression of such images unless it can be demonstrated that the >reconstructed image when subtracted from the original yields ABSOLUTELY ZERO >different pixels. That is a legal aspect. However, an overall approach considering the whole imageing process might lead to better results, if only the regu- lations can be changed. >After years of looking at image compression I have not seen a compression >technique which will exceed 10 to 1 on the average image (not one specially >selected to compress well) and still be completely non-destructive. >..... > voder!wlbr!etn-rad!jru Even 10 to 1 is very good, 3 to 1 or four to 1 are more usual. If you have a good reference on an exact compression algorithm that acchieves 10 to 1, please mail me or post it. Martin J. Duerst Dept. of Inf. Sc., Faculty of Sc. University of Tokyo 7-3-1 Hongo, Bunkyo-ku 113 Tokyo, Japan