Path: utzoo!attcan!uunet!philmtl!philabs!ttidca!schear From: schear@ttidca.TTI.COM (Steve Schear) Newsgroups: comp.graphics Subject: Re: SigGraph Fractal Compression Message-ID: <5455@ttidca.TTI.COM> Date: 17 Aug 89 23:39:12 GMT References: <444@mit-amt.MEDIA.MIT.EDU> <20400001@inmet> <13147@well.UUCP> Reply-To: schear@ttidca.tti.com (Steve Schear) Organization: Citicorp/TTI, Santa Monica Lines: 11 In article <13147@well.UUCP> gors@well.UUCP (Gordon Stewart) writes: >It seems to me that this is an ENCODING technique -- one that does not >preserve 100% of the information input. > >Am I wrong? > No, you are correct. Fractal compression attempts to vectorize a continuous image by identifying objects or segments of objects which can be coded with coefficients of an affine transform. It works well on most "natural" objects, since they are generally contructed from processes which are recursive (e.g., cell division, crystal growth) or operate locally (e.g., weathering). Natural objects have been found to display a great deal of fractional dimensional similarity over several orders of magnitude. So, once you can identify a natural object in an image, for which you have already determined its afine transform and other IFS (Iterated Function System) information, you can (with relatively good accuracy) render that object at various resolutions, even ones which greatly exceed the resolution of the original image capture system. One may argue, and quite correctly, that these reconstructed images are not identical to the original images (esp. those which have enhanced resolution). I think that they are more akin to an artist's, abeit a very detail oriented artist's, rendition of the original scene. This may in many applications be more than sufficient.