home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!ukma!asuvax!ncar!news.miami.edu!newssun.med.miami.edu!dbrown
- From: dbrown@newssun.med.miami.edu (Daniel Brown)
- Subject: Image Input for a Neural Net
- Message-ID: <1992Nov12.181625.4135@newssun.med.miami.edu>
- Organization: University of Miami, School Of Medicine
- References: <1992Nov12.140317.7643@debbie.cc.nctu.edu.tw>
- Date: Thu, 12 Nov 1992 18:16:25 GMT
- Lines: 22
-
- I'm looking for information about preprocessing for image
- inputs into a neural-net.
-
- Our project requires the use of a 480x480 24-bit color
- image file in the tiff format. Even after converting to
- a greyscale image, that's over 65,000 inputs if used
- directly. I don't know too many systems that can handle
- that much information; certainly ours can't.
-
- I've looked a couple of compression techniques: i) summing
- and averaging over 100 pixels at a time (a coarse
- version of the former image), ii) coarse coding similar
- to the latter), and iii) sampling of a FFT transform of
- the input data.
-
- The obvious goal is to use as much important information
- without overloading the system. Does anyone have any other
- suggestions?
-
- Dan
- dbrown@newssun.med.miami.edu
-
-