Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!pt.cs.cmu.edu!nl.cs.cmu.edu!mlm
From: mlm@nl.cs.cmu.edu (Michael L. Mauldin)
Newsgroups: comp.graphics
Subject: Laplacian Pyramid Image Compression?
Message-ID: <8795@pt.cs.cmu.edu>
Date: 9 Apr 90 22:08:56 GMT
Organization: Carnegie-Mellon University, CS/RI
Lines: 43

Has anyone in UseNetLand implemented the Laplacian Pyramid image
compression algorithm?  I believe it was developed at RCA -- I just saw
a segment about it last night on "For All Practical Purposes" with Sol
Garfunkel.

The researchers were claiming 8 to 1 compression for greyscale images.

Anyone have any pointers to relevant literature for this algorithm?

--
Synopsis:

To encode an image, you repeat the following process:

	1. Generate a 1/4 size image which is the 'smooth' of
	   the full image by taking a Laplacian of the original.

	2. Blow up the 1/4 size image to full size (how was not
	   clear), and then take the difference of the original
	   and the enlarged, smooth image.

	   For regions of low detail, the difference image is
	   mostly zeroes and small integers.

	3. Recursively endcode the 1/4 size Laplacian by repeating
	   steps 1 & 2 on the smaller image, until you get down to
	   a single pixel (or some trivially small image).

The size requirements are

	1 + 1/4 + 1/16 + 1/64 + ... = 4/3

But this image is mostly zeroes and small numbers, and can be compressed
easily by other methods.

You transmit an image by starting with the lowest resolution images, so
the image builds up in resolution.  They showed this happening in color
for a real estate photo library on the show.

--
Michael L. Mauldin (Fuzzy)		Center for Machine Translation
ARPA: Michael.Mauldin@NL.CS.CMU.EDU	Carnegie Mellon University
Phone: (412) 268-5293			Pittsburgh, PA  15213-3890