grant@hp-pcd.UUCP (06/11/84)
For a different approach yet, see the first article in the
June, 84 "Computer" magazine (IEEE). It gets around 2x compression
without any prior knowledge of character/word probabilities.
The algorithm is to start with a set of symbols which represent the
input characters. As a message is read in, characters are parsed off
to match the longest symbol defined. When this string is ended, a new
symbol is defined which is the parsed symbol followed by the next character.
The parsed symbol is then sent. Decoding rebuilds the symbol table from
the input stream, so the translation table does not need to be transmitted.
One question I had: what does the algorithm do when it runs
out of symbols? I think this method has good potential.
Grant Garner
hp-pcd!grant