Browse Prior Art Database

Method for Adaptively Initializing a Source Model for Symbol Encoding

IP.com Disclosure Number: IPCOM000042424D
Original Publication Date: 1984-May-01
Included in the Prior Art Database: 2005-Feb-03

Publishing Venue

IBM

Related People

Authors:
Arps, RB Langdon, GG Rissanen, JJ [+details]

Abstract

This invention relates to an adaptive string encoding method in which a priori statistics are used to start up and possibly modify the downstream encoding estimate. The method uses the binary events of ascertaining the ratio of new symbol occurrence to the total number of symbols in a startup string. With respect to the next encoding event, if the symbol is "new", a flat probability of occurrence is used, whereas if the symbol is "old", then it is encoded with a distribution based on counts. The method steps include (a) partitioning the symbols into two classes, "the observed" and "unobserved"; (b) assigning conditional probabilities to the symbols within each class; and (c) using an age-dependent scheme to assign probabilities in a code space to the observed characters.