RECURSIVE GROUP CODING WITH DYNAMIC FREQUENCY MODELING

Abstract

Task of entropy group coding of data for reduce of its statistical redundancy on base of recursive group coding is considered. Recursive group coding provides effective coding of symbols of large alphabets. It is fast and in some cases more effective alternative of such methods as arithmetical coding and Huffman coding. However applications of recursive group coding are restricted by usage of statistical frequency modeling. For such modeling frequencies (or probabilities) of symbols are calculated for entire text as well as information about size sand contents of groups are stored in compressed file. In this work a modification of recursive group coding are proposed which is able effectively compress texts statistically heterogeneous by frequency characteristics. A model for synthesizing of test texts is considered. Also a new metric for estimates of uniformity of a given text is proposed. It is shown that for considered test texts the proposed modifications of recursive group coding provide up to 65% less compressed data size than conventional method.

Keywords

• рекурсивне групове кодування, ентропійне кодування, арифметичне кодування, кодування хафмана.