An efficient Huffman table sharing method for memory-constrained entropy coding of multiple sources

Abstract

For optimal entropy coding of multiple sources, the encoder and the decoder have to retain a separate Huffman table for each source. In many practical cases, however, available memory is usually restricted and therefore it is necessary for some sources to share a Huffman table. Recently, we developed an iterative algorithm, which leads to locally optimal sharing of Huffman tables (Lee et al., 1995). In this paper, we examine the iterative algorithm in detail and present some modification methods which improve the sharing performance and reduce computational complexity. First, considering that the iterative algorithm provides only locally optimum, we introduce an unused table processing method and two initialization methods, splitting and merging, for the iterative algorithm. And fixed-length coding is introduced for the encoding of some sources, which provides considerable performance improvement in case only a small number of Huffman tables are used. In addition, we present a simple approximation method of codelengths, which reduces the computational complexity of the sharing algorithm with little performance degradation. Simulations show that performance can be further improved with these modifications. For practical applications, we also present a search method for fast determining the number of Huffman tables and the Huffman table size, which, under a given memory constraint, minimize average bit-rate. The proposed sharing method can be widely applied to many high-order entropy coding systems with memory constraints.