Conditional source coding with competitive lists

Abstract

Summary form only given. A new lossless source coding algorithm was developed that achieves a compression ratio slightly better than the Lempel-Ziv-Welch algorithm, but requires as little as 250 kBytes of storage. The algorithm is based on the context-tree approach, encoding one input symbol at a time. Thus, its throughput lies in a range comparable to the PPM algorithm. The very low memory requirement is achieved by eliminating the costly probability estimation commonly performed at every context in context-tree algorithms. The algorithm uses a competitive list at every context. The competitive list is an invertible device that converts the output stream of an unknown discrete memoryless source into a stream of integers whose first-order probability distribution is monotone. The output of all the lists is encoded using a single arithmetic encoder.