Interpolative Coding as an Alternative to Arithmetic Coding in Bi-Level Image Compression

Abstract:
Huffman and arithmetic coding have established a firm status as statistical ’entropy coders’ in data compression. Both are in a sense optimal: Huffman in symbolwise coding, and arithmetic in ’continuous’ coding. Optimality is always relative to the ability of the source model to reveal the redundancies occurring in the data. The probabilities of elements produced by the model are important information for a statistical coder, but extraction of reliable probability values may be quite timeconsuming. This paper suggests the usage of interpolative coding as a simple, non-probabilistic alternative to arithmetic coding, with bi-level image compression as a sample application. According to our experiments, a basic context model, combined with interpolative coding, can give the same or better compression than the standard JBIG and JBIG2 coders, but requires only half of the processing time. For cases where speed is not critical, an extended source model is described giving still better compression.