On Optimality of Context Modeling for Bit-Plane Entropy Coding in the JPEG2000 Standard

Abstract

The JPEG2000 image compression standard exploits two empirically designed context models for conditional entropy coding of bit-planes of wavelet transform coefficients. The models use the so called significance (binary) values of eight adjacent coefficients as a context template, which are mapped into 9 conditional contexts. This paper addresses the problem of optimality of this approach. In other words, we answer the question: given the context template, is it possible to design models that would result in a better compression performance? In our work, we exploited optimization techniques for model design. We show that for the chosen context template, optimization results in only marginal improvement of compression performance (about 0.3%) compared to the JPEG2000 models for the class of natural images. Our conclusion is that the compression efficiency can be improved only by choosing a larger context template.