On Redundancy of Rice Coding

In this paper we derive the relative per-symbol redundancy of the Rice coding algorithm, which is a widely used technique in image compression for very fast entropy coding. We show, that for some important source models, such as the two-sided geometric distribution (TSGD), the redundancy depends on the source entropy H and it tends to zero if H -> oo. The redundancy is upper bounded by 50% if H -> 0.