Improved dictionaries for generalized bitplanes-based matching pursuits video coding using ridgelets

A novel decomposition of functions using generalized bit-planes has been recently proposed. Such a decomposition maps a function into a set of indexes, thus performing quantization and signal decomposition in a single step. In a previous work, this signal representation model has been successfully used replacing the decomposition/quantization operations in Neff and Zakhor´s matching pursuits video encoder. In this paper we investigate good codebooks for such decompositions whose construction is based on ridgelets. We show that by combining products of properly scaled and rotated ridgelets one can obtain codebooks which provide improved rate-distortion performances. Also, we confirm the theoretical prediction that good codebooks are the ones that have a small maximum angle between any vector in R^N and the closest one in the codebook.