Fast algorithm for pyramid vector quantization

Abstract

This paper presents an efficient implementation of a pyramid vector quantization coding/decoding algorithm developed by Filip and Ruf [2], where the original classical calculation of pyramid numbers is replaced by a sliding calculation. The sliding calculation allows movement in any direction of a lattice so that the calculation of in each algorithm iteration can be continued from the previous pyramid number. These properties can, under certain conditions, significantly (up to 25 times) accelerate the coding/decoding process. The condition under which the sliding calculation is more effective than the original one is also presented. The simulation results confirm the expected improvement over the original algorithm.

Vertex data compression through vector quantization

“…Once the nearest codevector is determined, its binary codeword, or index value, can be computed directly from the codevector itself, as first shown by Fischer [23]. Since then, other enumeration schemes have been developed that can offer increased error-resiliency [27] or improved performance [28].…”

A Kind VLSI Design of Vector Quantization Algorithm for Image Compression

“…Furthermore, VQ has qualities such as less sensitivity to channel errors and subjective performance improvements not evident in objective performance measurements, etc [2]. This paper focuses on the image encoding part of the VQ compression system.…”