A Massively Parallel Algorithm for Vector Quantization

Overview

A Vector Quantizer can be defined as a mapping of input vectors of a particular size into a finite set of codevectors in a codebook. Signal compression is achieved in VQ by using a codebook with relatively few codevectors compared to the number of possible signal vectors.

This package contains programs for realizing full-search vector quantizers, the Linde-Buzo-Grey algorithm for codebook design and decoders for full-search vector quantizers. The first two are written using the MasPar programming language (MPL), and runs on MasPar MP-2 massively parallel computers. The decoder is written in ANSI C. The programs are written to accomodate a range of codebook and vector sizes. In addition to these, there are also programs to convert 2-D signals of specified dimensions into a stream of vectors and vice versa. All the programs assume the image to be 512 x 512 monochrome images stored in the Sunraster format. Additional modifications are required to accomodate images of arbitrary sizes and other raster formats.

Availability

Retrieve the package (20058 bytes in the compressed tar file)