An Improved Discrete Particle Swarm Optimizer for Fast Vector Quantization Codebook Design

For tree-structured vector quantizers (TSVQ), the codebook quality highly depends on the splitting criterion and the approach by which a specific node is selected and then be partitioned into new ones. Among several proposed TSVQs, maximum descent (MD) algorithm can produce high quality code-books and reduce the computation time simultaneously. In this paper, under the basic structure of MD algorithm, we propose an improved discrete particle swarm optimizer with less computation cost and faster convergence rate than the conventional one, and then, based on which, a novel binary partitioning scheme for MD algorithm is presented. Experimental data show that the newly proposed algorithm can further improve the codebook quality while the computation time is almost equivalent to that of the MD algorithm.