A Comparison Study Of Vector Quantization Codebook Design Algorithms based on the Equidistortion Principle

H. Takizawa, T. Nakajima, K. Sano, and H. Kobayashi (Japan)


competitive learning, vector quantization, neural networks, the equidistortion principle


This paper discusses vector quantization codebook design algorithms based on the equidistortion principle, which is effective to minimize quantization error evaluated by the mean squared error. In this paper, we review some repre sentative vector quantization codebook design algorithms including our law-of-the-jungle algorithm. Those algo rithms are then compared from the viewpoint of vector quantization performance. In addition, computational ef ficiency of each algorithm is also examined. Experimen tal results show that the law-of-the-jungle algorithm is the most promising for practical vector quantization applica tions, in terms of both vector quantization performance and computational efficiency.

Important Links:

Go Back