K-means algorithm applied in vector quantization strongly depends on the selection of the initial codebook

and if not given a good initial codebook it can easily be trapped in local minima. Furthermore

Bezdek's fuzzy K-means algorithms are computationally expensive so that they are impractical in codebook design. So

people have been researching those algorithms which can achieve good performance in the convergent speed of algorithms and the quality of the reconstructed image. Analyzing the fuzzy vector quantization algorithm (FVQ) presented by Nicolaos.B.K. and aiming at the irrational convergent procedure of the algorithm

from the aspect of convergent structure and strategy the paper presents a dynamic fuzzy vector quantization algorithm (DFVQ) and gives two concrete methods based on the idea of the presented algorithm. Experiments show the presented methods markedly accelerate the convergent procedure and improve the quality of convergence.