Fuzzy learning vector quantization (FLVQ) algorithm outperforms the hard-competitive vector quantization in that it reduces the dependence of the resulting codebook on the initial codebook selection

yet it has the disadvantages of slow convergence and easy to be trapped in local minima. In this paper

the principle of fuzzy learning vector quantization for image coding is reviewed. Followed by a discussion of the possible ways for optimizing the FLVQ algorithm

a new fuzzy learning vector quantization algorithm based on tabu search(TS-FLVQ) is then proposed. In this algorithm

we firstly constructed a table listing oriented to global search by the tabu search algorithm

and afterwards took advantage of fuzzy learning to reach the global minimum point of the predefined objective function. The algorithm with a detailed description of the procedure involved was simulated in the computer finally. The algorithm differs from a standard greedy search in that the best move is executed also if it leads to a configuration with a greater energy than the current one; this is necessary to be able to escape from local minima. Experimental results show that TS-FLVQ has much better coding performance over FLVQ with remarkably faster convergence.