A two-stage vector quantization approach via self-organizing map

In this paper, a two-stage algorithm for vector quantization will be proposed based on self-organizing map (SOM) neural network. Firstly, a conventional self-organizing map will be modified to deal with dead codebooks in the learning process and then will be used to obtain the codebook distribution structure for a given set of input data. Secondly, sub-blocks will be classified based on the previous structure distribution with a prior criteria. Then the conventional LBG algorithm will be applied to these sub-blocks for data classification with initial values obtained via SOM. Finally, extensive simulations illustrate that the proposed two-stage algorithm is very effective.