SEQUENTIAL AND PARALLEL NEURAL-NETWORK VECTOR QUANTIZERS

This correspondence presents novel sequential and parallel learning te chniques for codebook design in vector quantizers using neural network approaches. These techniques are used in the training phase of the ve ctor quantizer design. Our learning techniques combine the split-and-c luster methodology of the traditional vector quantizer design with neu ral learning, and lead to better quantizer design (with fewer distorti ons). Our sequential learning approach overcomes the code word underut ilization problem of the competitive learning network. As a result, th is network only requires partial or zero updating, as opposed to full neighbor updating as needed in the selforganizing feature map. The par allel learning network, while satisfying the above characteristics, al so leads to parallel learning of the codewords. The parallel learning technique can be used for faster codebook design in a multiprocessor e nvironment. It is shown that this sequential learning scheme can somet imes outperform the traditional LBG algorithm, while the parallel lear ning scheme performs very close to the LGB and the sequential learning algorithms.