Fast Vector Quantization Algorithm Based on Vector Features

The codebook design of vector quantization (VQ) based on a training set is computationally very expensive due to a lot of distance computations in the process of its clustering. In order to speed up the process of VQ codebook design, a fast algorithm is proposed in this paper. The proposed algorithm puts the training vectors into an orderly whole according to the characteristic values of training vectors. An ordered initial codebook is got from the ordered training sets. The clustering process of VQ is speed up by employing the fast kick-out conditions in the ordered codebook. Experimental results confirmed that the proposed method can speed up the design process and improve the codebook performance.