Estimation of clustering algorithms to design code books for practical usage of vector quantization

Vector quantization (VQ) is useful technology for communication terminals with both small payloads and limited computational abilities. Clustering algorithms are mainly used to design a code book (CB) for VQ. In most previous studies to estimate performance of clustering algorithms, learning images to design the CB for VQ were the same as the test images to examine their performances. However, it is necessary to encode/decode unseen images for practical usage of VQ. The performance of the clustering algorithms therefore has to be estimated under the condition that learning images are different from test images. This comparative study is indispensable to examine effectiveness of clustering algorithms for practical usage of VQ. We selected mainly used four clustering algorithms and estimated performance of these algorithms to design a CB for practical usage. A set of computational experiments showed that there was marginal difference in the performance of clustering algorithms.