A codebook design algorithm for vector quantization of images

The authors present a new codebook design algorithm for the vector quantization of images. A two-dimensional discrete cosine transform (DCT) is used to extract the feature vectors of training images. Using the good energy compact capability of DCT, certain significant components of the feature vectors are extracted and then used to partition the training vectors into a binary tree. This results in a significant savings of computation in the clustering process. Simulation results show that the new algorithm requires about 51% of the amount of time needed by the pairwise nearest neighbor (PNN) algorithm and a better reconstructed picture quality is obtained