Image compression using block pattern-vector quantization with variable codevector dimensions

A new coding scheme for image compression using vector quantization (VQ) is proposed. The new technique offers: (1) excellent edge reproduction to the extent that the reproduced edges are not staircased, (2) no noticeable blockiness in the reproduced image, and (3) very robust performance. All the above advantages are obtained at bit rates of 0.7-0.8 bpp with moderate computational complexity. The novelty of the new technique lies in encoding two vectors for each 4*4 edge block. This is in contrast with the case of the ordinary VQ where only one vector is used to represent each edge block. One of the proposed two vectors corresponds to the light pixels in the 4*4 edge block and the other vector corresponds to the dark pixels in the same block.<
>