Novel codebook design techniques for vector quantization image compression

In this paper new techniques for codebook design are presented. A new algorithm for codebook design is described. Specific characteristics of the blocks of the training sequence are taken into consideration during generation of the initial codebook. Beginning with a representative initial codebook the iterative optimization procedure converges quickly in representative final codebook which in turn leads to high output image quality. A technique for computational extension of small codebooks is also proposed. It is based on the application of simple transformations on the codewords. The reduced memory requirements of the proposed technique makes it very useful for applications requiring low-power consumption.