Scene adaptive vector quantization for image coding

The authors propose a scene-adaptive vector quantization (VQ) scheme based on component VQ which encodes an image block into three fundamental components: mean, gain, and index of normalized output vector. The scheme provides scene adaptability by incorporating a dynamic codebook which is successively self-generated through the intellectual coding process. The latest codebook set with learning function can be stored and used for subsequent VQ operations. A hierarchical fast search algorithm is derived from a structural combination of a fixed codebook and a dynamic one. Excellent coding performance is demonstrated through computer simulation