An improved classified vector quantization(CVQ) algorithm is proposed for application to the hybrid coding of motion-compensated prediction and discrete cosine transform(MCP/DCT) for video image sequences.
As a classification measure, five DCT coefficients are used. According to the magnitude of these coefficients, each 8x8 image block is classified into one of 9 classes. These classes consist of non-edge types, ordinary edge types and strip types. Using energy distribution of each class, different forms of vectors are also investigated, where each vector consists of DCT coefficients of similar energy.
Proposed method is compared with two conventional schemes. One is MCP/DCT-VQ schemes without classification. Another is MCP/DCT-CVQ schemes with classification but without the class of strip types.
