Motion-compensated vector quantization with a dynamic codebook

An interframe coding scheme, motion-compensated vector quantization (VQ) with a dynamic codebook, has been proposed. The basic idea of this algorithm is to combine the concepts of motion-compensation on a block basis and vector quantization. In the scheme, three steps are involved. First, each input vector is compared with the one on the same spatial position in the previous frame. If the difference is smaller than a preset threshold value, this input is simply replaced by the one in the previous frame. If not, the second step follows. A dynamic codebook is first generated for each input vector at this step. The codewords (or representative vectors) in the codebook consist of the vectors extracted from a motion estimation search window on the previous frame. As the motion search window varies with the spatial position of the input vector, the codebook is automatically updated without the need of any side information and overhead. If the difference between the input vector and the best matched codeword in the codebook is smaller than the preset threshold, the index or label of the codeword is transmitted. If not, the encoding goes to the third step, in which the differences between the the blocks on the consecutive frames are further encoded by the transform coding. The experimental results demonstrated that good reconstructed pictures can be obtained at a bit rate as low as 0.4 bit/pixel