In the CS-MSVQ(Constrained Storage-Multistage Vector Quantization) algorithm, a codebook sharing technique is adopted in the MSVQ by carefully grouping residual vector sources into certain classes at every stage and allocating muliple codebooks for each class. It improves the coding performance with a marginal storage increment compared with the conventional MSVQ. In this paper, an efficient variable rate CS-MSVQ algorithm is proposed for image coding based on the greedy bit allocation algorithm. The proposed algorithm is a natural extention of the straightforward codebook allocation algorithm for the CS-MSVQ. In order to design efficient variable rate CS-MSVQ, the codebooks allocated in each group are selected one by one so as to optimize the rate-distortion relationship rather than all the codebooks in a group are allocated at a time. Because of the optimal codebook assignment, the performance of the proposed method can be improved at a given bit rate compared with the one of the conventional CS-MSVQ.
The performance of the proposed method is investigated by computer simulation for various test images and compared with the ones from convential MSVQ algorithms such as the fixed rate MSVQ, variable rate MSVQ, and fixed rate CS-MSVQ. The simulation results show that the proposed one outperforms the conventional ones over all the range of bit rates with marginal storage increment in tems of the subjective quality as well as the PSNR. For an example, the proposed method shows 1.5dB improvement over the variable rate MSVQ at 0.25bpp on "Lena" image.