Summary form only given. We first classify VQ indexes into smooth and non-smooth groups by using the relative variance of each code vector. Based on the smoothness of neighboring indexes, we define three different probability models. These models describe the correlations between the current VQ index and its neighboring indexes more precisely, and adaptive arithmetic coding schemes can be applied more efficiently. Furthermore, the size of each model is guaranteed not to be greater than the square of the number of code vectors, and there is no difficulty in implementation of arithmetic coding. In one of the models we generalize the idea of gradient match in Juan and Lee (1998) to predict the current index by use of its four known neighbors. We compare our method with conditional entropy coding of VQ indexes.

Summary form only given. Entropy-constrained residual vector quantization (EC-RVQ) has been shown to be a competitive compression technique. Its design procedure is an iterative process which typically consists of three steps: encoder update, decoder update, and entropy coder update. We propose a new algorithm for the EC-RVQ design. The main features of our algorithm are: (i) in the encoder update step, we propose a variation of the exhaustive search encoder that significantly speeds up encoding at no expense in terms of the rate-distortion performance; (ii) in the decoder update step, we propose a new method that simultaneously updates the codebooks of all stages; the method is to form and solve a certain least square problem and we show that both tasks can be done very efficiently; (iii) the Lagrangian of rate-distortion decreases at every step and thus this guarantees the convergence of the algorithm. We have performed some preliminary numerical experiments to test the proposed algorithm. Both random sources and still images are considered. For random sources, the size of training sequence is 2500 and the vector size is 4. For still images, the training set consists of monochrome images from the USC database and the vector size is 4/spl times/4.

We describe a method of using a Lagrange multiplier to make a locally optimal trade off between rate and distortion in the motion search for video sequences, while maintaining a constant bit rate channel. Simulation of this method shows that it gives up to 3.5 dB PSNR improvement in a high motion sequence. A locally rate-distortion (R-D) optimal mode selection mechanism is also described. This method also gives significant quality benefit over the nominal method. Though the benefit of these techniques is significant when used separately, when the optimal mode selection is combined with the R-D optimal motion search, it does not perform much better than the codec does with only the R-D optimal motion search.

This paper introduces an adaptive search vector quantization (VQ) algorithm, hierarchical adaptive search VQ (HASVQ) and compares its performance against several previously proposed adaptive search image compression algorithms. These algorithms are: the JPEG standard, adaptive tree search vector quantization (ATSVQ), multi-stage adaptive search vector quantization (MASVQ). The bit-rate examined is from 0.625 to 1.25 bits per pixel (bpp). The results indicate that HASVQ outperforms all the other algorithms over the entire range of bit-rates examined. For higher bit-rates MASVQ performs next to the HASVQ algorithm and is, in-turn, followed by the JPEG standard. ATSVQ performs well at lower bit-rates, however its performance at higher bit-rates is suppressed by not having an adequate codebook size.>

This article discusses bit allocation and adaptive search algorithms for mean-residual vector quantization (MRVQ) and multistage vector quantization (MSVQ). The adaptive search algorithm uses a buffer and a distortion threshold function to control the bit rate that is assigned to each input vector. It achieves a constant rate for the entire image but variable bit rate for each vector in the image. For a given codebook and several bit rates, we compare the performance between the optimal bit allocation and adaptive search algorithms. The results show that the performance of the adaptive search algorithm is only 0.20-0.53 dB worse than that of the optimal bit allocation algorithm, but the complexity of the adaptive search algorithm is much less than that of the optimal bit allocation algorithm.

This paper discusses some algorithms to be used for the generation of an efficient and robust codebook for vector quantization (VQ). Some of the algorithms reduce the required codebook size by 4 or even 8 b to achieve the same level of performance as some of the popular techniques. This helps in greatly reducing the complexity of codebook generation and encoding. We also present a new adaptive tree search algorithm which improves the performance of any product VQ structure. Our results show an improvement of nearly 3 dB over the fixed rate search algorithm at a bit rate of 0.75 b/pixel.>