High Efficient Image Coding System Using Vector Quantization in Transform Domain

本计画系统地研究向量量化影像编码方案。研究结果如下。（1）关于图象信号矢量量化的考虑指出了矢量量化应用中的两个问题。一个是矢量量化器的复杂性，另一个是矢量量化器性能的鲁棒性。（2）基于离散余弦变换的矢量量化编码（DCT-VQ）提出了一种新的基于离散余弦变换的矢量量化编码方案，称为DCT-VQ。为了达到通用矢量量化器，矢量量化器被设计用于独立于图像信号的标准概率分布。由于将变换域分解为向量，大大降低了计算复杂度。（3）提出了自适应DCT矢量量化（Adaptive DCT-VQ），它根据图像信号的局部特性自适应地控制编码参数，以提高DCT矢量量化的性能。仿真结果表明，自适应DCT-VQ编码优于传统的自适应变换编码。最后给出了该方案在彩色图象编码中的应用。（4）差分矢量量化（Differential Vector Quantization）为了实现鲁棒量化，本文还提出了差分矢量量化（Differential Vector Quantization），它利用为标准分布设计的通用矢量量化器（Universal Vector Quantization）对预测误差信号进行矢量量化。（5）正交变换域矢量量化的快速搜索算法为了实现矢量量化的快速搜索，提出了正交变换域的部分距离算法和最近邻搜索算法。对于在变换域功率分布集中的图像信号等信号源，这些算法比树搜索算法需要更少的乘法运算。

In this project, image coding schemes using vector quantization are systematically studied. The results of this research are as follows.(1)Consideration concerning Vector Quantization fo Picture SignalsTwo problems are pointed out in application of vector quantization. One is the complexity of vector quantizer and the other is robustness of vector quantizer performance.(2)Discrete Cosine Transform Coding with Vector Quantization (DCT-VQ)A new vector quantization scheme using discrete cosine transform, named DCT-VQ is proposed. To arrive at an universal vector quantizer, vector quantizers are designed for a standard probability distribution independent of picture signals. The complexity is greatly reduced because of a decomposition of transform domain into vectors.(3)Adaptive DCT-VQAdaptive DCT-VQ is presented, which employs adaptive control of coding parameters according to the local property of picture signals in order to improve the performance of DCT-VQ. The simulations have shown that the adaptive DCT-VQ outperforms the conventional adaptive transform codings. An application of this scheme to color image coder using color coordinate conversion is also presented.(4)Differential Vector QuantizationTo realize robust quantization, differential vector quantization is also proposed, which vector-quantizes predictive error signals using universal vector quantizer designed for standard distribution.(5)Fast Search Algorithms for Vector Quantization in orthogonal Transform DomainIn order to attain fast search at vector quantization, a partial distance algorithm and nearest-neighbor search algorithm are proposed, both in the orthogonal transform domain. These algorithms needs less multiplications than the tree-search algorithms for signal sources such as picture signals which have concentration in the transform domain power distribution.