用于图像编码的相关矢量量化研究

当相邻的图像块用矢量量化（ＶＱ）编码时可能出现编码地址相同的情况,尤其是在图像的平滑区。为了减少相邻块间编码地址的相关性,本文提出了一种相关矢量量化方案,采用相关码书与改进的自组织特征映射（ＩＳＯＦＭ）码书同时编码一个窗口内的四个邻域块,与无记忆类ＶＱ相比,对一幅典型的“Ｌｅｎｎａ”图象,编码过程中所需计算量减少一半,比特率减少４０％,由于在Ｋｏｈｏｎｅｎ自组织神经网络的训练过程中,对边缘类矢量采

The Study of Correlation Vector Quantization for Image Coding

Some adjacent image blocks may have the same address when ceding them with vector quantization(VQ),especially in the stationary and uniform regions of an image. This paper proposes a correlation vector quantization scheme to decorrelate the addresses of adjacent coded blocks.Correlation codebook and improved self-organizing feature maps(ISOFM) codebook are sued to encode the four adjacent blocks simutaneously in a window.For a typical image"Lenna",the computational quantity during the coding process is reduced by a factor of 2 and the bit rate is reduced by about 40% over the memoryless VQ. Since preferential measures are taken to the vectors belonging to the edge classes in the training process of Kohonen's self-organization neural network,the subjective quality of the decoded image is significatly improved.