基于神经网络的多级矢量量化图象编码

基于神经网络的多级矢量量化图象编码 MULTI-STAGE VECTOR QUANTIZATION IMAGE GODING BASED ON NEURAL NETWORK 程道伟 Cheng Daowei 1 first-author 陈淑珍 Chen Shuzhen 1 张梦涛 Zhang Mengtao 1 王淑融 Wang Shurong 1 武汉大学空间物理与电子信息学系 武汉大学空间物理与电子信息学系 Deocrtment of Soace Physics and Electronics Information,Wuhan University Deocrtment of Soace Physics and Electronics Information,Wuhan University 本文用改进的Kohonen神经网进行码书设计,并对误差图象再编码传送,形成多级矢量量化编码结构,算法简便,并可在码率低于1.0bpp的情况下得到较好的图象质量。 Vector quantization image coding is more efficient than conventional scalar quantization image coding. The key problem of this technique is how to form the codebooks. In this paper, the codebooks are designed using an improved Kohonen neural network, and the error images are also coded and transmitted thus a multi stage image coding structure is formed. This method is very sim ple, and can offer high quality image when the bitrate is below 1, Obpp. 矢量量化 图象编码 码书 神经网络 码字 1992-01-01 2021-04-01 1