文中根据视觉生理学和心理学特性 ,采用现有人类视觉系统模型及研究结果和一系列独立于自然图像统计特性及与人类视神经元相关的特性 ,把图像分成菱形块。并考虑图像的观察者———人眼的时 -空频率具有各向异性 ,根据韦伯定理设计一个滤波器 ,把图像块分成三个区域 ,运用改进的自适应矢量量化神经网络 (MSOFM )来训练各分类块并编制码书。计算机仿真结果表明 :重构图像的“块效应”明显得到改善。 According to the characteristics of visual physiology and psychology, this paper divides the image into diamonds based on the existing human visual system model and research results and a series of characteristics independent of the natural image statistical characteristics and human visual neurons. Considering the anisotropy of the human eye’s time-space frequency, a filter is designed according to Weber’s theorem, and the image block is divided into three regions. Using improved adaptive vector quantization neural network (MSOFM) Train each classification block and code book. Computer simulation results show that the “block effect” of the reconstructed image is obviously improved.