HDTV 编码中利用联想存储和神经网络的运动估计方法  

Associative Memory and Neural Network Approach for Motion Estimation in HDTV Image Coding

作　　者：顾文涛[1] 郑志航[1] 

机构地区：[1]上海交通大学电子工程系

出　　处：《上海交通大学学报》1997年第12期24-28,33,共6页Journal of Shanghai Jiaotong University

基　　金："八五"攀登计划资助

摘　　要：采用以内容寻址的联想存储结构，在保证估计精度的前提下实现高度并行的处理，适合于ＶＬＳＩ实现．在块内存在运动边界时，块匹配法的运动一致性前提不再满足，本文提出利用神经网络模式分类的分割匹配方法，能够提高含运动边界块的运动估计性能．An associative memory architecture is proposed to execute fullsearch matching. The new method can reduce computation complexity greatly without degradation in image quality. Besides, when there exist moving edges inside a block, the motion homogeneity assumption for block matching algorithm is no longer satisfied. To deal with this situation, a segmented block matching algorithm employing neural network for pattern classification is presented, which can hence improve the estimation property.

关 键 词：运动估计 联想存储 图象分割 神经网络 HDTV编码 

分 类 号：TN941.1[电子电信—信号与信息处理]