一种新的基于自组织神经网络的运动估计算法

A New Motion Estimation Algorithm Based on Self-Organizing Neural Network

提出了一种新的基于自组织网络的CFSSOM-VQ运动估计算法,新的帧间预测编码方案采用基于自组织特征映射算法(SOM)的矢量量化(VQ)作为帧间预测,以取代目前常用的运动补偿帧间预测(ME+MC).并对SOM算法进行了改进,提出了一种分类频率敏感自组织特征映射(CFSSOM)算法.将该算法应用到会议电视视频编码的实验结果表明,与ME+MC算法相比,CFSSOM-VQ算法具有更好的预测编码性能.

To improve the performance of interframe prediction coding, this paper presented a new motion estimation algorithm based on self-organizing feature map （SOM） algorithm. We used vector quantization （VQ） to predict present video frame instead of the general motion estimation and motion compensation （ME ＋ MC） algorithm. Then a classified frequency sensitive self-organizing feature map （CFSSOM） was proposed for the codebook training. Experiment results showed that CFSSOM-VQ algorithm had better prediction coding performance than ME ＋ MC algorithm for conference video coding.