一种基于运动模式识别的视频对象编码算法

An Object-Based Video Coding Using Motion Pattern Recognition

针对块匹配运动估计补偿(ME+MC)算法的不足,提出了一种基于自组织特征映射(SOM)的运动模式识别(MPR)算法,应用于会议电视的视频对象编码.为了进一步提高SOM算法的性能,提出了频率敏感自组织特征映射(FSSOM)算法.实验结果表明,FSSOM-MPR算法较ME+MC算法具有更好的编码性能,在压缩比为180∶1时,重建视频图像的平均峰值信噪比(PSNR)有3.3 dB的改善.

To improve the performance of block matching algorithm in motion estimation and motion complement （ME ＋ MC）, a motion pattern recognition （MPR） algorithm based on self-organizing fea- ture maps （SOM） is proposed and tested in object-based conference video coding. A frequency sensitive self-organizing feature maps （FSSOM） algorithm is proposed to improve the SOM algorithm＇s perfor- mance. Experimental results show that FSSOM-MPR algorithm has better coding performances than ME ＋ MC algorithm. When the compression ratio is 180：1, the peak signal to noise ratio（PSNR） improvement of reconstructed video images is about 3.3 dB.