基于人脸检测的视频压缩

本文在研究监控系统的视频图像压缩基础上，提出了给予人脸对象分割的视频压缩。首先利用运动信息分割出入体轮廓，并综合运用人体的特征与形态数学的方法成功地分离出入脸大致区域；然后采用基于改进统计彩色信息模型方法，精确分割出入脸区域，去掉了不相关的冗余信息；最后提出了利用图像小波变换结合差分链码技术描述了人脸对象，并实现了高效的视频压缩。本文提出的视频压缩算法在同等压缩效率下，与MPEG-2算法相比提高了图像PSNR近3dB，而且经过本算法处理的视频序列，可以方便地用于人脸检测，为基于内容的检测打下良好的基础。

Human face detect based video compression

On studying the picture character of monitor system,a human face central video compressing architect is proposed.First human body contour is extracted by exploited moving information,furthermore,applying both Morphologic Translation and human body feather analysis on binary graph produce a perfect human face region segment result.Secondly,in order to form a accurate border,we present a improved statistical color model,which has remove redundancy successfully.Finally,a high compression rate is achieved by combining wavelet transform and differ chain code.The approach we proposed has improve PSNR nearly 3dB compared with MPEG standard at same data rate and the video stream produced by our arithmetic is perfect input for human face recognization which will convenience content-based detection considerably.