Human Face Detect Based Video Compression

Video compression plays an important role on digit moving picture processing ,since it affects the picture's quality ,comprcssing rate and visual degree in video sequence . Adopting the same aphesis on the different objects, this approach which has employed by MPEG-2 standard has led to a blur image of key objects,especially for those of the human faces .By means of studying the picture characters of the monitor system , the human -face-central video compressing architects are proposed . Firstly, with the application of both Morphologic Translation and human body feather analysis on binary graph ,the human body contours are extracted by exploited moving information , producing perfect human face region segments ; Secondly , in order to form an accurate border ,the author presents an improved statistical color model, which has removed redundancy successfully; Finally, a high compression rate is achieved by way of combining wavelet transform and different chain codes.Some novel enhancing Morphologic Translation algorithms such as Mark , AreaSelect, Bylsland, Island are implemented in the experiment of this thesis. And what's more, making full use of overlapping movement compensation , embedded zerotree wavelet translation encoding and contend-based wavelet translation , this thesis has achieved the human-face-central video compression of the key objects.Compared with MPEG-2 standard at the same data rate ,this approach we have proposed has improve PSNR nearly 3dB and the video stream brought out by our arithmetic produces a perfect input for human face recognition which will provide convenience for contend-based searches considerably.