Technology Of Content Based Skin Detection In Videos

With the development of Wide-band network,video compression and storage technologies,video,as a media style with huge information,is being applied more and more widely. Furthermore,it's spreading speed becomes much faster and the spreading rage becomes much wider. However,while we are enjoying these benefits from videos,some deleterious videos are spreading more and more widely,including some erotic and exposure scenes. So,it's necessary for us to recognize and control these videos. The content based video skin detection system presented in this paper is designed to recognize skin regions in videos,compute the skin exposure degrees and provide some brief judgments.Firstly,we do video segmentation and key frame extraction. Through the segmenting video into shots,we can represent the visual content of every shot with key frames abstractly. We can divide every frame into Similar Chromatic Regions,Luminance Changed Regions and Chroma Changed Regions and resolve the problem of inaccurate segmentation within luminance changing conditions. Based on this method,we also design a key frame extraction algorithm,which can transform the problem of recognition in videos to the problem of recognition in key frame images.The key frames are detected using the motion features,color features and texture features of skin in key frame images. Firstly,we directly use the motion vectors of macro-blocks defined in MPEG-I/II compressing standards and filter the immobile macro-blocks. Then,we build a skin color model in YCbCr color space using the convergent property of skin color,and we present the Gaussian Model Skin Recognition Method and Positive-Negative Look-up Table Method in details. And we analyze the texture of skin after wavelet transform and present a Bayesian Method Based Texture Recognition Method and a High Texture Filtering Method. The result of experiments shows that integrated application of skin's color and texture features can improve the skin recognition in images.Based on the foresaid stages,we do some discussions on how to estimate the video skin exposure degree and present an algorithm. Also,we indicated that such an exposure degree detection method is to detect "erotic" videos,a very high semantic concept.Finally,we present an experimental system and introduce the process and result of detection. The results shows that this method can effectively recognize skin regions in videos and detect the skin exposure degree successfully.