Design And Implementation Of A Pornographic Video Detection System

The rapid progress of Internet brings a lot of convenience to our life and working,however, the spreading of bad content on the internet become more convert anddiversified. The spreading of bad content on the internet (especially pornographic video)severely impede the healthy growth of Internet industry, debase the social climate. How tosuppress the spreading of bad content on the Internet and to effectively detect the badcontent has become an important task.In this thesis, we design and implement a system to detect pornographic video. It iscomposed of three parts: Key-frame extract module, Skin-detect module and Porn-detectmodule. First of all, the system extract key frame from the video according to a certaintime interval. Then the key frame is detected by the Skin-detect module, the frames thatwith a small coverage of skin color are considered as healthy pictures so that can befiltered out and pictures with a large skin color coverage are sent to the Porn-detectmodule for further detection. In Porn-detect module, the system use picture detectiontechnique based on SIFT and Bag-of-visual-words to detect pornographic images. Oursystem considers the ratio of pornographic images detected and the whole picturesdetected as a judgment for pornographic video, and give a final result for a video.Function test shows that our system can work properly with Key-frame extractmodule, Skin-detect module and Porn-detect module. Performance test shows that thoughthe allocation of sampling rate of Key-frame extract module and the arrangement ofSkin-detect module, the time consumption for detection can be obviously reduced whilethe detection accuracy is not greatly influenced.