Research On Erotic Video Detection Based On Skin And Sensitive

Video is the most popular mobile applications now, especially when the3G network is available for people. There are thousands of new videos appearing in internet every day, among which may be some porn videos to be uploaded to the video websites. If porn videos are published by a legal video website, it will cause bad influence on this website, causing some policy risk. So many internet video providers must pay effort to examine every new uploaded video by manual, which causes much workload, and waste a large amount of time. In this situation, implementing a porn video detecting system to help the video examining will ease the verifiers’ load. The verifiers will not have to watch all the videos, and just need to examine those videos judged as porn by the system. It will save much manpower and time, also will increase the efficiency of the video examination obviously.In this paper, we first introduce the problem the video websites are facing, then show the necessity of developing a porn video examine system. Then, we analyzed the existed ways of detecting porn images and videos as well as their defects. Finally, we discuss some potential way for better dealing with the issues. The main work in this paper focuses on:1. Finding the skin-color pixels in the key frames by color information2. Find the texture features of the frames, and filter some non-skin pixels whose color is close to skin.3. Extracting the shape features of the skin-color regions in the images4. Detecting whether the image contains the sensitive human body parts such as breasts5. Detecting whether the video contains periodic motion using optical flow.6. Combining all the information above, using SVM to judge if the video contains adult contents.