Implementation of high performance objectionable video classification system

Sex video files being spread via Web site and P2P networks have caused social problems. It is necessary to prevent more people, especially teenagers, from getting a distorted view of sex through the influence of the obscene content. We propose and implement the obscene video classification system using two visual features to determine a video's lewdness with very excellent performance. In our approach, the two features are first built for each video and a final decision function is made to classify an obscene video. One of the visual features is a single frame based decision variable and the other is a group frame based decision variable. Then the final decision function is developed to optimize high classification performance using the variables by the discriminant analysis. Experimentally we show that the proposed method performs excellently in classifying videos into porno and the other, thus enabling automatic filtering of obscene videos