Design And Implement Of Violent Animation Video Detection System

In recent years, the animation industry has been concerned by more and more countries, every kind of animation emerge in an endless stream. With the popularity of the Internet, animation is becoming more popular as a kind of media form of broadcast. Not only the source of animation is not limited, the average age of its audiences is also gradually rising. Such a thriving industry has both advantages and disadvantages, many parents started to worry about the influence of animation to their children. In the face of the mass animation video on the Internet, it is not enough to rely on manual recognition. Instead, we need a technology which is able to automatically identify the animation content and judge it whether violence or not. It should filter out the part of unsuitable for young children. For this purpose, we design and realize a violent animation video detection system.Different from the low efficiency of traditional manual identification, the detection system described in this paper applies the computer vision domain knowledge and the pattern recognition algorithm to detect violent on a large number of animation video extracted from the network, using which we got a relatively accurate result in the experiment. The detection process is divided into three parts:database establishing, feature extraction and classification.(1) Animated Database Establishing:Through analyzing the data of the animation which are marked to violence in the major video sites, the main characteristics of violent video animation are summarized. According to these features, I extracted a large number of violent and non violent video clips from a mass of video, as the positive and negative samples in detection and learning of computer.(2) Feature Extraction:In order to make better use of the low dimensional feature in video, I extract feature from two aspects:static and dynamic. According to the characteristics of animation video color characteristics significantly, I choose the color information as image feature extraction. In order to detect violent video more accurately, I use the method combining Pyramid Lucas-Kanande optical flow and motion template to extract video motion information..(3) Classification:Through the experimental analysis to various types of classification algorithms, I decided to use Support Vector Machine(SVM) as the classification algorithm of the detection system. Then I used Cross-validation to study and detect the feature and got the experimental results. Based on the experimental results, the detection system can efficiently divide animate video to violence and nonviolence, and had a high enough accuracy. It can not only avoid the limitation of single feature, but also directional reduce the amount of data, so that we can get more accurate results in the shorter time.