Design And Implementation Of Apparel And Race Detection In Sensitive Video

With the rapid development of the Internet industry,especially the appearance of network video brings a lot of convenience to people’s life, work and study. However, along with these developments brought many discordant things, especially now on the Internet jurisprudence, involving violence, terrorism, porn etc,such sensitive videos have brought many security hidden danger to the society and brought a bad influence to non adults’physical and mental health. These sensitive video have a stronger temptation and agitation to other media. According to incomplete statistics, only a year by 2015 the East Turkistan Islamic movement released more than 100 violent terrorist videos through various channels to the Internet, which brought a very bad influence to the stability of the state and society.Based on this background our laboratory team develop a complete sensitive video recognition system. The author implements two modules of this system:apparel detection and race detection.The author uses the knowledge of the field of computer image processing, and uses the algorithm of pattern recognition to detect the kinds of costumes and the human race from the video, and the effectiveness of the method is verified by experiments.Apparel and race detection module mainly contain three steps:video preprocessing,positioning and identification. In the video preprocessing the author list some common shot segmentation and key frame extraction algorithm and through the experiment to compare. The positioning step is to locate the position of the dress or the human face in the image frame from the key frames extracted from the video. Recognition process for the detection of apparel the first step is to get characteristics of key areas of the apparel and then compared with the feature library in order to detect the apparels’type and similarity, for the race detection the first step is face correction and then extract the characteristics of the key area of the face.at last use the trained classifiers for binary classification (violent or non-violent). In this paper, the author discusse the implementation process of the two modules.Through the analysis of the experimental results of these two modules, the algorithm can be used in the two modules to detect the types of clothing and the human race for two classification, and both of them have a high degree of accuarcy.The result of these two modules also provide a strong basic for the decision of the whole set of sensitive video recognition system.