The Research And Implement Of The Porn Video Detection By The Clasification Using Both Audio And Video Features

With the popularization of computer network, a large number of porn videos which do much harm to teenagers are massively spread and downloaded. So we urgently need to develop a monitoring system according porn videos to promote the healthy development of Internet environment and protect customers from the harm of porn videos. In this paper, we propose an approach to detecting porn video by fusing video and audio information.In the video domain, this paper proposes a pornographic frame identification method using SVM (Support Vector Machine). A feature vector is formed based on the output of the skin detector and foreground detection, and then a SVM classifier is trained on a set of labeled feature vector. As a result, we obtain the detector which is used to detect pornographic image frames. In the audio domain,this paper proposes an audio identification method based on classification tree, which is used to recognize mute, noise, porno-sound and normal tone. Finally, fusion rules are employed to combine the recognition results from video and audio and then the video shot's pornographic score is generated. Experimental results demonstrate that our framework is more robust than the one which uses either video features or audio feature alone.