Multi-layer Objectionable Video Classification System Using Local-Global Information

In this paper, we propose multi-layer objectionable video classification system using local information and global information simultaneously. We also analyze the additional information of video files through the Internet for use in the objectionable video classification from a statistical point of view. The proposed system consists of 3 analyzers and uses MPEG-7 visual descriptors as features for content-based analyzer. To gain the local information of a video, we extract 200 representative frames from video file by uniform sampling and classify every frame using MPEG-7 visual descriptors and a SVM. To avoid the misclassification by a few frames, results of frame classification is used as the global information. Experiment results show that the proposed system has an excellent performance in classifying a video as the objectionable or as the unobjectionable. Key-Words: Objectionable video, video classification, support vector machine, content-based analyzer