Automatic System for Filtering Obscene Video

This paper proposes an automatic system for filtering obscene video sequences. At present, an enormous number of video files are available on the various channels such as P2P networks, spam emails, and web sites. Especially recently, many video files are propagated via UCC community sites with the manual filtering process to screen the harmful videos. Thus, automatically recognizing obscene pictures in video files becomes an extremely important problem. In this paper, we propose a new, efficient, and fast algorithm for detecting obscene videos. Our algorithm is based on segmentation. We first extract the frames of a video, detect global motions, separate the foreground objects, and classify each frame as an objectionable or a benign frame based on skin color, texture, and shape features. Our algorithm shows a remarkable capability to detect obscene frames in the video files of various video edits such as UCC contents.