A practical system for detecting obscene videos

We propose a practical system for detecting obscene videos. Our goal is to identify whether a video is obscene or not whereas the obscenity of the individual frames is of little interest. The main emphasis is on fast processing. We introduce several assumptions for bypassing frames with negligible performance degradation. We extract sizable body regions and determine the obsceneness of a frame based on shape matching using Zernike moments. Experimental results on 1,000 sample videos show that the proposed system is comparatively fast with a remarkable capability to detect obscene videos.