PORNOGRAPHY DETECTION IN VIDEO USING CHARACTERISTIC MOTION PATTERNS

A mechanism for detecting pornographic content in a video based on an analysis of characteristic motions in the video. The motion-based mechanism is to appearance-based and audio-based approaches, and therefore can be employed in combination with such approaches, or by itself. In one implementation, the mechanism employs a first phase of unsupervised learning, and a second phase of supervised learning. In the first phase, characteristic motion patterns are discovered and clustering is performed. In one implementation, characteristic motion patterns are discovered by analyzing the relative displacements of large numbers of ordered trajectory pairs over time. In the second phase, a classifier is trained to associate the clusters identified in the first phase with pornographic content. A motion pattern descriptor (e.g., a feature vector, etc.) for a video (e.g., a newly-uploaded video, etc.) is obtained and is provided to the trained classifier to obtain a pornography score for the video