Research On Audio-Visual Feature-Based Video Content Detection

With the development of modern computer and internet technology, video will be the main carrier for information and will spread more broadly and rapidly, which will enhance people's life quality, but at the same time, it brings new risks, especially for the younger. The possibility for them to come into contact with pornographic videos is much larger. So, this paper researches the features of this kind of videos, and gives out a method to recognize and filter it.Based on the features of video, which haves both visual and audio messages, we proposed an audio-visual feature-based framework. Audio and visual features are analyzed and extracted independently, and their results are combined to give out a final detection result.In visual field, our research focuses on skin detection. Firstly, we import skin color model, lighting compensation and texture model to abstract skin message from a frame image, and then build a skin color mask. Secondly, to different images with pornography from normal images, we analyze their skin distribution, and abstract six effective parameters for the following classification. In audio field, our research focuses on well-regulated sexy female-voice. Firstly, we segment the audio stream with audio short-time features, to get all dubious audio sects, and then, analyze and abstract some segment features which can different sexy female-voice...