Detection of Violent Video with Audio-Visual Features Based on MPEG-7

In order to solve the problem that there isn’t an effective way to detect the violent video in the network, a new method using MPEG-7 audio and visual features to detect violent video was put forward. In feature extraction, the new method targeted chosen the features about audio, color, space, time, motion. Parts of MPEG-7 descriptors were added and improved: instantaneous feature of audio was added, motion intensity descriptor was customized, and a new method to extract dominant color of video was proposed. BP neural network optimized by GA was used to fuse the features. Experiment shows that these selected features are representative, discriminative and can reduce the data redundancy. Fusion model of neural network is more robust. And the method of fusing audio and visual features improves the recall and precision of video detecting.