TV commercial detection using audiovisual features and support vector machine

Automatic detection of commercials is a important step of commercial management, and has multiple potential applications in TV video content analysis. In this paper, we transform the problem of commercial detection to commercial shot detection, and propose a novel method to fuse audio and visual features to detect commercial blocks. An intermediate feature POIM (Program-oriented Informative Images) is introduced, which contains the information of product or program. The contextual features for each shot are generated with time expansion from POIM image probability and other basic audiovisual features. And then, these contextual features are utilized to predict the probabilities of commercial shot using SVM (Support Vector Machine) classifier. Experiments get promising performance on a database from TV channels in China.