基于镜头的鲁棒的视频广告检测

Automatic detection of commercials embedded in digital video materials is a challenging task with development of retrieval of multimedia. However, because of diversity of commercials, the performances of detection systems are not good enough. In this paper, a Three-phase system for commercial detection with its own characteristics is constructed to improve the robustness of the detection system. Firstly, two Region-based Feature Importance schemes are proposed to detect hard cut and dissolve cut shots respectively and some statistical features for shot characterization are also extracted. Secondly, a SVM classifier is applied to classify these shots. At last, in order to make the commercial segments more accurate, we use a fusional continuance of time and content of commercial shots to eliminate false positive commercial ones. After that, commercial shots are integrated into the commercial sequences. The final test results on 30 clips of TV videos show the practicality of the constructed system.