Novel shot boundary detection method based on support vector machine

A novel algorithm about Shot boundary detection based on Support Vector Machine is proposed in this paper. The algorithm utilizes SVM, which is trained by using of some features, to classify videos so as to test the change of shots, and realizes shot boundary segmentation by distributing video frames into three categories: Normal Frame, Gradient Frame and Switched Frame. The features adopted here consist of two parts: one is the features extracted from pixel domain which includes mean luminance, brightness variance, edge variance ratio, block histogram and so on, and the other is the ones extracted from compressed domain which mainly involves DC coefficient and motion vector. Experimental results show the novel algorithm possesses good robustness on the motion of camera and the admittance of big objects, and is simpler than most of the other methods.