A robust shot transition detection method based on support vector machine in compressed domain

In this paper we propose a new algorithm for shot transition detection. A multi-class support vector machine (SVM) classifier is constructed to differentiate frames of a video into three categories: abrupt change, gradual change and non-change. This approach enables us to integrate many kinds of features into a uniform structure and to eliminate arbitrary selection of thresholds. To enhance the robustness of the algorithm, we form the feature vector from all frames within a temporal windows, each frame represented by six features in compressed domain. Experimental results on TREC-2001 video data set have shown that the result of our algorithm is 8% higher than the best result of 2001 TREC evaluation in F1 comparison when cut and gradual changes are both considered.