Shot Boundary Detection Based on SVM Optimization Model

Categorizing the consecutive video frames into shots is the first step for content-based video retrieval. Recently, more and more research has made use of support vector machine to improve the performance of shot boundary detection. However, there has not been a uniform standard for selecting parameters of support vector machine kernel so that it relies on numerous experiences to try, which is not only time-consuming, but also can hardly obtain satisfactory results. In this paper, two novel algorithms for shot boundary detection are proposed, which based on support vector machine optimized by particle swarm and Tabu search respectively. The features are organized into a multi-dimension vector by using the method of sliding window. Experimental results show the effectiveness and robustness of the proposed algorithms, and the performance of support vector machine optimized by Tabu search is better than that of Particle swarm optimization algorithm.