MCMC-based scene segmentation method using structure of video

Abstract

Video scene segmentation and classification are fundamental steps for multimedia retrieval, browsing and indexing. In this paper, we present a robust scene segmentation approach based on the Markov Chain Monte Carlo (MCMC) method using the structure of video sequences. In our method, there are two novel approaches to segment video sequences into scenes. The first approach is the use of the video structures to extract scene boundary candidates from shot boundaries. Then using the MCMC method to select the true scene boundaries from these candidates, highly-accurate scene segmentation becomes possible. It should be noted that when the prior probability concerning the number of scenes in a target video sequence is given correctly, the MCMC method can provide a more accurate scene segmentation result. Therefore, in the second approach of the proposed method, the parameter utilized in the prior probability is set to the optimal value by using Multiple Regression Analysis (MRA). Consequently, accurate scene segmentation becomes possible by using the above novel approaches. Experimental results performed by applying the proposed method to actual broadcast programs are shown to verify the effectiveness of the proposed method.