Video segmentation using spatial and temporal statistical analysis method

Video segmentation is a critical component for video shot detection, video data analysis and information retrieval. In this paper, a new statistical method for video segmentation based on the spatial and temporal analysis of the video data is proposed. A statistical hypothesis testing framework is described for video shot detection and video sequence segmentation. The proposed approach is based on the salient statistical features utilizing both local and global spatial and temporal information in the video image sequence. The statistical hypothesis testing framework in our approach provides a flexible way of handling the variabilities of the video data, and it avoids the pitfalls of heuristic methods of pre-defined threshold in traditional frame-difference and histogram based approaches. This algorithm is applied to real-time broadcasting news video and movie video clips. Experimental results indicate that the proposed algorithm is more robust than previous approaches. Significant performance improvements are observed and the efficacy of the algorithm is verified on the real unconstrained video sequence.