Content-Based Video Segmentation using Statistical Motion Models

We present in this paper an original approach for content-based video segmentation using motion information. The method is generic and does not require any knowledge about the type of the processed video. Its relies on the analysis of the temporal evolution of the dynamic content of the video. The motion content is characterised by a probabilistic Gibbsian modelling of the distribution of local motion-related measurements. The designed statistical framework provides a well formalised similarity measure according to motion activity that we exploit to derive criteria for segmentation decision. Then, the considered merging criteria are sequentially applied between every two successive temporal units of the video to progressively form homogeneous segments in term of motion content. Experiments on real video documents demonstrate the ability of the proposed approach to provide a concise and meaningful overview of a video.