Video summarization using shot segmentation and local motion estimation

The video used by professionals from various fields and the general public is constantly growing, it's important to develop efficient systems to classify and index them. The video summarization is an essential step to construct a system of indexing and searching video, it's the process to create a short video presentation and a global vision on the important scenes of the video. This summarization consist to extract keyframes to present video content, it's similar to extract the keywords from text document. Mainly summarization methods apply clustering algorithms to all video frames. There are computations expensive because they use a dissimilarity matrix which implies a quadratic calculation. In this context, we present a new technique for extracting video summary using an adapted shot segmentation. We apply a hierarchical clustering in each shot to extract the keyframes; the number of these keyframes is proportional to variations and movements in the shot. We propose a local motion estimation, we use a co-occurrence matrix to measure the dissimilarity between shot frames and to take account the motion in the shot.