Video Key-Frame-Extraction Based on Block Local Features and Mean Shift Clustering

Key-frame-extraction has been recognized the important research issue in the content based on video retrieval. And the effectiveness of the key frames will directly influence on video retrieval. This paper proposes a new method of video key-frame-extraction based on block local features and mean shift clustering. Firstly, we partition the image by the block-weighted strategy, and then extract the color moments and the texture feature of each block image, which increases the spatial information, by gray level co-occurrence matrix. Secondly, we extract the key-frame by clustering in the color and texture information space using mean shift algorithm. The mean shift can automatically determine the cluster number and has strict convergence, only to set empirical bandwidth. So, it greatly reduces the computation and avoids human factors. The experiment has proved that the key frames extracted by this method can more accurately describe the content of the shots in details.