A Key Frame Extraction Algorithm Based on Clustering and Compressive Sensing

Key frame extraction is considered as one of the most critical issues in content-based video retrieval technology (CBVR). In this paper, an efficient key frame extraction algorithm based on unsupervised clustering and compressive sensing is proposed. Firstly, three types of filters with various scales are employed to generate high dimensional feature of each frame in one shot, which will be projected to low dimensional feature by a very sparse random projection matrix that satisfies the condition of Restricted Isometry Property (RIP), and then sub-shot segmentation is conducted by an unsupervised clustering method in order to divide one shot into sub-shot collections, in which each class of clustering represents one sub-shot. Finally, the Bhattacharyya coefficient is used to measure the similarity between frame and class center, the frame with the maximum similarity value is selected as the key frame in each sub-shot. The experimental results demonstrate that the proposed method could extract key frames efficiently and properly.