Hierarchical Keyframe-based Video Summarization Using QR-Decomposition and Modified k-Means Clustering

We propose a novel hierarchical keyframe-based video summarization system using QR-decomposition. Specially, we attend to the challenges of defining some measures to detect the dynamicity of a shot and video and extracting appropriate keyframes that assure the purity of video summary. We derive some efficient measures to compute the dynamicity of video shots using QRdecomposition, and we utilize it in detecting the number of keyframes that must be selected from each shot. Also, we derive a theorem that illustrates an important property of QR-decomposition. We utilize it in order to summarize video shots with low redundancy. The proposed algorithm is implemented and evaluated on TRECVID 2006 benchmark platform. Compared with other algorithms, our results are among the best.