Row Echelon based Spectral Clustering Framework for Key Frame Extraction

Video key frames are the abstraction of content rich frames of a shot or a video that reflects the nature of the whole video without redundancy. Key frame extraction plays a vital role in video summarization, content based video retrieval, video indexing, etc. Key frame extraction techniques utilize the features such as color, texture, edge, motion, object, events, etc. Among the existing key frame extraction techniques, the performance of the cluster based key frames extraction techniques is better due to its computational simplicity. However, the missing temporal order of the frames results in poor accuracy. To enhance the accuracy, this paper proposes a Row Echelon based Spectral Clustering framework (RES C) which includes the temporal grouping constraints by means of permutation matrix. The proposed technique is tested on the benchmark datasets viz. VSUMM, UCLA, VIRAT, TvSum, and WEB datasets and the accuracy is improved to 90%.