Forward-Backward Nonlinear Sparse Dictionary Selection Based Video Summarization

Abstract

The explosion of big video data has posed great challenges in video browsing, managing and storing, such that effective and efficient video summarization is urgently required. Recent years have witnessed the promising advancements of sparse representation based video summarization. In order to explore the nonlinearity among video frames, the nonlinear sparse dictionary selection has been attempted, however, using only the forward strategy cannot remove the poor selections. Inspired by the backward strategy in linear dictionary selection, a forward-backward nonlinear dictionary selection algorithm is proposed for the nonlinear video summarization. Experimental results on a benchmark dataset demonstrate that the proposed algorithm outperforms some state-of-art video summarization algorithms, including the nonlinear algorithm only with a forward strategy.