Representative and diverse video summarization

Video summarization usually refers to produce a summary preserving essential content of the original video. Many existing methods have been developed to select representative frames by a dictionary learning model, which have led to a state-of-the-art performance. However, learning dictionary without considering relationship between samples of the original data space would lead to imprecise representation. To address this problem, in this paper, geometrical distribution information of samples is incorporated into the dictionary learning process. A graph based learning strategy is employed to draw the geometrical distribution information. Meanwhile, the diversity criteria is considered as important as representativeness, which can reduce redundant frames to be selected in final summary. Thus similarity measuring is imported to guarantee that a final summary contains diversity contents within the original video. The proposed method is validated on a challenging and widely used dataset, and state-of-the-art performance is achieved in contrast to other methods.