Video summarization using a visual attention model

This paper presents a method of video summarization based on a visual attention model. The visual attention model is a bottom-up one composed of two parallel ways. A static way, biologically inspired, which highlights salient objects. A dynamic way which gives information about moving objects. A three steps summary method is then presented. The first step is the choice between the two kinds (static and dynamic) of saliency maps given by the attention model. The second step is the selection of keyframes. An “attention variation curve” which highlights changes on frames content during the video is introduced. Keyframes are selected on this variation attention curve. To evaluate the summary a reference summary is built and a comparison method is proposed. The results provide a quantitative analysis and show the efficiency of the video summarization method.