A Novel Graph-Based Structural Dissimilarity Measure for Video Summarization

Effective detection of shot boundaries is important for video summarization methods based on shot boundary detection. However, various gradual shot boundaries (such as, fade in, fade out, dissolve) pose a great challenge to shot boundary detection. Previous work constructs graph models based on feature histograms from images and analyzes the structural changes of graphs, thus improving the detection of gradual shots. In this paper, we develop a new quantization method to calculate the structural change of the graph so as to more accurately locate the gradual shot boundaries. Statistical analysis methods are performed to analyze the data to be detected with past data to achieve real-time shot boundary detection. Experimental results on the VSUMM dataset show that our method outperforms some state-of-the-art methods on the F-Score.