Characterization of abrupt/gradual video shot transitions as unsmoothed/smoothed singularity

Abstract

The multitude of high-level video operations demands sophisticated low-level video techniques. Detection (or segmentation) of video shot transitions (or boundaries) is one of the crucial low-level operations towards automatic video indexing, video editing, video abstracting or preview, and so on. Since content-based multimedia processing has been the focus of MPEG7, we shall develop a scheme to precisely characterize video shot transitions for further applications. Varieties of shot transition detection techniques have been reported in literature. However, no work was really done by taking the benefit of analyzing shot transitions from the local singularities of the signal. In this paper, we propose to characterize shot transitions from their time-frequency variations. First, the color differences between every two neighboring frames are collected as a discontinuity sequence. Second, the time frequency analysis of 1D continuous wavelet transform is employed to locate and interpret the local singularities of the resultant discontinuity sequence. Results reveal that abrupt transition could be characterized as an unsmoothed singularity while gradual transition could be characterized as a smoothed singularity.