Multidimensional comparison of shot detection algorithms

Shot change detection in video sequences mainly involves two tasks: a numeric evaluation of the degree of disparity between temporally related frames, and deciding whether a value or a series of values of that disparity is or is not a shot change. This paper focuses on the first task, and presents a methodology for comparison between different criteria to evaluate disparity between frames. The main innovation respect to other comparative studies, is that it allows a multidimensional comparative analysis based on the simultaneous consideration of several criteria. In order to overcome the problems inherent to a multidimensional approach, this method is independent of any decision process, and evaluates the performance of every criterion or set of criteria via a process that results in the obtaining of a. single numerical value, which allows an extensive study and a clear and straightforward analysis of the comparative possibilities of each technique.