Kirsch Directional Derivatives Based Shot Boundary Detection: An Efficient and Accurate Method

Abstract

The segmentation of video into its basic structural unit is an essential step for video content analysis. In this paper, we propose a shot transition detection approach based on Kirsch directional derivatives. The ability of Kirsch operator to capture the maximum gradient value amongst different orientations is explored for the task of shot transition detection. Each frame of a given video is convolved with Kirsch operator in eight orientations and compute first and second order moments. These moments are used as features for shot boundary detection. Experiments are conducted on a subset of TRECVID dataset and comparison is done with some common shot boundary detection algorithms to exhibit the performance of the proposed approach.