Visual significance model based temporal signature for video shot boundary detection

Video shot boundary detection (VSBD) is the fundamental step for video processing algorithms. The goal of any VSBD algorithm is to detect the transitions (abrupt or subtle) in the given video precisely. In this paper, a visual significance model that is suitable for describing transitions in video is introduced. The proposed visual significance model is composed of parameters like color, texture, edge, motion and focus computed over the frames in the video. Once the visually significant region is identified in each frame of the video, the temporal signature is generated through the dissimilarity measure of the visual significance model. The temporal signature is further examined using standard Random Vector Functional Link (RVFL) networks for categorizing the transitions as Abrupt Transitions (AT), Subtle Transitions (ST) and No Transitions (NT). To validate the performance of the proposed visual significance model based VSBD Framework, it is evaluated on benchmarks like VIDEOSEG2004 and TRECVID2001 to detect and categorize the transitions. Comparison of F1-Score measure with prominent early methods reveals that the proposed framework is a promising model for detecting the transitions in videos even in the presence of varying illumination conditions, fast camera and object motion.