Camera break detection by partitioning of 2D spatio-temporal images in MPEG domain

In this paper, we propose a new approach to detect camera cuts and wipes. The approach projects a video into two images representing the spatio-temporal continuity of a video life. Visually, these images are composed by regions of different patterns, each region represents the temporal location of a shot. The shape of a, region boundary inherently classifies cut and wipe, and most importantly, marks the start and end of a wipe sequence. We hence formulate algorithms to locate the color and texture discontinuities occurring at the boundaries of image regions