A scene boundary detection method

Abstract

A visual rhythm is a special 2D image reduced from a 3D video data in a way that the pixels along a vertical line of the visual rhythm are the pixels uniformly sampled along the diagonal line of a video frame. Using the distinct visual patterns to appear on the visual rhythm, we propose a video segmentation method to effectively detect both abrupt and gradual shot transitions. Specifically, the algorithms to detect cut, wipe and dissolve transitions are outlined. Note that the proposed method operates on a small fraction of the original video data, thus offers remarkable computational savings.