Effective Scene Change Detection by Using Statistical Analysis of Optical Flows

Abstract

We present a novel method that exploits the statistical properties of optical flows to find representative video frames that contain scene change moments in video contents. For effective scene change detection, we first divide the optical flows into background and foreground groups. Optical flow is a useful and effective method for tracking object motion between consecutive video frames. By analyzing the variation of optical flows, we can detect rapid scene change between consecutive frames. A scene change probability for each frame is computed by applying some basic statistical methods, such as average and standard deviation. Starting from the selected frames with high probability, we find a clear image that contains no overlapping contents by inspecting the moment that optical flow values changes slowly and steadily. Experimental results show the robustness and effectiveness of our method.