Scene-Cut Detection using Best Fitting Kernel and Modified Phase Correlation Methods

In this paper, a novel scene-cut detection system based on best-fitting kernel and modified phase correlation is proposed. Initially, the best-fitting kernel coefficients are obtained for two consecutive sub-sampled image frames in least square sense using the best-fitting kernel method. This kernel is then used to estimate frame t+1 from frame t. Then, modified phase correlation is applied to the estimated and the orginal frame for t+1 as a decision metric and a double threshold strategy is applied to get candidate scene-cuts. Finally the candidate scene-cuts are applied to a false frame removal algorithm to prevent false detections.