Shot Boundary Detection Using Directional Empirical Mode Decomposition

In this paper, directional empirical mode decomposition is introduced into shot boundary detection method. In the method, every frame in videos is decomposed by DEMD into a series of Intrinsic Mode Function images and a residue image. The IMF image with the lowest frequency of each frame is removed and a substitute frame is recomposed by the other IMF images and the residue image. These substitute frames are used to calculate differences of adjacent frames in order to avoid mistaken detections caused by abrupt illumination changes. A region-based object tracking method is also applied to avoid mistaken detections caused by object motions. The results of experiments show that the precision of detecting shot boundaries is significantly improved.