Effective prediction mode decision scheme concerning scene switch

In HEVC, prediction mode decision is essential for efficient coding in applications. While many fast algorithms have been designed to reduce the candidate modes, the features of scene switch frames have seldom been considered. These frames usually have no relationship with the previous frames. Thus, intra modes should be used. In this paper, we propose a simple and efficient Hash-based algorithm. It evaluates the motion intensity of videos and can detect the occurrence of scene switch efficiently. Then chooses the best prediction modes among intra ones directly for CUs in these frames. Experimental results show that it achieves 42% to 53% time saving of the switch frame prediction. Meanwhile, as the prediction time of a switch frame is about 5 to 13 times over that of a normal continuous frame, the algorithm can achieve an outstanding improvement in the prediction of the whole video, especially when dealing with multi-scene videos.