Scene change detection on H.264/AVC compressed video using intra mode distribution histogram based on intra prediction mode

This paper presents a novel scene change detection method using complexity and edge direction based on intra prediction mode in H.264/AVC compressed video. When scene change occurs, there is less temporal correlation between consecutive frames. Therefore, most of macro-blocks are encoded in intra mode. Using this property, the proposed method calculates the percentage of intra mode blocks in each predictive frame in order to get candidates of scene change frames. To detect accurate scene change frames, we generate the Intra Mode Distribution Histogram (IMDH) representing edge directionality and image roughness. By similarity matching of consecutive two IMDHs, we decide scene change frames. The distance between two IMDHs is measured using χ2-statistic. The experimental results show that our method is efficient and robust.