Summary: | 碩士 === 國立交通大學 === 資訊科學與工程研究所 === 97 === Shot change detection algorithm is a fundamental step in automatic video content analysis, indexing, and retrieval. Shot change detection can apply to many kinds of video technology fields, for example, video codec compression quality and performance and transcode preparation…etc. This thesis proposed a new shot change detection method using edge direction similarity based on intra macroblcok prediction mode and the ratio of intra to inter macroblocks in H.264/AVC compression domain. The proposed shot change detection algorithm includes gradual shot change and cut shot change detection. In gradual shot change detection, we proposed a fade shot change detection algorithm, using intra macroblock ratio and P_SKIP macroblock ratio to decide start point and end point. Then edge direction similarity is used to judge if current detected frame is included in the gradual shot or not. In cut shot change detection, we use P-MB and B-MB ratio to detect candidate cut change frame, and then use proposed edge direction similarity to judge the candidate frame is a cut change or not.
The experimental result shows that our proposed algorithm has a high accuracy in terms of recall and precision for a variety of video streams even they are variable bit rates. Our proposed edge direction similarity approach also can enhance shot change detection algorithms and achieve satisfactory performance.
|