基于MapReduce的抗闪光镜头边界检测

Anti-Flash Shot Boundary Detection Based on MapReduce

利用MapReduce编程模型并行计算的思想,将视频数据切割成若干个视频片段,分配给若干个任务节点主机进行视频特征值的提取和镜头边界检测。考虑到新闻视频中存在大量的闪光镜头,容易造成误检,用滑动窗口对帧间差异进行分析,修正帧间差值。实验结果表明,改进的镜头算法不仅大大提高镜头检测的效率,还可以有效地避免闪光镜头,取得较好的检测精度。

Using the distributed computing of MapReduce model, the video is splinted into several video clips, video feature extraction and shot boundary detection are assigned to several task node hosts. Considering the existence of a large number of flashes in the news video, it is easy to cause false detection, sliding window is used to analyze the differences between frames, then correcting the frame difference. Experimental results show that the improved shot detection algorithm is effective in accelerating the shot detection process, and it can also effectively avoid flash lens, achieve better detection accuracy.