基于两次检测曲线的压缩视频关键帧提取

摘要

The traditional key frame extraction algorithms under compressed domain have the problems that the selected feature is single, the accuracy of the extracted key frames is low, and the efficiency of the algorithms' performance is not high. This paper proposed a key frame extraction algorithm under the compressed domain based on curvature detection. The proposed algorithm denoted the significant changes of the curve using high curvature points on the curve. Then, it constructed the feature similarity curve by using the AC coefficient and DC coefficient of DCT which were the inherent characteristics of compressed video. Finally, the introduced algorithm extracted key frame by performancing two detection of the feature similarity curve. The experimental result shows that the proposed algorithm can automatic extract key frames fast and more efficient, and it can also improve the precision and recall of the key frames.%针对目前压缩域下提取视频关键帧的算法存在特征选取单一、提取的关键帧准确性不高、算法效率低的缺点,提出了压缩域下基于两次曲线曲率检测的关键帧提取算法.算法利用曲线上的高曲率点表示曲线的显著变化,并在此基础上利用压缩视频的固有特征,即离散余弦变换之后的AC系数和DC系数特征,构建特征相似度曲线,进而对曲线进行两次高曲率点检测并提取视频关键帧.实验表明,该算法能快速有效地实现关键帧的自动提取,并可以提高提取关键帧的查准率和查全率.