提出一种基于内容的视频检索系统的关键帧提取新算法,把关键帧提取问题建模为一个可以用动态规划算法求解的全局优化问题。首先建立二值的帧差矩阵来表示低维特征空间中帧与帧之间的相似性度量,然后使用动态规划算法分割帧差矩阵从而提取出关键帧。该算法具有低计算复杂度和对于视频内容的自适应性,而且保持了关键帧的时间顺序,可以方便地根据需要调节关键帧数目。
A new key frame extraction algorithm through global optimization using dynamic programming method is proposed for content-based video retrieval system. Binary semantic difference matrix is established to give similarity measure of frames in low- level feature space. Global optimization is then imposed on the matrix to approximate semantic segmentation using dynamic pro- gramming. The main merits of the proposed method include a good trade-off between computational complexity and global temporal segmentation, the maintenance of sequential ordering constraint and the adaptability to extensive video contents. The convenience to adjust the number of key frames considering the storage burden is also satisfactory.