中文摘要：

提出一种利用重构的静态关键帧和动态关键帧来表示视频镜头内容的检索算法．该算法充分利用视频序列在时间轴上的特性构建静态关键帧，并利用三维小波变换构建动态关键帧，将运动信息和静态信息相结合以更好地反映视频序列的内容．实验结果表明，该算法的性能优于基于时域最大值关键帧的算法及其改进算法．

英文摘要：

A video retrieval approach based on reconstructed frames is proposed. The approach makes full use of the representative characteristics along time axis to construct static frame, and utilizes three dimensional wavelet transform to create the dynamic frame. The proposed method provides a better representation of the video content by combining the motion features and static information. Experimental results illustrate that the proposed approach outperforms the temporally maximum occurrence frame （TMOF） algorithm and its variations.