论文部分内容阅读
【目的/意义】随着社会信息化水平的不断提高,视频资源逐渐成为数字图书馆资源建设中的重要组成部分,如何从海量视频资源中快速准确地检索到用户需要的视频信息已成为数字图书馆亟待解决的问题。【方法/过程】针对该问题构建了一种基于镜头的视频检索框架。该框架通过镜头边界检测实现视频分割,然后运用3-D Surfacelet变换对视频镜头进行2层分解,通过对各高频方向子带求均值和标准差,生成镜头的特征向量,再计算所有镜头特征向量的均值生成视频指纹,最后利用欧式距离来度量视频的相似度,再在相似视频里检索相似镜头。【结果/结论】通过实验验证表明该方法具有较好查全率和查准率,并且对数字图书馆中常见的视频编辑变换具有较好的鲁棒性。
[Purpose / Significance] With the continuous improvement of the level of social information, video resources have gradually become an important part of the construction of digital library resources. How to quickly and accurately retrieve the video information that users need from the vast amount of video resources has become a digital book Urgent problem to be solved. [Method / Procedure] A lens-based video retrieval framework was constructed for this problem. The framework uses the 3-D Surfacelet transform to decompose the video shots in two layers. The average and standard deviation of the subbands in each high frequency direction are used to generate the eigenvectors of the shots, and then all the shots are calculated The mean of the vectors is used to generate the video fingerprints. Finally, the Euclidean distance is used to measure the similarity of the videos, and the similar shots are retrieved from similar videos. [Results / Conclusion] The experimental results show that the proposed method has good recall and accuracy, and is robust to the common video editing and transformation in digital library.