基于压缩传感和EMD距离的视频镜头关键帧提取

关键帧提取是视频内容分析与检索技术的核心问题。提出一种基于压缩传感和EMD距离的关键帧提取方法，首先构造一个符合有限等距性质的稀疏矩阵，将帧高维特征投影到低维空间，然后通过计算帧低维特征之间的调节余弦相似度完成子镜头分割。在各子镜头中，利用EMD距离计算帧与子镜头中心的差异，并选择差异最小值所对应的帧作为该子镜头的关键帧。实验结果表明，该方法提取的关键帧能够对视频内容进行准确的描述。

Video Shot Key Frame Extraction Based on Compressive Sensing and EMD Distance

Key frame extraction is the key issue of video content analysis and retrieval technology. A key frame extraction method based on compressive sensing and Earth Mover's Distance is proposed. Firstly, high dimensional feature of each frame is projected to low dimensional space by a sparse matrix that satisfies Restricted Isometry Property (RIP), and then sub-shot segmentation is accomplished by computation of adjusted cosine similarity between low dimensional features. In each sub-shot, EMD distance is used to compute the difference between sub-shot center and each frame, the frame with minimum value is selected as the key frame. The experimental results show that key frames extracted by the proposed method could precisely describe video contents.