Video summarization has received widely attention as a technique to quickly display the main video content.Existing graph model based method takes video frames as vertices and uses the edges to build the relationship between two vertices,which may not well capture the complex relationship among the video frames.To overcome this drawback,we present a novel method based on hyper-graph ranking to generate a static video summarization,and name it Hyper-Graph Ranking based Video Summarization (HGRVS).Specifically,HGRVS first builds a video hyper-graph model to connect the video frames that have internal relations with hyper-edges;then classifies the video frames with an idea borrowed from the hyper-graph ranking method to provide the candidate keyframes;the video summarization is finally determined with the keyframes chosen by an objective function.Extensive subjective and objective experiments on the popular Open Video Project and YouTube datasets clearly demonstrate the superiority of HGRVS to the state-of-the-art approaches.%视频摘要技术作为一种快速感知视频内容的方式得到了广泛的关注.现有基于图模型的视频摘要方法将视频帧作为顶点,通过边表示两个顶点之间的关系,但并不能很好地捕获视频帧之间的复杂关系.为了克服该缺点,本文提出了一种基于超图排序算法的静态视频摘要方法(Hyper-Graph Ranking based Video Summarization,HGRVS).HGRVS方法首先通过构建视频超图模型,将任意多个有内在关联的视频帧使用一条超边连接;然后提出一种基于超图排序的视频帧分类算法将视频帧按内容分类;最后通过求解提出的一种优化函数来生成静态视频摘要.在Open Video Project和YouTube两个数据集上的大量主观与客观实验验证了所提HGRVS算法的优良性能.
展开▼
