Existing clustering methods have the problems when extracting the key frames, such as the division sensitively and unable to express the high-order correlations of a large number of frame images in a shot. In view of this, in the paper we propose a key frame extraction method based on the probabilistic hypergraph clustering. Firstly, a shot frame probabilistic hypergraph is constructed, then the clustering learning algorithm of probabilistic hypergraph spectrum is applied to the frames in the shot for clustering, and finally the frames in each clustering centre are selected as the key frames of the shot. Experiments show that the proposed method is computationally simple and has high accuracy in extracted key frames; moreover, the extracted result can better reflect the main content of the video.%现有聚类方法在提取关键帧时存在着划分敏感、无法表达镜头内大量帧图像高次相关关系等问题.鉴于此,提出基于概率超图聚类的关键帧提取方法.该方法首先构建镜头帧概率超图,然后使用概率超图谱的聚类学习算法对镜头中的帧图像进行聚类,最后选取各聚类中心的帧图片作为该镜头的关键帧.实验表明,该方法计算简单,所提取的关键帧准确性高,提取结果能够更好地反映视频的主要内容.
展开▼
