【作者】 张维刚； 黄庆明； 何辉； 王孝朋； 高文；

【Author】 Weigang zhang1, 2 +, Qingming Huang3, Hui He1, Xiaopeng Wang1, Wen Gao2,4 1 School of Computer Science and Technology, Harbin Institute of Technology at Weihai, Weihai 264209, China 2 School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001, China 3Graduate University of Chinese Academy of Sciences, Beijing, 100190, China 4Institute of Digital Media, Peking University, Beijing, 100871, China

【机构】 哈尔滨工业大学(威海)计算机科学与技术学院； 哈尔滨工业大学计算机科学与技术学院； 中国科学院研究生院； 北京大学数字媒体研究中心；

【摘要】 视频关键帧提取的目的是用尽可能少的视频帧来表征尽可能多的视频内容,减少冗余视频帧,降低计算量,以便于视频的快速浏览、内容摘要、索引和检索。本文提出了一种通用的基于无监督合并聚类的视频关键帧提取方法。首先把视频划分为镜头,再对各镜头内的视频帧进行合并聚类。每次选择最相似的两帧或两个子类进行合并,此过程循环进行,直到满足合并停止规则后自动停止合并,得到视频帧聚类结果。然后对聚类结果进行筛选,选出包含足够多视频帧的、能够表征镜头大部分内容的关键聚类。最后从关键聚类中选择与同类中其他所有视频帧最相似的一帧作为关键帧。实验结果表明该方法能够快速有效的提取出视频关键帧。更多还原

【Abstract】 The aim of video key frame extraction is to use as few video frames as possible to represent as much video content as possible. In this way, redundant video frames and the computation are reduced for facilitating the quick browsing, summarization, indexing and retrieving of videos. In this paper, a general and unsupervised clustering method is adopted to extracting key frames. Firstly, the video clip is divided into shots. Then, the video frames in each shot are clustered by merging. Every time the two most similar frames or clusters are merged into a new cluster to instead of the original ones. This procedure is repeated until the merging stop criterion is satisfied. The merging is stopped and the clustering results are obtained. After that, those achieved clusters which have enough frames and cover most content of the shot are selected as key cluster. For each key cluster, the frame which is closest to all other frames in the same cluster is selected as key frame. The experimental results show the efficiency of the proposed method.更多还原