基于距离阈值聚类的关键帧提取方法  

Method of Key Frame Extraction Based on Distance Threshold Clustering

作　　者：马振磊 丁昕苗 柳婵娟[1] 李亚庆 MA Zhenlei;DING Xinmiao;LIU Chanjuan;LI Yaqing(School of Information and Electrical Engineering,Ludong University,Yantai 264039,China)

机构地区：[1]鲁东大学信息与电气工程学院,山东烟台264039

出　　处：《鲁东大学学报（自然科学版）》2018年第2期128-135,共8页Journal of Ludong University:Natural Science Edition

基　　金：国家自然科学基金(61303086;61603172)

摘　　要：传统聚类方法在关键帧提取过程中对初始中心敏感,须人工估计聚类阈值.对此,本文提出了一种基于距离阈值聚类的关键帧提取方法,实现了关键帧提取中聚类中心的有效确定和阈值选取.该算法首先提取视频帧的特征,基于特征向量密集采样定义距离阈值,依据帧间相似度和阈值确定初始聚类中心位置和个数;然后使用K-means算法优化初始类心,对优化后位置相近的聚类簇进行合并得最终聚类结果;最后提取距离聚类中心最近的帧作为视频关键帧.实验表明,该方法适应性强,能根据不同视频类型有效地提取出合适数目的关键帧.The traditional clustering method is sensitive to the initial center and needs to set the clustering threshold manually in the process of key frame extraction.Therefore,a key frame extraction method based on distance threshold clustering was proposed in this paper,which realized the effective determination of the clustering center and the threshold selection in the key frame extraction.The algorithm extracted the feature of video frames firstly,the distance threshold was defined based on intensive sampling of the feature vector,and according to the inter-frame similarity and threshold the initial cluster centers position and number were determined.Then,the initial cluster centers were optimized by K-means algorithm.The final clustering results were obtained by merging closer clusters.At last,the frames which were nearest to the clustering centers were extracted as key frames.Experiments showed that the method was not only adaptable but also could extract the appropriate number of key frames effectively according to different video types.

关 键 词：聚类分析 距离阈值 关键帧 特征提取 K-MEANS 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]