基于多层核心集凝聚思想的视频关键帧提取

摘要

针对当前传统的关键帧提取算法准确度低，视频检索的查全率和查准率不高的问题，提出一种基于多层核心凝聚思想的视频关键帧提取算法。利用K-medoids算法优化多层核心集凝聚算法，用真实数据作为聚类中心的特性，对 MUL-CA算法的凝聚粗化和凝聚细化两个重要过程进行改进，用其替代原粗化过程得到的顶层核心集，设计一种多层核心集凝聚算法(IMULCA)，实现顶层核心集的快速准确定位，适当减少凝聚层数，简化算法的计算复杂性。将改进算法应用到视频关键帧提取中，实验结果表明，改进算法相对于原有算法能够更加有效地对视频关键帧进行提取。%The traditional key frame extraction algorithm has low accuracy,video retrieval recall and precision.To solve the problem,a video key frame extraction algorithm based on multi core condensed ideas was proposed.K-medoids algorithm was used to optimize multi core condensed algorithm,and the real data was used as the characteristics clustering center of MULCA algorithm,agglomeration coarsening and condensed refinement were improved,replacing the top core set obtained in the raw coarsening process.A multilayer core set of condensing algorithm (IMULCA)was designed,realizing the fast and accurate posi-tioning of the core set,reducing the aggregation layer,simplifying the computation complexity of the algorithm.The improved algorithm was applied to the video key frame extraction.Experimental results show that the proposed algorithm can extract video key frame more effectively compared with the original algorithm.