基于多层核心集凝聚思想的视频关键帧提取

摘要

Key frame extraction is a key technology in video retrieval.Traditional key frame extraction algorithm has the problems of low accuracy and low recall and precision in video retrieval.For these problems,we proposed a key video frame extraction algorithm which is based on multilayer core agglomeration thoughts.First,we studied the multilayer core set agglomeration algorithm (MULCA)presented in literature [1 ],and utilised the characteristic of K-medoids algorithm-it uses real data as the clustering centre ! to improve two important processes of MULCA algorithm,the roughening of agglomeration and the refinement of agglomeration,we also substituted the original top layer core set derived from roughening process with it,designed a new multilayer core set agglomeration algorithm (IMULCA),and realised the fast and accurate positioning of top layer core set,which can appropriately reduce the agglomeration layer and simplify the computation complexity of the algorithm.Then,we applied the IMULCA algorithm to the extraction of key video frames.Experimental results showed that the proposed algorithm,compared with the previous algorithm,could more effectively extract the key video frame.%关键帧提取是视频检索的一项关键技术。针对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。首先，对文献[1]提出的多层核心集凝聚算法（MULCA）进行研究，并利用K-medoids算法用真实数据作为聚类中心的特性，对MULCA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，设计了一种新的多层核心集凝聚算法（IMULCA），实现了顶层核心集的快速准确定位，并可适当减少凝聚层数，简化了算法的计算复杂性。然后，将IMULCA算法应用到视频关键帧提取中，实验结果表明所提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。