采用自适应聚类的教学视频关键帧研究  

Adaptive AP clustering algorithm based teaching video key frame extraction

作　　者：王华秋[1] 殷志恒[1] 

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《计算机时代》2016年第12期90-94,97,共6页Computer Era

基　　金：国家社会科学基金一般项目(14BTQ053);重庆市研究生教育教学改革研究项目(yjg143090)

摘　　要：视频镜头分割和关键帧提取是当前数字视频系统发展的关键步骤。在AP聚类算法之上做了两点改进:一是在初始相关系数矩阵中增加权重,提高聚类精度;二是自适应调整阻尼系数,提高收敛速度。先利用颜色信息加权和相邻帧间差方法把视频分割成镜头,再利用改进的AP聚类算法对镜头提取关键帧。实验结果表明,所提出的方法有效地解决了关键帧提取方法中耗时高和视觉信息低效的问题。Video shot segmentation and key frame extraction is the critical step in the current digital video system development. In this paper, the AP（affinity propagation） clustering algorithm is improved in the two points： first, the weight is increased in the initial correlation coefficient matrix to improve the clustering accuracy; second, the damping coefficient is adjusted adaptively to improve the convergence rate. Use the weighted color information and adjacent frame difference method to divide the video into shots, and then use the improved AP clustering algorithm to extract the key frames from the shots. The experimental results show that the method proposed can effectively solve the problems of high time consuming and low efficiency of capturing visual information in the key frame extraction methods.

关 键 词：数字视频 镜头分割 关键帧提取 AP聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]