一种基于局部二值模式的关键帧提取方法  

A Method for Key-frames Extraction Based on LBP Feature

作　　者：张芯[1] 谢毓湘[1] 栾悉道[2] 

机构地区：[1]国防科学技术大学信息系统工程重点实验室,湖南长沙410073 [2]长沙大学信息与计算科学系,湖南长沙410073

出　　处：《计算机与现代化》2013年第11期8-12,共5页Computer and Modernization

基　　金：国家自然科学基金资助项目(60802080);长沙市重点科技计划资助项目(k1205045-11)

摘　　要：基于旋转不变均衡局部二值模式,提出一种均衡摘要生成速度和摘要信息量的视频摘要算法。首先,使用预采样方法降低视频处理数据量,在此基础之上提取图像的局部二值模式特征;然后对两帧图像相似度进行分析,获取聚类数目。获取聚类数目之后,使用k均值算法对镜头关键帧进行聚类;最后,使用"重要度"函数评测聚类重要度,从"重要"聚类中选取聚类中心最近帧为摘要关键帧。实验结果表明,该算法生成的视频摘要在保证摘要实时性的同时,提高摘要的信息量,较好地表达了视频的内容。Aiming at making an improvement on the information capacity of video summary under the premise of summary speed, a new method for video summarization based on rotation invariant uniform pattern feather is presented in this paper. First of all, the amount of frames to be processed is decreased with the pre-sampling approach, upon which local binary pattern feature is ex- tracted. Then, the cluster number is generated according to the similarity between each two adjacent frames. With the cluster number, the k-means algorithm is applied to cluster similar frames. Finally, the frame which is nearest to the cluster center is used to construct the video summarization. The importance of cluster is evaluated by the function introduced. Experiment results indicate that the proposed method can making an improvement on the information capacity of video summary under the premise of generation speed, the summary performs well to the source video.

关 键 词：局部二值模式 K均值算法 视频摘要 聚类分析 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]