摘要
针对基于内容的视频检索中的镜头聚类问题,采用了一种基于模拟退火思想改进的K均值聚类算法.该方法提取视频帧的时间信息、均值、方差、偏度和信息熵等颜色直方信息作为特征,利用模拟退火算法全局寻优的能力来改善K均值聚类易陷入局部极值的缺点,从而提高视频镜头聚类的准确性.理论分析和实验结果表明该方法是一种有效的视频镜头聚类算法.
This article proposes a simulated annealing K-means clustering algorithm for video shots,which is very useful and important in CBVR(Content-based Video Retrieval).In the algorithm,time features and color features that containe mean values,variance,skewness,entropy of information and so on were used as feature vectors.Then we use the global optimization ability of simulated annealing to remedy the local extremum shortcoming of K-means,so that we can increase the clustering accuracy for video shots.The theoretical analysis and experimental results witness that our algorithm is an efficient shot clustering algorithm.
出处
《哈尔滨理工大学学报》
CAS
北大核心
2010年第6期13-16,共4页
Journal of Harbin University of Science and Technology
关键词
基于内容的视频检索
模拟退火
视频镜头聚类
K均值聚类
content-based video retrieval
simulated annealing
clustering algorithm for video shots
k-means