一种图像空间下的ICM聚类方法
An ICM clustering method based on image space
谢昭 1谢年群 1姚婷婷1
作者信息
- 1. 合肥工业大学计算机与信息学院,安徽合肥230009
- 折叠
摘要
针对图像聚类问题,提出了一种基于图像空间关系的聚类方法,采用场模型描述图像之间的空间关系,利用K-近邻思想构建图像邻域系统,聚类过程中无需手动标记特征表示的图像类别信息,只需要给定初始类别数,通过条件迭代算法(ICM)对图像进行聚类.该文通过实验分析了图像样本大小、图像特征维数、图像特征类型、初始类别标签对聚类结果的影响,通过与多种经典聚类算法进行对比,实验结果充分验证了该方法的有效性.
Abstract
Focusing on the image clustering problem,this paper proposes a new clustering method based on image space relationship.In the paper,the Markov field model is used to build the relationship between image spaces,and the K-nearest neighbor-theory is adopted to construct the image neighborhood systems.Without feature hand-annotation during the clustering process,the method only needs to provide the number of initial categories,then iterative condition model (ICM)algorithm is used to achieve the image clustering procedure.By changing the image numbers,characteristic dimension size,characteristics and initial category labels,we analyze the clustering results comprehensively.By comparing with some other classical clustering methods,the experiments demonstrate that the proposed method outperforms these classical clustering methods on OT scene dataset.
关键词
图像聚类/MRF模型/ICM算法/K-近邻邻域系统Key words
image clustering/MRF model/ICM algorithm/K-nearest neighborhood systems引用本文复制引用
基金项目
国家自然科学基金(60905005)
国家自然科学基金"辅助驾驶车载视频信息的结构场模型与理论研究"(61273237)
出版年
2013
NETL
NSTL
维普
万方数据