基于成对约束Info-Kmeans聚类的图像索引方法

doi:10.3969/j.issn.1000-436x.2013.07.018

通信学报 ›› 2013, Vol. 34 ›› Issue (7): 159-166.doi: 10.3969/j.issn.1000-436x.2013.07.018

基于成对约束Info-Kmeans聚类的图像索引方法

刘文杰¹,伍之昂²,曹杰²,潘金贵²

¹ 南京大学软件新技术国家重点实验室，江苏南京 210046;
² 南京财经大学江苏省电子商务重点实验室，江苏南京 210003

出版日期:2013-07-25 发布日期:2017-06-24
基金资助:
国家自然科学基金资助项目;国家自然科学基金资助项目;江苏省省属高校自然科学研究重大基金资助项目;国家科技支撑计划基金资助项目;江苏省自然科学基金资助项目;江苏省自然科学基金资助项目

Image indexing method based on clustering via Info-Kmeans under pair constraints

Wen-jie LIU¹,Zhi-ang WU²,Jie CAO²,Jin-gui PAN²

¹ State Key Lab for Novel Software Technology,Nanjing University,Nanjing 210046,China;
² Jiangsu Provincial Key Laboratory of E-Business,Nanjing University of Finance and Economics,Nanjing 210003,China

Online:2013-07-25 Published:2017-06-24
Supported by:
The National Natural Science Foundation of China;The National Natural Science Foundation of China;Key Project of Natural Science Research of Jiangsu Provincial Colleges and Universities;National Key Technologies R＆D Program of China;The Natural Science Foundation of Jiangsu Province;The Natural Science Foundation of Jiangsu Province

摘要/Abstract

摘要：

针对图像数据噪声大和高维稀疏的特点，提出了一种基于噪声过滤和Info-Kmeans 聚类的图像索引构建方法。首先，利用余弦兴趣模式过滤噪声。其次，提出了一种新的Info-Kmeans聚类算法，该算法不仅避免KL-divergence计算过程中的零值困境问题，还能融合以成对约束出现的先验知识。最后，在LFW和Oxford_5K 2个图像数据集上的实验表明：噪声过滤能显著提高聚类性能；Info-Kmeans比已有聚类工具具有更优越的性能。

关键词: 图像索引, 兴趣模式, 噪声过滤, 聚类分析

Abstract:

Constructing high-quality content-based image indexing is fairly difficult due to the large amount of noise in the data set and the high-dimension and the sparseness of the image data.To meet this challenge,a novel noise-filtering and clustering was proposed using Info-Kmeans based image indexing construction method. Firstly,a noise-filtering me-thod using the cosine interesting patterns was presented. Secondly,a novel Info-Kmeans algorithm was proposed which could avoid the zero-feature dilemma caused by the use of KL-divergence and exploit the prior knowledge in the form of pair constraints. The experimental results on the two image data sets,LFW and Oxford_5K,well demonstrate that: noise filter can improve the clustering performance remarkably and the novel Info-Kmeans algorithm yields better results than the existing clustering tool.

Key words: image indexing, nteresting pattern, noise filtering, cluster analysis

刘文杰,伍之昂,曹杰,潘金贵. 基于成对约束Info-Kmeans聚类的图像索引方法[J]. 通信学报, 2013, 34(7): 159-166.

Wen-jie LIU,Zhi-ang WU,Jie CAO,Jin-gui PAN. Image indexing method based on clustering via Info-Kmeans under pair constraints[J]. Journal on Communications, 2013, 34(7): 159-166.

图/表 12

图1

图2

图3

图4

图5

表1

图6

图7

表2

图8

图9

图10

参考文献 21

[1]	YANG C W , SHEN J J . ecover the tampered image based on VQ indexing[J]. Signal Processing, 2010,90(1):331-343.
[2]	WEINBERGER K , BLITZER J , SAUL L . Distance Metric Learning for Large Margin Nearest Neighbor Classification[M]. Cambridge: MIT Press, 2006.
[3]	GUILLAUMIN M , VERBEEK J , SCHMID C . Is that you? metric learning approaches for face identification[A]. Proceedings of the In-ternational Conference on Computer Vision[C]. Kyoto, 2009. 498-505.
[4]	DAVIS J , KULIS B , JAIN P , et al . Information-theoretic metric learn-ing[A]. Proceedings of the 24th International Conference on Machine Learning[C]. New York,USA, 2007. 209-216.
[5]	CAO J , WU Z A , WU J J , et al . Towards information-theoretic K-means clustering for image indexing[J]. Signal Process ng, 2013,93(7):2026-2037.
[6]	李国波, 陈钢, 吴百锋 . 基于特征聚类的图像错误检测及掩盖算法通信学报[J]. 2010,31(12):1-11. LI G B , CHEN G , WU B F . Error detection and concealment based on characteristic clustering of image[J]. Journal on Communica ions, 2010,31(12):1-11.
[7]	YANG J C , YU K , GONG Y , et al . Linear spatial pyramid matching using sparse coding for image classification[A]. Proceedings of the IEEE Int'l Conf on Computer Vision and Pattern Recognition[C]. Miami,USA, 2009. 1794-1801.
[8]	尹学松, 胡恩良, 陈松灿 . 基于成对约束的判别型半监督聚类分析软件学报[J]. YIN X S , HU E L , CHEN S C . Discriminative semi-supervised clus-tering analysis with pairwise constraints[J]. Journal of Software, 2008,19(11):2791-2802.
[9]	TAN P N , STEINBACH M , KUMAR V . Introduction to Data Min-ing[M]. Boston: Addison-Wesley, 2005.
[10]	WU J J , XIONG H , CHEN J , et al . A generalization of proximity functions for K-means[A]. Proceedings of the IEEE International Conference on Data Mining[C]. Omaha, 2007. 361-370.
[11]	付爱英，曾勍炜，徐知海等 . 基于聚类的关联规则挖掘算法的研究及应用[J]. 通信学报, 2006,27(z1):177-180. FU A Y , ZENG Q W , XU Z H , et al . Research and application of the algorithms for mining association rules based on clustering[J]. Journalon Communications, 2006,27(z1):177-180.
[12]	WU J J , ZHU S W , XIONG H , et al . Adapting the right measures for pattern discovery: a unified view[J]. IEEE Trans on Systems,Man,and Cybernetics, 2012,42(4):1203-1214.
[13]	TAN P N , KUMAR V , SRIVASTAVA J . Selecting the right interes-tingness measure for association patterns[A]. Proceedin of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining[C]. Edmonton, 2002. 32-41.
[14]	XIONG H , TAN P N , KUMAR V . Hyperclique pattern discovery[J]. Data Mining and Knowledge Discovery Journal, 2006,13(2):219-242.
[15]	WU J J , ZHU S W , LIU H F , et al . Cosine interesting pattern discov-ery[J]. Information Sciences, 2012,184(1):176-195.
[16]	LI F F , PIETRO P . A Bayesian hierarchical model for learning natural scene categories[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. San Diego,USA, 2005. 524-531.
[17]	CLUTO a clustering toolkit[EB/OL]. 2012.
[18]	ZHONG S , GOHOSH J . Generative model-based document cluster-ing: a comparative study[J]. Knowledge and Information Systems, 2005,8(3):374-384.
[19]	VINH N X , EPPS J , BAILEY J . Information theoretic measures for clusterings comparison: is a correction for chance necessary?[A]. Pro-ceedings of the International Conference on Machine Learning[C]. Montreal, 2009. 1073-1080.
[20]	CAO J , WU J J , et al . SAIL:summation-based incremental learning for information-theoretic text clustering[J]. IEEE Trans on Cybernetics, 2013,43(2):570-584.
[21]	WU J J , XIONG H , CHEN J . COG: local decomposition for rare class analysis[J]. Data Mining and Knowledge Discovery, 2010,20(2):191-220.

数据集	实例数量	特征数量	簇数量	密度%
LFW	13 233	3 456	未知	11.886
Oxford_5K	5 062	658 346	11	0.022 8

		参数设置					NMI
数据集	场景			模式数量	图像数量
数据集	场景			模式数量	图像数量
		min_supp	min_cos			Info-Kmeans		CLUTO
	场景1	—	—	—	5 060	0.185		0.157
Oxford_5K	场景2	0.08%	0.45	3 553	1 352	0.594		0.534
( =11)K	场景3	0.09%	0.47	3 383	1 009	0.625		0.483
	场景4	0.10%	0.48	839	559	0.397		0.331
	场景1	—	—	—	13 233	0.176		0.184
LFW	场景2	20%	0.50	907	11 389	0.801		0.678
( =800)K	场景3	35%	0.55	321	6 235	0.846		0.692
	场景4	40%	0.60	21	3 653	0.885		0.703

基于成对约束Info-Kmeans聚类的图像索引方法

Image indexing method based on clustering via Info-Kmeans under pair constraints

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 21

相关文章 9

Metrics

推荐阅读 0

[1]	李骜, 冯聪, 牛宇童, 徐士彪, 张英涛, 孙广路. 面向视角非对齐数据的多视角聚类方法[J]. 通信学报, 2022, 43(7): 143-152.
[2]	迟荣华,程媛,朱素霞,黄少滨,陈德运. 基于快速高斯变换的不确定数据聚类算法[J]. 通信学报, 2017, 38(3): 101-111.
[3]	赵春晖,李雪源,崔颖. 混合编码方式的图像聚类算法[J]. 通信学报, 2017, 38(2): 1-9.
[4]	丁伟,徐杰,卓文辉. 基于层次聚类的网络流识别算法研究[J]. 通信学报, 2014, 35(Z1): 41-45.
[5]	丁伟，徐杰，卓文辉. 基于层次聚类的网络流识别算法研究[J]. 通信学报, 2014, 35(Z1): 9-45.
[6]	刘文杰1，伍之昂2，曹杰2，潘金贵1. 基于成对约束Info-Kmeans聚类的图像索引方法[J]. 通信学报, 2013, 34(7): 18-166.
[7]	闫健恩1，袁春阳2，许海燕1，张兆心1. 基于多维流量特征的IRC僵尸网络频道检测[J]. 通信学报, 2013, 34(10): 6-55.
[8]	闫健恩,袁春阳,许海燕,张兆心. 基于多维流量特征的IRC僵尸网络频道检测[J]. 通信学报, 2013, 34(10): 49-55.
[9]	饶翔,王怀民,蔡华,周琦,孙廷韬,史殿习,尹刚. 云计算系统中基于噪声模板跳表的日志过滤方法[J]. 通信学报, 2011, 32(7): 103-113.