a novel duplicate images detection method based on plsa model

ISCAS OpenIR

	a novel duplicate images detection method based on plsa model
	Liao Xiaofeng; Wang Yongji; Ding Liping; Gu Jian
	2012
会议名称	4th International Conference on Machine Vision: Machine Vision, Image Processing, and Pattern Analysis, ICMV 2011
会议录名称	Proceedings of SPIE - The International Society for Optical Engineering
页码	-
会议日期	December 9, 2011 - December 10, 2011
会议地点	Singapore, Singapore
收录类别	EI
ISSN	0277-786X
ISBN	9780819490254
部门归属	(1) Institute of Software Chinese Academy of Science Beijing 100190 China; (2) Graduate University of Chinese Academy of Sciences Beijing 100049 China; (3) Information Engineering School Nanchang University Nanchang Jiangxi 330031 China; (4) Key Lab. of Information Network Security of Ministry of Public Security Third Research Institute of Ministry of Public Security Shanghai 200031 China
摘要	Web image search results usually contain duplicate copies. This paper considers the problem of detecting and clustering duplicate images contained in web image search results. Detecting and clustering the duplicate images together facilitates users' viewing. A novel method is presented in this paper to detect and cluster duplicate images by measuring similarity between their topics. More specifically, images are viewed as documents consisting of visual words formed by vector quantizing the affine invariant visual features. Then a statistical model widely used in text domain, the PLSA(Probabilistic Latent Semantic Analysis) model, is utilized to map images into a probabilistic latent semantic space. Because the main content remains unchanged despite small digital alteration, duplicate images will be close to each other in the derived semantic space. Based on this, a simple clustering process can successfully detect duplicate images and cluster them together. Comparing to those methods based on comparison between hash value of visual words, this method is more robust to the visual feature level alteration posed on the images. Experiments demonstrates the effectiveness of this method. © 2012 Copyright Society of Photo-Optical Instrumentation Engineers (SPIE).; Web image search results usually contain duplicate copies. This paper considers the problem of detecting and clustering duplicate images contained in web image search results. Detecting and clustering the duplicate images together facilitates users' viewing. A novel method is presented in this paper to detect and cluster duplicate images by measuring similarity between their topics. More specifically, images are viewed as documents consisting of visual words formed by vector quantizing the affine invariant visual features. Then a statistical model widely used in text domain, the PLSA(Probabilistic Latent Semantic Analysis) model, is utilized to map images into a probabilistic latent semantic space. Because the main content remains unchanged despite small digital alteration, duplicate images will be close to each other in the derived semantic space. Based on this, a simple clustering process can successfully detect duplicate images and cluster them together. Comparing to those methods based on comparison between hash value of visual words, this method is more robust to the visual feature level alteration posed on the images. Experiments demonstrates the effectiveness of this method. © 2012 Copyright Society of Photo-Optical Instrumentation Engineers (SPIE).
关键词	Affine Transforms Clustering Algorithms Image Retrieval Semantics
主办者	Int. Assoc. Comput. Sci. Inf. Technol. (IACSIT)
语种	英语
内容类型	会议论文
URI标识	http://ir.iscas.ac.cn/handle/311060/15725
专题	中国科学院软件研究所
推荐引用方式 GB/T 7714	Liao Xiaofeng,Wang Yongji,Ding Liping,et al. a novel duplicate images detection method based on plsa model[C],2012:-.