Object recognition as many-to-many feature matching

被引:94
作者
Demirci, M. Fatih [1 ]
Shokoufandeh, Ali
Keselman, Yakov
Bretzner, Lars
Dickinson, Sven
机构
[1] Drexel Univ, Dept Comp Sci, Philadelphia, PA 19104 USA
[2] Depaul Univ, Sch Comp Sci Telecommun & Informat Syst, Chicago, IL 60604 USA
[3] KTH, Computat Vis & Act Precept Lab, Dept Numer Anal & Comp Sci, Stockholm, Sweden
[4] Univ Toronto, Dept Comp Sci, Toronto, ON M5S 3G4, Canada
基金
加拿大自然科学与工程研究理事会; 美国国家科学基金会;
关键词
graph matching. graph embedding; Earth Mover's Distance (EMD); object recognition;
D O I
10.1007/s11263-006-6993-y
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Object recognition can be formulated as matching image features to model features. When recognition is exemplar-based, feature correspondence is one-to-one. However, segmentation errors, articulation, scale difference, and within-class deformation can yield image and model features which don't match one-to-one but rather many-to-many. Adopting a graph-based representation of a set of features, we present a matching algorithm that establishes many-to-many correspondences between the nodes of two noisy, vertex-labeled weighted graphs. Our approach reduces the problem of many-to-many matching of weighted graphs to that of many-to-many matching of weighted point sets in a normed vector space. This is accomplished by embedding the initial weighted graphs into a normed vector space with low distortion using a novel embedding technique based on a spherical encoding of graph structure. Many-to-many vector correspondences established by the Earth Mover's Distance framework are mapped back into many-to-many correspondences between graph nodes. Empirical evaluation of the algorithm on an extensive set of recognition trials, including a comparison with two competing graph matching approaches, demonstrates both the robustness and efficacy of the overall approach.
引用
收藏
页码:203 / 222
页数:20
相关论文
共 36 条
[1]   On the approximability of numerical taxonomy (fitting distances by tree metrics) [J].
Agarwala, R ;
Bafna, V ;
Farach, M ;
Paterson, M ;
Thorup, M .
SIAM JOURNAL ON COMPUTING, 1999, 28 (03) :1073-1085
[2]  
[Anonymous], IEEE C COMP VIS PATT
[3]  
ATHITSOS V, 2004, P IEEE C COMP VIS PA
[4]   USING GEOMETRY TO SOLVE THE TRANSPORTATION PROBLEM IN THE PLANE [J].
ATKINSON, DS ;
VAIDYA, PM .
ALGORITHMICA, 1995, 13 (05) :442-461
[5]  
Bell ET., 1934, AM MATH MONTHLY, V41, P411, DOI DOI 10.1080/00029890.1934.11987615
[6]   Shape matching and object recognition using shape contexts [J].
Belongie, S ;
Malik, J ;
Puzicha, J .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2002, 24 (04) :509-522
[7]  
Buneman P., 1971, Mathematics in the Archaeological and Historical Sciences, P387
[8]   Correspondence matching with modal clusters [J].
Carcassoni, M ;
Hancock, ER .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2003, 25 (12) :1609-1615
[9]  
Cohen S., 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision, P1076, DOI 10.1109/ICCV.1999.790393
[10]  
Conway JH., 1998, SPHERE PACKING LATTI