Learning Cost Functions for Graph Matching

Rafael de O Werneck (1) , Romain Raveaux (2) , Salvatore Tabbone (3) , Ricardo da S Torres (1)

1 UNICAMP - Universidade Estadual de Campinas = University of Campinas
2 LIFAT - Laboratoire d'Informatique Fondamentale et Appliquée de Tours
3 LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery

During the last decade, several approaches have been proposed to address detection and recognition problems, by using graphs to represent the content of images. Graph comparison is a key task in those approaches and usually is performed by means of graph matching techniques , which aim to find correspondences between elements of graphs. Graph matching algorithms are highly influenced by cost functions between nodes or edges. In this perspective, we propose an original approach to learn the matching cost functions between graphs' nodes. Our method is based on the combination of distance vectors associated with node signatures and an SVM classifier, which is used to learn discrim-inative node dissimilarities. Experimental results on different datasets compared to a learning-free method are promising.

Mots clés: Graph matching; Cost learning; SVM

Domaines: Traitement des images [eess.IV]; Vision par ordinateur et reconnaissance de formes [cs.CV]

Rafael de O Werneck, Romain Raveaux, Salvatore Tabbone, Ricardo da S Torres. Learning Cost Functions for Graph Matching. Joint IAPR International Workshops on Statistical Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition (SSPR), Aug 2018, Beijing, China. ⟨hal-01889964⟩