Learning error-correcting graph matching with a multiclass neural network

Many tasks in computer vision and pattern recognition are formulated as graph matching problems. Despite the NP-hard nature of such problems, fast and accurate approximations have led to significant progress in a wide range of applications. However, learning graph matching from observed data, remains a challenging issue. In practice, the node correspondences ground truth is rarely available. This paper presents an effective scheme for optimizing the graph matching problem in a classification context. For this, we propose a representation that is based on a parametrized model graph, and optimize the associated parameters. The objective of the optimization problem is to increase the classification rate. Experimental results on seven public datasets demonstrate the effectiveness (in terms of accuracy and speed) of our approach compared to four reference graph classifiers. (C) 2018 Elsevier B.V. All rights reserved.