An evaluation of recent graph matching algorithms

Abstract

Graph matching is a fundamental problem in computer vision and image processing and is widely used in object detection. Recently, many methods formulate it as integer quadratic programming problem to find inexact solutions by relaxing it in continuous domain. In this paper we classify these methods in 3 categories based on the relaxed constraints, hypothesis, solving methods, and convergence properties separately. For evaluation purpose we modify these methods and add some toy modifications to compare the detail configuration of these algorithms under different situations. Finally we try to give some explanation based on experimental results.