A New Method Based on Spatial Relations for Feature Matching

Jin-jian LV,Gong-jian WEN,De-ren LI,Ji-yang WANG

DOI: https://doi.org/10.3321/j.issn:1001-1595.2008.03.017

2008-01-01

Abstract:Feature matching is a difficult problem of feature-based image registration methods and it can be solved reliably by algorithms using spatial relations. In this paper, a uniform mathematical model of the algorithms based on spatial relations for feature matching is firstly abstracted in theory; and then the shortcomings of various global consistency measure criterions are anatomized in detail, which defined by a few existing classical methods; finally, a new one is proposed to overcome their shortcomings. The Gauss distribution is used in the proposed method to describe the disparity of space position of the matched feature pairs under the limitation of the transform model and then a novel global consistency measure criterion is defined which has a more improvement than ones of the existing methods in smoothness, consistency and prominence of its optimal solution. The variance of Gauss distribution which is the exclusive parameter in the new criterion affects the prominence of the optimal solution slightly and so the proposed criterion is robust. A two step iterative local optimal method is devised to solve the optimal question of the Criterion. Some images were used to demonstrate the proposed method's performance. Simultaneously, compared with some classical methods, the proposed one is accurate and robust.