一种新的基于空间关系的特征匹配方法

特征匹配是基于特征图像配准方法的难点之一,利用空间关系的策略是解决它的一种可靠途径.首先从理论上统一基于空间关系的特征匹配方法的数学模型,然后详细剖析已有几种经典方法定义的全局一致性度量准则存在的问题,最后提出一种新的解决方法.该方法采用高斯分布描述匹配特征对的空间位置在变换模型约束下存在的不一致性,由此定义全局一致性度量准则,使它在平滑连贯性和全局最优解的突出性等方面比已有准则有较大改善,同时该准则中涉及的惟一参数--高斯分布的方差--对它的全局最优解的突出性影响较小,保证提出准则的稳健性.针对新定义准则的最优化求解问题,本文设计一个两步迭代优化算法.用大量实际图像测试本文提出的方法,并将它与几种经典方法进行比较,实验结果表明本文提出的方法是正确和稳健的.

A New Method Based on Spatial Relations for Feature Matching

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.