基于结构风险最小化的图像配准之改进方法

目前的基于多项式、正交多项式和加权正交多项式的图像配准方法是以误差平方和为度量标准的．此度量会导致图像配准结果出现过溢出现象，并且泛化能力差．根据结构风险最小化原则，重新定义误差函数，即在误差函数中引入系数惩罚项，提出了改进的多项式、正交多项式和加权正交多项式三种图像配准方法．实验结果表明，基于新的误差函数的图像配准方法较好地提高了图像配准的准确度。

Novel Structural-Risk-Minimization Based Approach for Image Registration

Current polynomial, orthogonal polynomial and weighted orthogonal polynomial approaches for image registration are based on MSE, which often leads to so-called overflowing phenomenon and the poor generalization capability. In this paper, based on the principle of the structural risk minimization, the new error function is defined by introducing the coefficient penalty term. The improved polynomial, orthogonal polynomial and weighted orthogonal polynomial approaches for image registration are accordingly presented. The experimental results demonstrate that the improved approaches enhance the accuracy of image registration.