基于似然函数EM迭代的红外与可见光图像配准

IR/visible image registration based on EM iteration of log-likelihood function

为了实现红外与可见光图像的自动配准,提出了基于似然函数EM迭代的图像配准算法。该算法以图像边缘作为配准点特征,将异源图像配准转化为边缘点集配准。通过点集的高斯混合建模建立了点集配准似然函数,以该函数作为目标函数,仿射变换参数作为优化变量,利用EM迭代优化方法进行最优变换参数求解。迭代过程中,引入基于概率密度自适应阈值分割的外点剔除机制,解决了外点对目标函数的干扰问题,实现了边缘点集的精确配准。利用实测的可见光和红外图像进行了算法验证,证明了该算法的有效性。

In order to realize the automatic image registration for infrared images and visible images, an image registration method based on the Expectation Maximum（EM） iteration of the log-likelihood function is proposed. This method utilizes the image edge as the registration point, and thus the image registration is transferred into an edge point set registration. The point set is modeled as Gauss Mixture Model （GMM）, and the likelihood function of the point set registration is obtained. To solve the affine transform parameter, the likelihood function is maximized with EM iterations. And during the EM iterations, the probability density of edge point is segmented with an adaptive threshold to elimi- nate the outer points, and the interference of outer point with the likelihood function is overcome and the affine transform parameter is determined accurately. The experiments on image registration for infrared images and visible images are verified, and the results indicate that the proposed method is effective.