基于似然函数最速下降的红外与可见光图像配准

IR/Visible Image Registration Based on the Steepest Descent of the 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最速下降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.基于点集的高斯混合模型建立了边缘点集配准似然函数,以该函数作为目标函数,仿射变换参量作为优化变量,利用最速下降方法进行最优变换参量求解,从而实现边缘点集配准.同时,将多分辨率金字塔引入迭代配准框架下,实现了高分辨率图像配准的加速.实验结果表明:该算法精度高,运算速度快,可以很好地完成可见光与红外图像的自动配准.

In order to realize automatic image registration for infrared image and visible image,an image registration algorithm based on the steepest descent of the likelihood function was proposed.Image edge was selected as the registration point,and thus the image registration was transferred to edge point set registration.The likelihood function of edge sets registration was established on the basis of Gauss Mixture Model（GMM） of point sets.In order to resolve the optimum transformation parameter by using the steepest descent method,the likelihood function was regarded as objective function and the affine transformation parameter was regarded as the optimization variance.Meanwhile,the multi-resolution pyramid was induced into iteration registration and the speed of registration algorithm for high resolution image was increased.The experiment results show that the algorithm can well complete automatic registration of infrared image and visual image at high registration accuracy and fast registration speed.