基于轮廓多边形拟合的红外与可见光图像配准算法  

Image registration algorithm for infrared and visible images based on contour polygon fitting

作　　者：李振华[1] 江耿红[1] 徐胜男 刘允刚[1] 

机构地区：[1]山东大学控制科学与工程学院,山东济南250061 [2]济南幼儿师范高等专科学校,山东济南250307

出　　处：《系统工程与电子技术》2015年第12期2872-2878,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(60904097);中国博士后科学基金(2013M541914;2014T70637);山东省优秀中青年科学家科研奖励基金(BS2011DX018);国防技术基础科研计划(GKY201201);山东省科技发展计划(2013YD01036);济南高校自主创新计划项目资助课题

摘　　要：针对红外图像和可见光图像配准问题,提出一种基于轮廓多边形拟合的图像自动配准算法。首先,为获得较好的主轮廓信息,对提取的轮廓进行多边形拟合,有效剔除轮廓上的冗余点和噪声。然后选取拟合轮廓上的多边形顶点为特征点,将轮廓划分为特征轮廓段,以特征轮廓段作为匹配单元匹配轮廓并得到匹配特征点。采用修剪的最小二乘法,在获取变换参数的同时不断剔除误匹配。实验证明,该算法可实现性强,配准精度高,配准速度快,较好地实现了刚体变换下红外图像与可见光图像的配准。This paper presents an automatic image registration algorithm based on the contour polygon fit- ting, which aims to align the infrared and visible images geometrically. Firstly, in order to get the well-defined main contours from the input images--the infrared and visible images, polygon fitting is performed to get rid of the redundant points and noises involved in the contours. Secondly, the polygon vertices along the contours are chosen as feature points, then the contours are diveded into some feature segments. The feature segments along every contour are used as matching primitives to match the contour in order to get the matched vertices. Accord ing to the correspondence between the matched vertices in the reference image and those in the sensed image, the trimmed least square method （LSM） is conducted to calculate the registration parameters and eliminate false matching points at the same time. Finally, the infrared and visible images with rigid transformation can be a- ligned precisely. Experimental results show the effectiveness of the proposed registration algorithm and demon- strate the superiorities of its alignment accuracy and alignment speed.

关 键 词：图像配准 多边形拟合 特征轮廓段 修剪的最小二乘法 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]