摘要
针对存在旋转的两幅图像,提出了基于主方向和灰度相关的角点匹配方法。构造每个角点邻域的梯度直方图,以确定角点的主方向。在匹配过程中,用极线约束确定一幅图像中的角点在另一幅图像中的候选匹配角点;建立待匹配角点和候选匹配角点的左右邻域窗口,并根据两角点的主方向旋转右邻域窗口,再进行灰度相关系数计算,确定最终匹配的角点;最后用松弛法消除虚假匹配。该算法首次将经典的灰度相关法应用在大旋转和平移的宽基线图像匹配中,克服了传统灰度相关法不能处理大旋转的图像匹配问题。实验结果表明:本方法具有稳定性、可靠性和快速性,有一定的实用价值。
A novel corner matching algorithm based on principal direction and cross correlation was proposed to solve the rotation problem between two images. The principal direction was determined with constructed gradient histogram of each corner region. In the matching process, the polar line constraint was utilized to achieve the candidate matching corners. The algorithm estabilished the left-neighbor and right-neighbor window between to-be matching corners and candidate matching corners, and rotated the right-neighbor window according to the corner＇s principal direction. Then the cross correlation coefficient was computed to conf＇n＇rn the matched comers. At last, the relaxation technique was used to remove the false matches. This algorithm applied the classic cross correlation in stereo images matching which contained large rotation and translation, and it overcame the disadvantage that the traditional correlation method didn＇t deal with the large rotation image matching. The experimental results show that this algorithm is stable, reliable, fast, and also has the practical value.
出处
《红外与激光工程》
EI
CSCD
北大核心
2008年第3期561-564,共4页
Infrared and Laser Engineering
基金
武器装备预研基金资助项目(6140517)
关键词
角点匹配
旋转不变
梯度直方图
主方向
相关
Corner matching
Rotation invariant
Gradient histogram
Principal direction
Correlattion
