摘要
为了解决旋转图像间的点特征匹配问题,提出了一种以空间位置分布关系为搜索依据的新的匹配算法.首先将已检测到的特征点生成三角形,随后利用相似三角形配对获取图像间的仿射变换关系,并根据这种空间位置分布关系进行进一步的对应点搜索,最后利用MLESAC算法进行对应点筛选,得到内点以及外点.用活动摄像机采集图像进行实验,首先验证了对于旋转图像点特征搜索的有效性,接下来又验证了图像间是平移关系但存在重影的点特征搜索,结果表明该方法能有效地对旋转和平移图像进行点特征匹配.
This paper presents a novel method by using spatial distribution to obtain matched points to solve the registration problem between rotated images. At first, triangles are formed from detected feature points, and then the similarity triangle matching algorithm is utilized to acquire affine transformation between images. According to this affine transformation matched points are retrieved. MLESAC algorithm is used to obtain inliers and outliers at last. Images shot by active camera are used to demonstrate the effectiveness of this method. Experiments first evaluate the matching results between rotated images, and then assess the matching results between translated images which are seriously blurred. Results show that this novel method can handle the matching problems between rotated images effectively.
出处
《东南大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2008年第A02期150-153,共4页
Journal of Southeast University:Natural Science Edition
关键词
图像拼接
特征点
三角形
MLESAC
内点
image mosaicing
feature point
triangle
maximum likelihood estimation sample consensus
inliers