摘要
针对可见光与SAR图像灰度差异大,共有特征提取难的问题,提出了一种基于k-均值聚类分割和形态学处理的轮廓特征配准方法。利用k-均值聚类算法对两类图像进行分割,得到图像分割区域;通过形态学处理,有效减少SAR图像斑点噪声影响,准确提取两类图像的封闭轮廓;采用轮廓不变矩理论,引入矩变量距离均值、方差约束机制和一致性检查的匹配策略,获取最佳匹配对,实现了两类图像的配准。通过实验,三组图像的配准精度分别达到0.3450、0.2163和0.1810,结果表明该法可行且能达到亚像素的配准精度。
In order to solve the problem of large difference of gray level and the difficulty of common feature extraction for optical and SAR image registration, an improved multi-model contour feature image registration method is proposed which is based on k-mean clustering segmentation and mathematical morphology. The k-mean clustering algorithm is used to get two kinds of image segmentation region, and through the mathematical morphology processing, accurate extraction of two classes of closed contour image has been realized, which can reduce the influence of the SAR image speckle noise effectively. The matching strategy of the mean and variance of torque variable distance constraint mechanism along with consistency check is bring in which aims to obtain the best match result. Through the experiment, image registration precisions of three groups reach 0. 3450, 0. 2163 and 0. 1810, respectively, which indicates that this method is feasible and can achieve sub-pixel registration accuracy.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2014年第2期176-182,共7页
Acta Optica Sinica
基金
国家自然科学基金(61171057)
山西省回国留学人员科研资助项目(20120706ZX)
关键词
机器视觉
图像配准
K-均值聚类
形态学
约束机制
一致性检查
machine vision
image registration
k-mean cluster
morphology
restriction mechanism
consistencycheck
