一种用于红外与微光图像配准的改进型SURF算法  

An Improved SURF Algorithm for IR and Low Light Image Registration

作　　者：杨欢[1] 石俊生[1] 字崇德 李希才[1] 

机构地区：[1]云南师范大学颜色与图像视觉实验室,云南昆明650500

出　　处：《红外》2017年第4期34-43,48,共11页Infrared

基　　金：国家自然科学基金项目(61178054);云南省教育厅重大专项(ZD2014004);云南省高校科技创新团队支持计划资助项目

摘　　要：针对红外与微光图像配准的特殊性,为了减少配准计算量,提出了一种从主方向确定和特征点描述两方面加以改进的加速鲁棒特征(Speeded Up Robust Feature,SURF)配准算法。首先检测微光图像和红外图像的边缘,然后用改进型SURF算法提取两种图像边缘上的特征点,并采用最近邻距离法对原始特征点进行筛选。在得到较高精度的特征点后进行粗匹配。接着用随机抽样一致性(RANdom SAmple Consensus,RANSAC)算法对一次筛选后的特征点进行精匹配。最后利用精确的特征点建立变换模型,并将重采样后的待配准图像与参考图像实现配准。实验结果表明,该算法不仅可以解决红外与微光图像的配准问题,而且在匹配精度和算法运算时间等方面的表现均优于原始SURF算法。According to the par t icular ity of infrared and low l ight image registration, to reduce the amount of registration, an Speeded Up Robust Feature （SURF） registration algorithm improved both in determination of main direction and in description of feature points is proposed. First ly, the edges of low l ight images and infrared images are detected respectively. Then, the improved SURF algorithm is used to extract the feature points of two kinds of image edges. Secondly, a nearest neighbor method is used to screen out the original feature points. Af ter the feature points with higher accuracy are obtained, rough matching is carried out on them. Then, the RANdom Sample Consensus （RANSAC） algorithm is used to carry out precise matching on the feature points screened out one time. Finally, the precise feature points are used to establish a transform model and the images to be registered after resampling are registered with the reference images. The experiment results show that this improved algorithm not only can solve the registration problem of infrared and low l ight images, but also has better performance than the original SURF algorithm.

关 键 词：图像配准 微光图像 红外图像 SURF算法 边缘提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]