多特征匹配的高光谱图像配准方法  

Method of hyperspectral image registration with multi-feature matching

作　　者：吴丹 张艳 WU Dan;ZHANG Yan(School of Electronic&Communication Engineering,Guiyang University,Guiyang 550005,China)

机构地区：[1]贵阳学院电子与通信工程学院,贵州贵阳550005

出　　处：《现代电子技术》2021年第21期56-59,共4页Modern Electronics Technique

基　　金：贵阳市科技局贵阳学院专项资金(GYU-KY-〔2021〕)。

摘　　要：针对传统高光谱图像配准方法适配性较差的问题,提出一种多特征匹配的高光谱图像配准方法,对高光谱图像进行预处理,具体流程包括漏行补充、暗电流消除、校正辐射、检测与消除坏点。以图像配准预处理结果为基础,通过多特征匹配定义高光谱图像的尺度空间,获取SIFT特征点的位置尺度,根据高光谱图像的梯度分布特性获取SIFT特征点的方向尺度,确定SIFT特征点,通过亮度自适应配准算法计算高光谱图像的亮度配准系数,实现高光谱图像的配准。为了证明该方法的适配性较强,将该方法与传统高光谱图像配准方法进行对比,实验结果证明该方法左视场与右视场辐射亮度差异较小,即该方法的适配性优于传统高光谱图像配准方法,实现了性能方面的提升。In view of the poor adaptability of the traditional hyperspectral image registration methods,a method of hyperspectral image registration with multi-feature matching is proposed to preprocess the hyperspectral image.The specific processes include omitted line supplement,dark current elimination,radiation correction,detection and bad point elimination.On the basis of the results of image registration pretreatment,the scale space of the hyperspectral image is defined by multi-feature matching to obtain the position scale of SIFT(scale-invariant feature transform)feature points,and the direction scale of SIFT feature points is obtained according to the characteristics of gradient distribution of the hyperspectral image.The SIFT feature points are determined on the basis of the above position scale and direction scale.The brightness registration coefficients of the hyperspectral image are calculated by the brightness adaptive registration algorithm,so as to achieve the image registration.This method is compared with the traditional hyperspectral image registration methods to prove its adaptability.The experimental results show that the radiation luminance difference between the left field of view and the right field of view of this method is small,that is,its adaptability is better than that of the traditional hyperspectral image registration methods,so it has improved the performance of hyperspectral image registration.

关 键 词：图像配准 高光谱图像 多特征匹配 图像预处理 SIFT特征点 辐射亮度 视场校正 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP311[电子电信—信息与通信工程]