基于FDR-Net的红外-可见光图像像素级配准方法研究  

Infrared-Visible Image Pixel Level Registration Based on FDR-NET

作　　者：丁钰 黄丹飞[1,2] 钟艾琦 陈思阳 DING Yu;HUANG Danfei;ZHONG Aiqi;CHEN Siyang(School of Opto-Electronic Engineering,Changchun University of Science and Technology,Changchun 130022;Zhongshan Institute of Changchun University of Science and Technology,Zhongshan 528437)

机构地区：[1]长春理工大学光电工程学院,长春130022 [2]长春理工大学中山研究院,中山528437

出　　处：《长春理工大学学报（自然科学版）》2023年第2期39-45,共7页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金重大项目(61893096010)。

摘　　要：针对异源图像配准任务中难以稳定提取有效特征点,且特征点误匹配率过大等问题,提出一种基于FDR-Net特征点检测网络的端到端红外-可见光图像配准方法。首先采用双色相机构建数据集,在所得图像上生成同分布的点阵作为伪特征点,将成对图像和伪特征点坐标共同输入FDR-Net进行训练,学习有效特征点的特征。在网络成功提取特征点后,对特征点坐标使用仿射变换,通过最小二乘法求取最佳解,获得变换矩阵完成配准。配准结果与SIFT、SURF等主流算法进行对比,在特征点数量、平均配准误差、平均角点误差等评价指标上均提升58%以上,具有广阔的应用前景。In order to solve the problem that it is difficult to extract effective feature points stably and the error matching rate of feature points is too high in heterogeneous image registration task,an end-to-end infrared and visible image registration method based on FDR-NET feature point detection network was proposed.First,a two-color camera was used to construct the data set.The same distribution point set was generated on the image as the pseudo feature points,and the coordinates of paired images and pseudo feature points were input into FDR-NET for training to learn the features of effective feature points.After feature points are successfully extracted from the network,affine transformation is applied to the coordinate of feature points,t he optimal solution is obtained by the least square method,and the transformation matrix is obtained to complete registration.The registration results are compared with mainstream algorithms such as SIFT and SURF,and the number of feature points,average registration error,average corner point error and other evaluation indicators are improved by at least 58%,which has broad application prospects.

关 键 词：FDR-Net 特征点 配准 异源 图像处理 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]