基于CAO的海域监控场景红外与可见光图像配准-《集美大学学报（自然科学版）》

文章信息/Info

Title:: CAO-Based Infrared and Visible Image Registration for Marine Surveillance Scenarios

作者:: 杨青霞1; 潘登1; 黄万林2; 陈尔康1; 黄斌2; (1.集美大学海洋信息工程学院，福建厦门361021；2.集美大学计算机工程学院，福建厦门361021)

Author(s):: YANG Qingxia1; PAN Deng1; HUANG Wanlin2; CHEN Erkang1; HUANG Bin2; (1.School of Ocean Information Engineering,Jimei University,Xiamen 361021,China;2.College of Computer Engineering,Jimei University,Xiamen 361021,China)

Keywords:: marine surveillance; contour angle orientation(CAO); infrared and visible images; image registration; histogram equalization

摘要:: 针对海域监控场景下图像对比度低和视角变化多样的挑战，对轮廓角方向（contour angle orientation，CAO）配准算法进行了改进,提出了海域监控CAO（CAO-marine surveillance scenarios，CAO-MSS）配准算法。该改进在特征点提取阶段中引入CLAHE(基于对比度受限的自适应直方图均衡化)的海域场景特征增强模块；同时在粗匹配阶段提出了基于累计分布函数（cumulative distribution function，CDF）的自适应倾斜角度误差阈值选择方法。为评估CAO-MSS算法，使用真实的海域监控数据构建了一个红外与可见光图像对数据集。实验结果表明：CAO-MSS算法能够得到更多、更准确的特征点匹配，其马赛克拼接图显示红外与可见光图像块衔接更连贯自然。定量分析显示，相较于原CAO算法，CAO-MSS算法的均方根误差(root mean aquare error,RMSE)平均值降低了46.26%，精度平均值提升了15.4%。

Abstract:: Infrared and visible light image registration in maritime surveillance is a highly challenging task.To address the challenges of low image contrast and diverse viewing angles in this scenario,this study has improved the contour angle orientation (CAO) registration algorithm and proposed the CAO-marine surveillance scenarios (CAO-MSS) registration algorithm for maritime surveillance.The improvements include two key innovations:introducing a feature enhancement module for maritime scenarios based on contrast limited adaptive histogram equalization (CLAHE) in the feature point extraction stage,and designing an adaptive tilt angle error threshold screening strategy based on the cumulative distribution function (CDF) in the coarse matching stage.To evaluate the CAO-MSS algorithm,this paper has constructed a dataset of infrared and visible light image pairs using real maritime surveillance data.Experiments on this dataset show that the CAO-MSS algorithm can obtain more and more accurate feature point matches,and the mosaic images demonstrate that the connection between infrared and visible light image blocks is more coherent and natural.In the quantitative analysis,compared with the original CAO algorithm,the average root mean square error (RMSE) of the CAO-MSS algorithm has decreased by 46.26%,and the average precision has increased by 15.4%.