SIA OpenIR  > 光电信息技术研究室
基于特征和互信息的红外与可见光人工场景图像匹配算法研究
Alternative TitleMatching Algorithm of Infrared and Visible Artificial Scene based on Feature and Mutual Information
张亚红1,2
Department光电信息技术研究室
Thesis Advisor夏仁波
ClassificationTP391.41
Keyword红外与可见光 特征提取 线段上下文 图像匹配
Call NumberTP391.41/Z36/2015
Pages59页
Degree Discipline模式识别与智能系统
Degree Name硕士
2015-05-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract异源图像匹配是对来自不同传感器、不同时间或不同视角的两幅或多幅图像,在空间中寻找一种变换,使其在空间位置上达到一致。红外与可见光图像的匹配属于异源图像匹配,在自主导航、末制导、变化检测等领域中均受到了广泛关注。由于成像机制和条件的不同,同一场景的红外与可见光图像存在比较明显的光谱差异和透视差异,使得红外与可见光图像的匹配具有非常大的挑战性。本文针对红外与可见光图像,提出了一种基于特征和互信息的人工场景图像匹配算法。主要内容包括: 首先,研究了图像中点特征和直线段特征提取方法。Harris算法是一种经典的角点检测算法,但在红外图像中检测出的特征点分布均匀性差,针对这一问题,本文提出了基于子区域划分的角点提取方法;针对红外与可见光图像提取的特征点分布一致性不强的缺点,采用LSD(Line Segment Detector)算法检测出图像中的直线段,按照几何约束规则挑选出关键直线段,并计算它们的交点,将交点与改进的Harris角点一起组成图像特征点。针对局部邻域线段提取效率低的问题,在对全局直线段检测结果像素化表示的基础上,通过检索对应局部区域内的非零像素值实现快速直线段提取。 其次,为了提高异源图像点特征的表征性能,在特征提取的基础上,构建基于线段上下文的特征描述子。本文基于线段的长度l、线段的方向θ、线段到中心点的距离d三个属性设计了得分函数,通过计算特征点四象限邻域内线段的得分,得到每条线段对特征点的贡献,在此基础上采用圆形阵列的方式,构建基于线段上下文的特征描述子。 最后,运用双向匹配策略和RANSAC算法实现红外与可见光图像的粗匹配。为了提高匹配精度,提出了基于互信息的精匹配方法。将粗匹配结果作为精匹配的输入参数,确保搜索结果为全局最优解。为了提高计算速度和收敛效率,采用多分辨率框架和改进的Marquardt-Levenberg 搜索算法进行寻优,使目标函数快速地找到最优解。 实验结果表明,所提方法能够对灰度差异较大的红外与可见光图像实现精确匹配,并且在鲁棒性和时间效率方面都要优于主流异源图像匹配算法。
Other AbstractThe purpose of Heterologous images matching is to align two or more images of the same scene at different times, in different perspectives or with different sensors. Matching the infrared and visible images belong to heterologous image matching, which has draw wide attention on autonomous navigation, homing, change detection and other fields. As Imaging mechanism and the condition is different, there exist obvious spectral and perspective differences of the same scene of infrared and visible image, making the infrared and visible image matching being a big challenge. In view of the infrared and visible images, this paper puts forward an artificial scene image matching method based on feature and mutual information. The main contents include following parts: Firstly, we study the point and straight line feature extraction method. Harris algorithm is a classic corner detection algorithm, but the points distribution in the infrared images is not uniform. In order to solve this problem, based on sub-region division this paper puts forward a new corner point extracting method. Owing to feature point distribution is not consistent in the infrared and visible image, LSD (Line Segment Detector) algorithm is used to detect the line segments, then picking out key line segments according to the rules of geometric constraint, and calculating their intersections. The intersections and the Harris corners together constitute the image feature points. Aimed at the problem of low efficiency of the local neighborhood line extraction, on the basis of pixelated representation of global line segment detection results, we search through corresponding non-zero pixels within the local area to realize fast line segment extraction. Secondly, in order to improve the representation performance of feature in different source image, we construct a context-based feature descriptor. Based on the length l, direction θ and the distance d of the line segment to the feature point, we design score function. By calculating the score of each line segment in four-quadrant neighborhood around the feature point, we can obtain the contribution of each segment to the feature point. Then on this basis, the context-based feature descriptors are constructed in circular array pattern. Finally, bidirectional matching strategy and RANSAC algorithm is used to realize coarse matching of infrared and visible images. In order to improve the matching accuracy, precise matching method is proposed based on mutual information. The crude matching results is used as input parameters of refined matching to ensure the search results being global optimal. In order to increase computing speed and convergence efficiency, multi-resolution framework and improved Marquardt-Levenberg optimization algorithm is used to find the optimal solution. Experimental results show that, the proposed algorithm can achieve exact match results in infrared and visible images with great gray difference. In terms of robustness and time efficiency it is better than the classical heterologous image matching algorithms.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/16775
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
张亚红. 基于特征和互信息的红外与可见光人工场景图像匹配算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2015.
Files in This Item:
File Name/Size DocType Version Access License
基于特征和互信息的红外与可见光人工场景图(3383KB) 开放获取CC BYApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[张亚红]'s Articles
Baidu academic
Similar articles in Baidu academic
[张亚红]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张亚红]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.