SIA OpenIR  > 光电信息技术研究室
大视角变换下的异源光电图像匹配算法研究
其他题名Research on Algorithms of Multi-modality Photoelectricity Image Registration under Large View Transformation
赵恩波1,2
导师史泽林
分类号TP391.41
关键词异源图像匹配 特征描述子 外点去除 仿射不变
索取号TP391.41/Z43/2018
页数75页
学位专业控制工程
学位名称硕士
2018-05-17
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门光电信息技术研究室
摘要异源图像匹配技术在遥感、安防和军事等领域得到了广泛应用,但是由于异源图像之间的灰度变化较为复杂,现有的异源图像匹配算法多是针对特定场景提出的,算法的通用性不强。本文着重对异源图像匹配算法的场景适用性及大视角变化下的图像匹配进行研究,论文的主要工作成果如下:首先,通过分析异源图像成像原理上的差异,总结出异源图像间的异同点,为进一步开展异源图像匹配算法研究提供了依据;其次,在边缘直方图理论的基础上,提出了基于多尺度支撑域描述子的异源图像匹配算法。该算法首先对典型的点特征提取算法在异源图像上的表现进行对比和分析,选择重复度较高的Harris角点作为特征点,并对经典的Harris算法改进,使其具有了尺度和主方向参数;针对单邻域边缘方向直方图描述子邻域范围难以选择的问题,提出了通过组合不同尺度邻域内的边缘方向直方图构建特征描述子的方法,该方法可以充分利用异源图像中的边缘信息,提高描述子的独特性;针对异源图像初始匹配结果低内点率的问题,在RANSAC算法的基础上,提取了一种新的外点去除算法。大量的可见光-红外图像对的实验结果表明,所提算法比传统异源图像算法适用的场景更多,获得的正确匹配点对更多。最后,分析了几种常用的具有仿射不变特性的算法,后将ASIFT算法的思想与所提出的基于多尺度支撑域描述子的异源图像匹配算法相结合,提出了一种能适应大视角变化的异源图像匹配算法。
其他摘要Multi-modality image matching technology is widely used in areas such as remote sensing, security and military. For the complexity of the gray level changes among multi-modality images, most of the existing multi-modality image matching algorithms are proposed for specific scenes. This thesis focuses on the applicability of the algorithm and the image matching under large view transformation. The main contents of the thesis are divided into three parts. Firstly, by analyzing the differences in imaging process, the similarities and differences between multi-modality images are summarized, which provides a basis for further research on multi-modality image matching. Secondly, based on the theory of edge direction histograms, a multi-modality image matching algorithm based on multi-scale support region descriptors is proposed. The algorithm firstly compares and analyzes the performances of several feature point detection algorithms on the multi-modality images, and selects the Harris algorithm that can make the extracted feature points have the highest repeatability. After that, the classic Harris algorithm is improved to have scale and main direction parameters. Then for the problem that the neighborhood range of the single neighborhood edge histogram descriptor is difficult to select, a method is proposed to construct the feature descriptor by combining the histogram of the edge direction in the neighborhood of different scales. This method can fully utilize the edge information in multi-modality images to improve the uniqueness of the descriptors. For the problem of low inlier rates in the initial matching results, a new outlier removal algorithm based on the RANSAC algorithm is proposed. The experimental results show that the proposed algorithm is more robust than the traditional multi-modality image matching algorithms, and can obtain more correct matches. Finally, after analyzing several commonly used algorithms with affine invariance, a new multi-modality image matching algorithm is proposed and can be invariant to large view transformation. The new algorithm combines the idea of ASIFT algorithm with the proposed multi-modality image matching algorithm based on multi-scale support region descriptors.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/21819
专题光电信息技术研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
赵恩波. 大视角变换下的异源光电图像匹配算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
大视角变换下的异源光电图像匹配算法研究.(4526KB)学位论文 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[赵恩波]的文章
百度学术
百度学术中相似的文章
[赵恩波]的文章
必应学术
必应学术中相似的文章
[赵恩波]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。