基于直线特征的异源景象匹配技术研究

	基于直线特征的异源景象匹配技术研究
其他题名	Research on Multi-sensor Image Matching Based on Line Features
	王力
学位类型	硕士
导师	贾平
	2015-10
学位授予单位	中国科学院大学
学位专业	光学工程
关键词	图像处理图像匹配异源图像线特征线段对特征提取仿射不变
摘要	在异源图像融合、导航系统等应用中，异源图像匹配技术是其中最重要的一个环节。由于不同图像传感器的成像机理不同，成像特点各异，图像间的灰度具有很大的差异，而传统的图像匹配算法在构建特征描述算子的过程中非常依赖于图像灰度分布，很难应用到异源图像匹配工作中去，因此，针对异源景象匹配的研究是非常有必要的。论文以具有复杂地物条件的异源景象匹配为背景，以开发实时性和可靠性满足实际应用需求的异源图像匹配算法为内容，采用近年来得到深入研究并广泛应用到计算机视觉各个领域中的图像局部特征方法来进行处理。论文主要工作如下：第一部分首先介绍了图像匹配的定义以及在匹配工作中需要用到的几种变换模型，接着对匹配算法按照所利用的图像信息的不同进行分类，并且对这几类算法中一些有代表性的算法进行了简要说明以及分析。基于特征的匹配算法主要优点在于时间复杂度低，准确性强，是当前图像匹配技术领域的热门研究方向。基于区域的匹配算法的优点是实现简单，但计算量过大时这类方法的致命弱点，基于变换域的图像匹配算法最大的优点是易于硬件实现，但在匹配精度上有待进一步提高。第二部分主要分析和对比了三种常用的直线提取方法。首先介绍了采用HOUGH变换提取直线的方法，该算法时间复杂度高，对直线特征的漏检率比较高；接着对基于边缘检测和相位编码的直线提取算法进行分析说明，该算法在检测效率和精度上对图像边缘检测算法和相位编组算法的依赖性过高，而且提取出的弱特征过多；然后对LSD直线提取算法进行了分析和说明，该算法可以在线性时间内提取出亚像素级别的图像直线特征，并且漏检率比较低，是这几种算法中性能最为优秀的算法。第三部分主要对基于直线特征的异源图像匹配算法进行了分析说明以及实验验证。首先介绍了线段特征的提取及筛选及线段对的构造规则和相关数据的计算方法；接着介绍了线段对之间的相似性度量规则、同名线段对搜索策略、线段对匹配规则、根据构成线段对的线段特征之间的关系进行线段特征精确匹配的方法以及误匹配消除方法；最后对本章提出的方法进行了试验验证，通过大量实验选取了最佳的阈值，通过与其他匹配方法的比较验证了该算法的实时性以及匹配性能。第四部分主要对论文的主要工作和创新点进行了总结说明，并对不足之处进行了展望分析。
其他摘要	In the application of multi-sensor image fusion, medical image analysis and navigation system based on computer vision, multi-sensor image matching is always needed. Because of the different imaging mechanism of different image sensors, the image features are different, and the gray level of the image is very different. The traditional image matching algorithm is dependent on the gray distribution of the image. It is difficult to apply to the multi-sensor image matching. This paper is based on the background of the heterologous scene matching with complex terrain conditions, and developing the real-time and reliable multi-sensor image matching algorithm to meet the requirements of the actual application of the heterologous image matching algorithm is research content. Local feature of image which is studied deeply and used widely in computer vision field in recent years is used to deal with the difficulty of multi-sensor image matching algorithm. The main work of this paper is as follows: In the first part, the definition of image matching and the transformation models used in the work of image matching are introduced, then the image matching algorithm is classified according to the difference of image information, and some representative algorithms of these image matching algorithms are briefly described and analyzed. The main advantage of the feature based image matching algorithm is that the time complexity is low and the accuracy is high, and it is a hot research direction in the field of image matching technology. The advantage of the region based image matching algorithm is simple, but the fatal weakness of this method is the computational time is too large. The biggest advantage of image matching algorithm based on transform domain is easy to implement, but it needs to be further improved in the matching precision. The second part mainly analyzes and compares three kinds of commonly used straight line extraction method. Firstly, the method of using Hough transform to extract line is introduced, the time complexity of this algorithm is high, the line features missing rate is relatively high. Then the linear extraction algorithm based on edge detection and phase encoding is analyzed, the algorithm is based on the detection efficiency and accuracy of the image edge detection algorithm and the phase grouping algorithm is too high, and the weak feature extraction is too much. Then the LSD line extraction algorithm is analyzed and explained, the proposed algorithm can extract the linear feature of the sub pixel level in linear time. And the missing rate is relatively low; this is the most outstanding performance of several algorithms. In the third part, the analysis and experiment verification of the image matching algorithm based on line feature are described. Firstly, the extraction and selection of line segment feature and the calculation method of the construction rules and relevant data are introduced; Then introduced the similarity measure between the line segments, the search strategy of the same name, line segments to the matching rule, the line segment characteristics based on the relationship between the line segment feature matching method and the method of eliminating the false match. Finally, the method of the proposed method is verified by experiments, and the optimal threshold is selected by a large number of experiments. The real-time performance and matching performance of the proposed algorithm is verified by comparison with other methods. The fourth part mainly summarizes the innovation points of the paper, and analyzes the shortcomings of the paper.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/49335
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	王力. 基于直线特征的异源景象匹配技术研究[D]. 中国科学院大学,2015.

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
王力.pdf（2250KB）	学位论文		开放获取	CC BY-NC-SA	请求全文