遥感影像特征点的精确匹配方法

发布时间：2018-04-09 07:47

本文选题：特征点匹配　切入点：尺度比　出处：《西南交通大学》2017年硕士论文

【摘要】：影像匹配技术是影像处理的一项重要技术,被广泛应用于多个领域。在遥感领域中,遥感技术已经应用于灾害、资源环境、农业、林业等多个方面,伴随着遥感技术的快速发展,遥感技术的应用范围在人类生活中将会进一步扩大。多数影像的处理与应用,需要多源数据、多幅影像的共同应用,如影像融合提高影像分别率,通过不同时段的同一地区影像叠加进行变化检测,立体像对空中三角测量等。在多源、多幅影像处理中,影像匹配技术是一项基本的处理技术,影像匹配中最常用且稳定的特征匹配是基于影像特征点的匹配。因此,稳定且性能好的影像特征点匹配可以实现影像的自动匹配,使影像得到更好的应用。影像特征点自动匹配中多采用Lowe D G提出的尺度不变特征变换(Scale-Invariant Feature Transform,SIFT)方法实现对影像的特征点提取并描述,进而实现特征点匹配。由于其性能优越,在影像匹配中得到广泛应用。然而,在影像纹理重复性高、影像非线性灰度差异等情况下,基于特征点描述的匹配方法可能出现特征点的错误匹配,将无法满足实际需求。为了实现影像特征点正确匹配,并且获得更多的正确匹配点对,以满足影像匹配的需求,本论文从特征点匹配阶段着手,优化了特征点匹配性能。本论文的具体工作如下:(1)在影像特征点匹配中,影像特征点匹配阶段采用特征向量的最近邻与次近邻比值约束,将会出现正确匹配点的遗漏以及错误匹配点的误检。为了实现正确匹配点尽量多地被检测出来且不存在特征点的错误匹配,论文充分利用特征点的尺度与定位信息,以尺度比和坐标偏移约束实现影像特征点的匹配。(2)以尺度比和坐标偏移约束进行影像特征点的匹配,并与SIFT方法进行实验比较。不同时间、存在模糊变化以及尺度与旋转变化影像的实验结果表明论文方法的最终特征点正确匹配数得到增加,匹配点的均方根误差有所下降。(3)在影像特征点匹配中,剔除匹配点集中错误的匹配点对,才能保证影像匹配的正确性。为了剔除误匹配的特征点,论文分析了特征点的空间关系,将空间关系融入到特征点匹配中,保证特征点集中匹配点的正确性。(4)论文在初始特征匹配点集的基础上,融入特征点的空间关系对初始点集进行优化,并进行了多组实验验证论文方法的可靠性。对存在尺度与旋转、模糊与尺度变化、异源的影像进行实验,并与GTM方法、RSOC方法进行实验对比,实验结果表明论文方法有较高的稳定性。
[Abstract]:Image matching is an important technology in image processing, which is widely used in many fields.In the field of remote sensing, remote sensing technology has been applied to disasters, resources, environment, agriculture, forestry and other aspects. With the rapid development of remote sensing technology, the application of remote sensing technology in human life will be further expanded.The processing and application of most images require the common application of multi-source data and multi-image, such as image fusion to improve the image separation rate, image superposition of different periods of time to detect changes, stereoscopic aerial triangulation and so on.In multi-source and multi-image processing, image matching technology is a basic processing technology. The most common and stable feature matching in image matching is based on image feature points.Therefore, the feature point matching with stable and good performance can realize automatic image matching and make the image be better applied.Scale-Invariant Feature transform (sift) method proposed by Lowe D G is used to extract and describe feature points in image feature points automatic matching, and then feature point matching is realized.Because of its superior performance, it is widely used in image matching.However, in the case of high texture repeatability and nonlinear gray difference, the matching method based on the description of feature points may have a false matching of feature points, which will not meet the actual needs.In order to achieve the correct matching of image feature points and obtain more correct matching point pairs to meet the needs of image matching, this paper starts with the feature point matching stage, and optimizes the feature point matching performance.The specific work of this paper is as follows: (1) in the image feature point matching, the nearest neighbor and the next nearest neighbor ratio of the feature vector are used in the image feature point matching stage, and the omission of the correct matching point and the false detection of the wrong matching point will occur.In order to realize the error matching of correct matching points which are detected as much as possible and there are no feature points, the paper makes full use of the scale and location information of feature points.The scale ratio and coordinate migration constraints are used to match the image feature points. (2) the scale ratio and coordinate migration constraints are used to match the image feature points, and the experimental results are compared with the SIFT method.At different times, the experimental results show that the correct matching number of the final feature points is increased, and the root mean square error (RMS) error of the matching points is decreased in the image feature point matching.The correctness of image matching can be ensured only by eliminating the matching point pairs of mismatch points in the set of matching points.In order to eliminate the feature points of mismatch, the spatial relationship of feature points is analyzed, and the spatial relationship is integrated into the matching of feature points to ensure the correctness of matching points in feature points set.The initial point set is optimized with the spatial relationship of feature points, and the reliability of the method is verified by many experiments.Experiments are carried out on images with scale and rotation, fuzzy and scale changes, and heterogenous images, and compared with GTM method. The experimental results show that the proposed method has high stability.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】