基于改进SIFT的图像配准方法研究

发布时间：2018-10-16 20:31

【摘要】：随着数字图像处理和计算机视觉技术的发展,图像配准技术已成为图像处理领域极为重要的一项技术,被广泛应用于遥感、军事、医学图像分析,机器视觉和模式识别等许多重要领域。近年来,诸多国内外学者围绕图像配准进行了深入研究,提出了多种高效的图像配准方法。其中最具经典的是Lowe提出的尺度不变特征变换(Scale Invariant Feature Transform,SIFT),SIFT特征在图像旋转、角度变换、仿射变换和尺度缩放条件下都保持良好的不变性。特征匹配作为图像配准过程中重要的一部分,一直以来都是学者们关注和研究的重点。本文的主要研究工作是基于改进SIFT算法的图像配准,旨在实现特征点精确搜索和有效匹配,本论文的主要内容如下:(1)详细介绍了图像配准的研究背景和国内外研究现状,给出了图像配准的原理和数学理论基础。介绍了几种常见的几何变换模型,为后续图像配准提供理论基础。(2)详细论述了SIFT算法原理,针对原SIFT算法在图像过程中存在误匹配、漏掉大量的正确匹配对以及正确匹配率低的问题,本文提出了一种基于尺度、方向和距离约束的改进的SIFT匹配算法,通过对特征点添加约束因子,剔除误匹配对。实验结果表明,本文方法能够增加正确匹配对数量并提高正确匹配率。(3)针对SIFT算法在遥感图像配准过程中,匹配对数量较少,正确匹配率低的问题,本文提出了一种基于局部仿射约束改进的SIFT遥感图像配准方法。给出了一种新的梯度算子,使用圆形邻域代替原SIFT算法中的方形邻域对特征点构造特征描述子,通过FSC(Fast Sample Consensus)算法和仿射变换局部区域搜索算法,实现特征点的匹配。实验结果表明,本文方法对遥感图像配准效果较好,正确匹配率和配准精度均有所提高。
[Abstract]:With the development of digital image processing and computer vision technology, image registration technology has become an extremely important technology in the field of image processing, and has been widely used in remote sensing, military, medical image analysis. There are many important fields such as machine vision and pattern recognition. In recent years, many domestic and foreign scholars have carried on the thorough research around the image registration, proposed many kinds of efficient image registration methods. The most classical one is the scale invariant feature transformation proposed by Lowe. The (Scale Invariant Feature Transform,SIFT), SIFT features are invariant under the conditions of image rotation angle transformation affine transformation and scale scaling. As an important part of image registration, feature matching has always been the focus of attention and research. The main research work of this paper is based on the improved SIFT algorithm for image registration, aiming at the accurate search of feature points and effective matching. The main contents of this paper are as follows: (1) the research background of image registration and the current research situation at home and abroad are introduced in detail. The principle and mathematical theory of image registration are given. Several common geometric transformation models are introduced, which provide a theoretical basis for the subsequent image registration. (2) the principle of the SIFT algorithm is discussed in detail, aiming at the mismatch of the original SIFT algorithm in the image process. This paper presents an improved SIFT matching algorithm based on scaling, direction and distance constraints. By adding constraint factors to the feature points, the mismatch pairs are eliminated. Experimental results show that this method can increase the number of correct matching pairs and improve the correct matching rate. (3) in the process of remote sensing image registration, SIFT algorithm has fewer matching pairs and lower correct matching rate. In this paper, an improved SIFT remote sensing image registration method based on local affine constraints is proposed. In this paper, a new gradient operator is proposed, which uses circular neighborhood instead of square neighborhood in the original SIFT algorithm to construct feature descriptors. The matching of feature points is realized by FSC (Fast Sample Consensus) algorithm and affine transform local region search algorithm. The experimental results show that the proposed method is effective for remote sensing image registration, and the correct matching rate and registration accuracy are improved.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】