【作者】 常丽萍；

【导师】 冀小平；

【作者基本信息】 太原理工大学 ， 信息与通信工程， 2013， 硕士

【摘要】 图像配准是数字图像处理的关键技术之一。在高度信息化的今天,图像配准技术已被应用于图像处理的各个领域中,如遥感,医学,计算机视觉等。通过对不同视点不同传感器在同一场景中的成像进行融合,从而得到更全面的信息。图像配准是实现图像融合的基础,是图像融合第一步要解决的问题。图像配准是指将不同成像环境(光线,视点,时间,传感器等)下的同一场景的两幅或多幅图像进行几何关系的配准。通过对其进行平移、旋转、缩放等变形,使待配准图像与参考图像保持几何一致性。本文研究了基于特征的图像配准方法,目的是解决存在复杂空间变换之间的快速、精准的图像配准问题。主要研究的内容包括以下几个方面：1.通过大量的实验研究,本文对图像一般配准过程进行阐述,并在这部分对各个步骤进行了分析和解释,尤其是几何变换和插值最为详细。2.对多种检测算法进行实验,并分析其主要的性能。主要改善了Harris算子,将多尺度空间和模糊系数引入到该算法中。在针对比较大的图像,对其进行分块处理,使用局部阈值,实验证明该方法明显改善了大图像特征点明显分布不均的问题。3.介绍了几种常用的相似性测度方法以及各自的计算方法和适用范围。4.对原有算法进行改善实现图像匹配。实验表明,与传统的方法相比,新的配准方法的匹配效果有明显的改善。对图像配准中各个步骤做了全面的分析和研究。对多种角点检测方法进行了实验和性能分析,改善了原有配准方法。该匹配方法,通过使用最近邻与次近邻距离的双向匹配来实现初始配准。然后利用随机抽样一致的方法对其进行优化并采用邻域灰度信息来进行一致性检测。最后用MATLAB软件,在计算机上验证了该算法。本文仍存在不足的地方,例如其所使用的图像均为数码相机采集的图像,没有使用医用图像、遥感图像等传感器的图像。对于多源传感器图像应用有一定的局限性。更多还原

【Abstract】 Image registration as the foundation of the realization of image fusion, is one of the key technologies of digital image processing. With the development of digital technology, image registration technology has been applied in various fields, such as remote sensing, medicine, computer vision, etc. For one scene, there will be many differences between images obtained from different viewpoints using different sensor. We use the technology of image registration to get more comprehensive information of the scene.Image registration is the process of transforming different images which obtained from different imaging conditions (light, viewpoints, time, sensors, etc.) into one coordinate system. Through translation, rotation and scaling distortion, keep stay images geometric consistency.This paper studies the image registration method based on point characteristics. The purpose is to realize image registration more fast and accurate between huge different images. The main content includes the following aspects:1. Based on a lot of experimental research, this paper puts forward the general process of image registration, and in this part of the various steps were analyzed and explained, especially the geometry transform and interpolation is the most detailed.2. Test a variety of detection algorithms and analyze their main performances. An improved Harris corner detection algorithm is proposed in this paper. And the multi-scale space and fuzzy coefficient is introduced into the algorithm. For larger images, chunk, use local threshold. The experiment has proved that the method significantly improves the uneven distribution of feature points in large image.3.This paper introduces several kinds of commonly used similarity measures, including respective calculation methods and application scope.4. This paper proposes a new feature point matching algorithm. The experiment shows that the matching effect is improved compared with traditional methods.In this paper, every step of the image registration and many corner detections have been comprehensive discussed. A new registration method is proposed. This paper adopts the ratio between the nearest distance and the second nearest distance from two directions to achieve the initial matching. Then use the method of random sampling consensus to optimize the matching. Last, test the conformance of the points by the neighborhood gray level information. Finally, verify the effectiveness of the proposed algorithm through the experiment based on the MATLAB.In this paper, there is still a shortage of the paper. All images used are from the digital camera, not to use the medical images, remote sensing images. For multi-source image sensor, the applications have some limitations.更多还原