本文为印度Rourkela国立技术研究所（作者：SOMARAJU.BODA）的硕士论文，共71页。

图像配准是用于匹配两个或多个部分重叠图像的基本任务，例如，在不同的时间、从不同的传感器或从不同的视角拍摄的图像，将这些图像整合成包含整个场景的全景图像。它是一种基本的图像处理技术，在集成不同传感器信息、发现不同时间拍摄图像的变化、从立体图像中推断三维信息、识别基于模型的物体等方面具有重要的应用价值。提出了一些技术来寻找一种几何变换，将一个图像的点与另一个图像的对应点联系起来。要配准两个图像，必须找到一对图像之间的坐标变换。

本文提出了一种基于特征的方法来有效估计图像对之间的八参数投影变换模型。该方法利用小波变换提取多个特征点作为配准基础。每个选择的特征点是边缘点，其边缘响应是邻域内的最大值。在实际的匹配过程中，我们会事先检查每一个候选对，看看它是否可能成为一个正确的匹配对。由于这种检查，许多涉及互相关的不必要计算可以预先筛选。因此，获得正确匹配对的搜索时间显著减少。最后，根据正确匹配的特征点对集，确定两幅部分重叠图像之间的变换。

Image registration is the fundamental taskused to match two or more partially overlapping images taken, for example, atdifferent times, from different sensors, or from different viewpoints andstitch these images into one panoramic image comprising the whole scene. It isa fundamental image processing technique and is very useful in integratinginformation from different sensors, finding changes in images taken atdifferent times, inferring three-dimensional information from stereo images,and recognizing model-based objects. Some techniques are proposed to find ageometrical transformation that relates the points of an image to theircorresponding points of another image. To register two images, the coordinatetransformation between a pair of images must be found. In this thesis, afeature-based method is developed to efficiently estimate an eight-parametricprojective transformation model between pairs of images. The proposed approachapplies wavelet transform to extract a number of feature points as the basisfor registration. Each selected feature point is an edge point whose edgeresponse is the maximum within a neighborhood. During the real matchingprocess, we check each candidate pair in advance to see if it can possiblybecome a correct matching pair. Due to this checking, many unnecessarycalculations involving cross-correlations can be screened in advance.Therefore, the search time for obtaining correct matching pairs is reducedsignificantly. Finally, based on the set of correctly matched feature pointpairs, the transformation between two partially overlapping images can bedecided.

1 引言

2 图像配准概述

3 配准算法

4 实验结果

5 结论与未来工作展望

更多精彩文章请关注公众号：