基于特征的彩色图像配准技术分析-analysis of color image registration technology based on features.docx

摘要图像配准技术作为数字图像处理领域的重要研究内容，在军事和民用领域均 有着广泛的应用前景。图像配准是依据一些相似性度量，选取合适的图像空间变 换参数，使不同成像条件、不同传感器和不同时间获取的同一场景两幅或多幅图 像变换到同一个坐标系下，达到像素灰度匹配和空间位置对齐，是图像拼接、目 标识别等技术的前提和基础。本文回顾了国内外图像配准技术的研究现状，介绍了图像配准的基本理论和 方法，并在此基础上改进了基于灰度图像的两种经典算法，实现了基于特征的彩 色图像配准。最后通过实验验证了算法的有效性。本文主要工作有：1. 介绍了图像配准技术的研究背景和国内外研究现状，阐述了图像配准技术 的基本理论和主要方法，并重点分析了基于特征的图像配准技术的基本框架和算 法流程。2. 改进了基于灰度图像的 SIFT 算法，该改进算法结合颜色不变量理论，利用 图像 RGB 色彩信息构建 CSIFT 描述符，通过特征匹配求解图像空间变换参数，实 现了彩色图像配准。描述符性能测试和图像配准实验证明：CSIFT 描述符不仅继 承了 SIFT 描述符良好的特征不变性，还具有更好的独特性和稳健性，提高了图像 配准的准确度。3. 在深入基于灰度图像 SURF 算法的基础上，改进了 SURF 算法，用局部颜 色描述向量扩展 SURF 描述符，再进行特征匹配和求解空间变换参数，完成彩色 图像配准。描述符性能测试和图像配准实验证明：ESURF 算法具有良好的性能， 比 SURF 算法在彩色图像尺度变化、旋转、模糊、视角变化特别是光照变化方面 有更高的稳健性。主题词：彩色图像，图像配准，图像拼接，CSIFT，ESURF，描述符ABSTRACTAs one of the important research issues in digital image processing, image registration is widely used in military and civil areas. Image registration is the process of overlaying images (two or more) of the same scene taken at different times, from different viewpoints, and/or by different sensors. It is a significant task in the fields of image mosaic, object recognization and so on.The fundamental theory and arithmetic of the image registration are introduced in this paper on the base of reviewing of image registration approaches in domestic and foreign academe. Furthermore, two classical algorithms based on gray scale images are improved. The methods of colored image registration based on feature are the key points in the article. The tests show that the two algorithms based on color image are efficient. This paper completes the following works.The research background of image registration is introduced. The research statues in domestic and foreign academe are summarized. The concepts, principles and common methods of feature-based registration and the main framework and process of image registration are studied.Improved the SIFT algorithms based on gray scale images and applied it to color image registration. Colored image registration is realized by building scal