【作者】 王立鹏；

【导师】 胡俊；

【作者基本信息】 北京交通大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 图像拼接技术是数字图像处理领域的一个重要的研究分支,它将一组相互间存在重叠部分的图像序列进行匹配对准,重新注册融合后,形成一幅包含各图像序列信息的宽视角场景的新图像。图像拼接可以用来建立宽视角的高分辨率图像,在虚拟现实领域、医学图像处理领域、遥感技术领域和军事领域中均有广泛的应用。本文首先介绍了图像拼接技术的意义、研究背景、研究现状及拼接技术的特点等,从而展示了图像拼接技术广泛的应用前景。然后对图像拼接的基本流程和关键技术做了概括和总结,重点介绍了图像拼接技术中的图像配准和图像融合等技术,分析和总结了目前主流的配准和融合方法。论文重点研究了模板匹配法和相位相关法,并针对传统方法的不足提出了改进的方法。针对传统模板匹配法中亮度对拼接造成干扰的问题,本文提出了基于Lab颜色空间下的模板匹配。Lab颜色空间下可以很容易的将亮度从像素中分离出来,只对a、b颜色分量进行差值计算,实验证明可以有效避免亮度带来的干扰。另外在对应比较区域中采用了设定阈值随机取点的方法,避免了将所有对应点进行颜色差值计算,提高了拼接速度。针对传统相位相关法中亮度对拼接造成干扰的问题,本文采用了基于轮廓的相位相关法。首先提取待拼接图像的二值轮廓图,然后对轮廓图进行相位相关计算,实验证明可以有效的避免亮度带来的干扰。另外在δ函数阵列最大峰值的选取中采用了迭代法二次计算寻找峰值,实验证明,这样可以有效的降低多个峰值带来的干扰。更多还原

【Abstract】 Image mosaics technology is an important research branch of digital image processing. It carries on the spatial matching to a series of image overlapped with each other, and finally builds a seamless image with higher resolution and bigger eyeshot compared with a single image. Moreover, image mosaics technology can be used to construct the large field of view and high-resolution image, and widely applied to virtual reality, medical image processing, remote sensing and military affairs.The thesis first introduces the background of image mosaics technology with great potential, and then demonstrates its characteristics and present research directions. In the second chapter, the author discusses the basic steps of image mosaic. Furthermore, the two pivotal methods of image mosaic, which are the image registration methods and images fusion methods, are introduced respectively. In the third and forth chapter, the thesis analyzes several traditional algorithms, such as the template matching and phase correlation algorithm. Moreover, in order to improve the above-mentioned algorithms, the modified methods are proposed backwards.Specifically, in the traditional template matching algorithm, light intensity always interferes with image mosaic. A modified template matching algorithm in the Lab color space is proposed to solve the problem. In the Lab color space, it is so convenient to divide the light intensity from pixels that we just need calculate the difference between color values a and b. The results show that the proposed method can significantly attenuate the interferences due to the light intensity. On the other hand, random points within the threshold are selected and calculated in a quite small region, rather than the whole region, which largely reduces the complexity and increases the calculation speed.To further reduce the interference from the light intensity, a contour phase correlation algorithm has been adopted. After picking up the binary image, a contour phase correlation is calculated to achieve the desired image. For another, when finding the maximum peak of the Dirac function array, we propose an iterative method with higher accuracy. The results of the experiments have significance for application in image mosaics technology.更多还原