【作者】 李会平；

【导师】 尉朝闻；

【作者基本信息】 西安科技大学 ， 计算机应用技术， 2010， 硕士

【摘要】 图像拼接的目的是将有衔接重叠的图像拼成一张高分辨率全景图像,它是计算机视觉、图像处理和计算机图形学等多学科的综合应用技术。图像拼接技术在虚拟现实、医学、遥感技术等诸多领域有着广泛的应用。图像拼接一般包括图像的获取、图像的预处理、图像配准和图像融合等步骤。图像拼接成功与否主要取决于图像的配准。在对图像配准技术进行研究的基础上发现,基于灰度信息的图像配准方法对灰度变化非常敏感,对图像灰度信息依赖较大;基于变换域的图像配准方法虽然可以为图像拼接提供一个良好的初始配准参数,但该方法计算量太大;而基于特征的图像配准方法不仅不易受光照、旋转等因素的影响,而且特征相对像素数量较少,有利于提高速度,故而本文采用这种方法来实现图像的配准。基于特征的图像配准技术中,重点研究了特征点的检测与匹配技术,分析和总结了各类方法的性能及存在的问题,并对Harris角点检测算法和归一化互相关匹配算法进行了仿真。仿真结果发现特征匹配点对中,存在着一定数量的误匹配点对。针对这个问题,文中给出了改进的匹配算法,该方法的依据是匹配点对连线的斜率之间的关系,目的是消除误匹配点对,提高特征点对的匹配精度。实验证明,该方法是可行并且有效的。为了解决直接将配准后的图像拼接在一起存在的拼缝问题,需要对图像融合算法进行研究。经研究发现平均值法不适合光照度相差比较大的图像,在融合部分有明显的带状感觉;多分辨率样条法虽然能实现平滑过渡,但工作量大,计算时间长;加权平均法虽然消除了图像的边界,但重叠区中仍然会出现重影、模糊现象;故本文采用基于重叠区线性过渡的融合方法,实现了图像的无缝拼接。更多还原

【Abstract】 The purpose of image mosaic is to make up a high-resolution panoramic image through convergence of overlapping images, which is a multi-disciplinary application technology on computer vision, image processing and computer graphics and so on. The technology of image mosaic has been widely used in the Virtual Reality, medical science, remote sensing technology and other fields.The steps of image mosaic include image acquisition, the pre-processing, image registration and image fusion in general. The success of image mosaic primarily depends on the image registration. We find these results below after studying several methods of image registration. Image registration method based on gray information is very sensitive to changes of gray, and primarily rely on the gray information of the image. The image registration method based on transform domain can provide good initial registration parameters for image mosaic, but the computation of this method is too much. The image registration method based on feature is not only vulnerable to the light, rotation, but also helpful to increasing efficiency because the number of features is quite fewer than image pixels. Therefore, this method is used to realized image registration in this paper.The detecting and matching methods based on feather are especially studied in the image registration technology based on feature, their performance as well as the existing problems are analyzed, and the realization of Harris detecting and Normalized Cross Correlation matching argorithms are proposed in this paper. The author find that there is a certain amount of error matching point pairs in the feature matching points during the process of experiment. The improved method based on the relation of the slope ratio is proposed to solve this problem. Its purpose is to remove the error matching points and improve the accuracy of the feature matching points. Experiments show that the method is feasible and effective.There exists an obvious weld if mosaic the pictures directly after registration, so the fusion algorithm is studied to sovel this problem.The author finds that the average method is not suitable for those images which have a large difference in light after study these methods. Smooth transition can be realized through Multi-resolution spline method, but its computation is too much and the operating time is too long. The weighted average method can eliminate the border of the image, but the ghosting and fuzzy phenomenon is still existed in the overlapping regions. So the linear transition based on overlapping areas method is used in this paper. This method achieves the requirement of seamless mosaic.更多还原