全景图像拼接关键技术研究

发布时间：2018-02-28 06:11

本文关键词： 畸变校正图像配准最佳缝合线泊松融合全景图像拼接　出处：《电子科技大学》2013年硕士论文　论文类型：学位论文

【摘要】：图像拼接技术是一种解决单幅相机视角不够问题的技术，它给人们的日常生活以及科学研究都带来了很大帮助。随着数字图像处理和微电子等技术的不断发展，全景图像已广泛应用于虚拟现实、计算机视觉、航空航天、军事应用、医学图像分析、遥感图像处理等多个领域。全景图像具有视角广阔，真实性高等优点，目前全景图像的拼接方法已经成为了虚拟现实和图像处理等领域研究的热点。复杂场景的配准，图像畸变的校正，大曝光差异图像的拼接，动态场景图像的拼接都是图像拼接的难点，而且目前仍没有一套完善的针对大部分常用图像的拼接方法。本文主要针对全景图像拼接时畸变的校正、图像的配准、图像的融合、动态场景中融合鬼影的消除等关键技术进行研究，本文的具体研究内容及贡献如下： (1)研究了图像几何畸变的校正，包括常用的图像几何变换模型、全景图像投影模型和摄像头失真的校正。引入一种基于几何模型的图像失真校正方法，快速有效地实现了摄像头失真导致的图像几何畸变的校正。 (2)分析了基于频域、基于灰度和基于特征的几类图像配准方法。比较了Harris角点，尺度不变特征变换和快速鲁棒性特征，确定了比较有效的特征提取方法。深入研究了特征点提纯和模型变换参数的估计，，实现了图像的配准，对手持相机拍摄的图像的配准也能取得不错的效果。 (3)针对动态场景中运动物体导致融合鬼影的问题展开了研究，总结现有几种最佳缝合线的搜索准则，针对它们存在的问题提出了一种改进的最佳缝合线。改进的最佳缝合线减小了曝光差异的影响并充分利用了相邻像素点间的相似性，同时提高了算法的速度。研究了多分辨率融合和泊松融合算法，利用傅里叶变换求解提高了泊松融合的速度。最后利用最佳缝合线与泊松融合实现了自然无缝的全景图像。
[Abstract]:Image stitching technology is a technology to solve the problem of single camera view is not enough, it has brought a lot of help to people's daily life and scientific research. With the development of digital image processing and microelectronics technology, Panoramic images have been widely used in virtual reality, computer vision, aerospace, military applications, medical image analysis, remote sensing image processing and other fields. At present, panoramic image mosaic method has become a hot spot in the field of virtual reality and image processing. The registration of complex scene, the correction of image distortion, the splicing of large exposure difference image, the mosaic of dynamic scene image are the difficulties of image mosaic. At present, there is still no perfect mosaic method for most commonly used images. This paper mainly focuses on the distortion correction, image registration and image fusion in panoramic image stitching. In the dynamic scene fusion ghost elimination and other key technologies are studied, the specific research contents and contributions are as follows:. 1) the correction of image geometric distortion, including image geometric transformation model, panoramic image projection model and camera distortion correction, is studied, and an image distortion correction method based on geometric model is introduced. The correction of geometric distortion caused by camera distortion is realized quickly and effectively. In this paper, several kinds of image registration methods based on frequency domain, gray level and feature are analyzed, and the Harris corner, scale invariant feature transformation and fast robust feature are compared. A more effective feature extraction method is determined and the feature point purification and model transformation parameter estimation are studied in depth. The image registration is realized and the registration of the image taken by the hand-held camera can also achieve good results. 3) aiming at the problem that moving objects lead to fusion of ghosts in dynamic scene, this paper summarizes the search criteria of several kinds of best stitching lines. An improved optimal stitching line is proposed to solve their problems. The improved optimal stitching line reduces the exposure difference and makes full use of the similarity between adjacent pixels. At the same time, the speed of the algorithm is improved. The algorithms of multi-resolution fusion and Poisson fusion are studied, and the speed of Poisson fusion is improved by Fourier transform. Finally, the natural and seamless panoramic image is realized by using the best suture and Poisson fusion.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【引证文献】