【作者】 刘超；

【导师】 黄东军；

【作者基本信息】 中南大学 ， 计算机科学与技术， 2011， 硕士

【摘要】 近年来,随着计算机技术的快速发展,图像处理应用越来越广泛和深入,对具有较大视域全景图像的需求也越来越迫切。全景图像拼接作为新兴技术,短短几年得到了快速发展,受到研究者越来越多的关注。目前全景图像已经成为计算机仿真、计算机视觉模拟、图像处理和计算机特效以及虚拟现实研究中的热点和关键技术,在地质勘测、军事侦查、医学微创手术、航空航天以及视频会议等多个领域发挥着重要作用。图像拼接是指把多个较小视域的图像拼接成一个较大视域的图像。本文在研究以往拼接技术的基础上,针对当前图像拼接算法运行效率低的现状,采用基于特征点的图像匹配方法和基于双线性插值的加权平均融合方法,通过改进角点检测以及图像特征点匹配策略,设计了新的图像拼接算法。实验表明,该算法能够显著提高图像拼接的速度。本文算法首先对采集到的图像校正去噪尽可能的去除图像噪声；然后改进角点检测过程,在检测角点前先检测图像边缘,仅对处于边缘点的像素使用Harris算法进行角点检测；在特征点匹配时,把待拼接图像按照顺序进行分组,每组以中间图像为基准,同时和两边的图像进行特征点匹配,这样每次都够完成三幅图像匹配,并且每个图像仅被加载和读取一次；最后使用基于双线性插值的加权平均融合法进行图像融合,降低了图像融合的计算量,并且消除了拼接后图像间的缝隙。实验表明该方法和传统的基于特征匹配的图像拼接算法相比,拼接速度快,具有较好的实时性能。更多还原

【Abstract】 In recent years, with the rapid development of computer technology, image processing applications increasingly extensive and in-depth,as the same time, on the horizon with a large panoramic image of the demand for more becomes more urgent.Panoramic image stitching technology as a rise of emerging technologies will attract more attentions of the researchers. Just a few years it has been rapid development, and become a key technologie in computer simulation, computer vision, simulation, image processing and computer special effects and virtual reality Panoramic image mosaic image processing technology must be as the future of one of the main research site.Panoramic image mosaic uses multiple images with smaller view to blend into a larger image with much wide view. Based on previous splicing algorithm and technique, this paper proposes a new panoramic image mosaic algorithm.It assembles multiple images with smaller view into a panoramic image which has larger field of vision, high resolution, smooth edge, and seamless.Before image registering, it removes the collected image correction and denoises image noise as much as possible, Then this algorithm detects cornors in images using Harris algorithm base on detecting image edge through Sobel algorithm.When macthing the images, mosaic images are grouped according to their orders,and uses the meddle image as the reference.In this way it can complete three images match each time, and each image is only load and read for one time. Finally, based on bilinear interpolation and using the weighted average fusion, the algorithm of this paper completes image blending.It reduces the amount of computation, and eliminats the gap between the images in stitched image.It shows that this method has more fast speed and better real-time performance comparing with traditional feature-based matching algorithm for image stitching.更多还原