【作者】 程鸿；

【导师】 韦穗；

【作者基本信息】 安徽大学 ， 信号与信息处理， 2006， 硕士

【摘要】 体视显示通过特定的软件和硬件构造使人的左、右眼得到不同的图像，这种双眼的视差经过大脑的融合使我们感知到具有景深的图像。它是一种重要的3D显示技术，由于在技术实现上比较简单，因此在广告宣传、教育、娱乐、机械设计、建筑设计、医疗等领域具有广阔的应用前景。 本论文的研究目标是根据华东光电技术研究所合作研究项目“基于图像对的体视显示和3D重构”的实际要求而设定的。论文首先阐述了人眼体视显示的视觉原理，然后从体视显示的要求出发，借助于计算机视觉中的多视图几何原理，研究了基于图像对之间几何约束的视图插补方法，由一般的图像对获得满足双眼特性的图像对，从而达到项目的要求。本论文对视图插补的研究分成三个部分：(1)矫正图像对，使矫正后的图像满足视点平移关系；(2)建立矫正图像对之间的对应关系；(3)获取新图像。最后给出真实实验结果。 本论文的主要研究内容如下： 1．研究了人眼体视显示的视觉原理，详细阐述了人眼的体视感知，包括零视差区、两眼同视区以及双眼视差等。 2．研究了基于对极几何的图像矫正方法，包括计算基本矩阵、寻找对极点、无穷远点的射影变换、匹配射影变换、矫正原图像对等。 3．提出了一种矫正图像对的密集匹配算法，包括检测边缘点、寻找匹配边缘点、分割扫描线、仿射变换建立较为准确的点与点的对应关系等。 4．用Matlab编程实现了视图插补和体视显示，给出了真实实验结果，验证了算法的正确性。 本论文的主要研究成果如下： 1．根据体视水平视差的特点，选择了基于对极几何的图像矫正方法，这种方法的目的是为了产生一对“匹配对极线的投影”。这些投影的对极线平行于x轴并且在视图之间相匹配。该方法不需要标定摄像机，只依赖图像信息，运算速度快、应用面广，达到了项目的实际要求； 2．提出了一种矫正图像对的密集匹配算法，该算法最大的优点是在匹配过程中，只需要沿着一条水平线找对应点，具有匹配时间短、精度高等特点。更多还原

【Abstract】 The basic theory of stereo display can be generalized as: with the help of certain softwares and hardwares, different images are sent to left and right eyes and then stereo images can be seen by cerebral amalgamation. Stereo display is an important technique of 3D techniques. It has been widely used in many fields, such as advertisement, education, entertainment, machine designment, architecture designment, medical treatment and so on.The object of this thesis is enacted on the practical request of "Stereo Display and 3D Reconstruction Based on Image Pairs", which is researched cooperantly with Hua Dong Photoelectric Technique Graduate School. In this dissertation, firstly, the vision principle of stereo display are introduced. Then, by virtue of multiple view geometry theory in computer vision field, the method of image interpolation based on geometry constraints between stereo images is researched, so images satisfying characteristic of eyes can be acquired from general image pairs and request of the project is reached. The research on the algorithm of view interpolation is separated into three parts in this dissertation: rectifying the stereo images and making the rectified images satisfying the relation of pure translation motion; establishing the corresponding relation between the two rectified stereo images; acquiring new images. At last, the real experimental results are presented.The main research works are outlined as follow:1. The vision theory on stereo display of human eyes is studied first, and the way how the human eyes produce the sense of third dimension is expatiated, which includes horopter, Panum’s area, binocular disparity, etc..2. The method of rectified images based on epipolar geometry is researched, which includes calculating fundamental matrix, finding epipoles in stereo images, a projective transformation mapping the epipole to a point at infinity, matching projective transformation, rectifying the former stereo images and so on.3. A dense matching algorithm of rectified stereo images is presented, which including detecting the edge points, finding the matched edge points, segmenting更多还原