【作者】 王宇；

【导师】 陈殿仁； 朴燕；

【作者基本信息】 长春理工大学 ， 物理电子学， 2010， 博士

【摘要】 全景成像技术(Integral Imaging,Ⅱ)是一种新型三维图像技术,它利用微透镜阵列来记录、显示全真的三维场景。这种技术在不需要任何观察设备的情况下,在空间上再现了三维图像,而且不需要辅助光源显示立体图像,能给观察者提供连续视点、全视差的真实的立体图像,克服了眼睛集中适应性调节冲突问题。因为全景成像技术具有上述优点,所以它吸引了越来越多的科技工作者和公司的重视,成为三维图像领域内的研究热点,但全景成像系统的重构图像分辨率较低一直是影响该技术发展的主要问题之一。在解决全景成像系统重构分辨率低的问题时,一般有两类方法：基于光学的方法和基于计算机的重构算法的改进。本文在现有的全景成像系统条件下,重点研究了如何根据全景成像特点,利用计算机改进重构算法,从而提高重构图像的分辨率。本文首先总结了目前存在的几类三维立体显示技术以及各自特点,并对全景成像技术的基本理论进行了分析,包括全景成像技术的工作原理、性能指标和全景成像系统的分类。接着,论文探讨了重构图像的分辨率限制问题,分析了Ⅱ分辨率的计算、衍射效应和聚焦误差对分辨率的影响等问题。在介绍了目前采用的改善重构分辨率的光学方法之后,本文对传统的计算机重构三维物体的方法进行了总结和实验。同时,重点研究了全景图像的去噪过程,根据全景图像的特点,提出一种基于元素图像的局部去噪方法,这种方法在取得与传统技术相同的去噪效果的同时,还能保持元素图像之间边界的清晰度,对全景图像的后续处理非常有利。通过改进重构算法来提高重构图像分辨率的方法不需要机械运动和额外设备,更具有灵活性和可行性。本文提出了一种基于相似像素块平滑过渡的图像后处理方法,使用该方法处理后,属于同一个灰度值变化平缓区域内的相邻像素块的灰度值就能平滑过渡,这样的图像后处理过程既缓解了像素块间的灰度不连续性,又能保持重要的边缘信息,改善了重构图像的视觉效果。另外,本文又提出一种改善计算机重构图像视觉质量的方法,该方法利用3D空间的物体部分在每个元素图像中形成的匹配区域的纹理特征,从两个相邻的元素图像中的匹配区域提取出多个像素,经过加权计算重构出相应的图像区域。该方法与传统的计算机重构方法相比,提高了图像分辨率,改善了重构图像的视觉质量。最后本文对图像的超分辨率重构技术进行了研究,提出将超分辨率处理技术引入到重构的视图序列的后处理过程中,并设计了单幅视图和多幅连续视图的超分辨率处理的方法。同时,文中对提出的各种算法进行了仿真实验,得到了较理想的结果。更多还原

【Abstract】 Integral Imaging (Ⅱ) is a technique capable of recording and displaying 3D images in natural color through micro-lens array. This technique is a promising technology because:a) it produces autosteroscopic images, thus not requiring special viewing devices, b) it does not require special illumination of the scene, c) it provides the observer images with full parallax and continuous viewing points, d) it can overcome the observer’s visual fatigue because of convergence-accommodation conflict. Integral imaging has attracted a great deal of attention in 3D imaging fields in recent years. Low resolution is always one of the most important factors which constrained the development ofⅡtechnology.There are two kinds of methods to resolve low resolution of reconstructed image in II. One is optical method and the other is to improve reconstructed algorithms based on computer. Under the existingⅡsystem conditions, this paper mainly studies on improvement of reconstructed algorithms to improve image resolution according to the features of integral imaging.Several kinds of 3D display techniques and their characteristics are concluded in this paper. Basic theories of II technology are analyzed, including work principles, performance indexes and categories of II system. Then, the resolution limitation of integral imaging is discussed. Calculation of the resolutionm, the affection of diffraction limitation and focusing error on the resolution of II are also analyzed.After introducing current optical methods of improving resolution of reconstructed images, traditional reconstruction methods are concluded and experimented in this paper. Based upon the detailed analysis of the noise removal of integral image, a local removing noise method in elemental images is proposed according to characteristics of integral image. This method can achieve the same affection as usual removing noise methods while maintain clear boundary between elemental images. It is good for later processing of integral images.The method of improving reconstruction algorithm to improve resolution of reconstructed image is more flexible and feasible without requiring mechanical movements and additional equipments. This paper proposes an image post-processing method based on smooth transition between the adjacent similar pixels blocks. The proposed method modifies gray values of the pixels in the similar pixels blocks according to the proposed algorithm. Thus it provides improvement of viewing image by smooth processing in similar pixel blocks to relieve discontinuous pixel values between the adjacent pixels blocks. In addition, this paper also proposes a method of improving visual quality of computational integral imaging reconstructed images. This method is based on texture features in the matching regions of the same object region through different lenslets. Several pixels are extracted from two matching regions in adjacent elemental images to reconstruct corresponding image region after weighted calculation. Compared with conventional reconstructed methods, this method removes pixel-blocking effect existing in direct reconstruction of pixel blocks and improves viewing quality of reconstructed image.Finally, image super-resolution reconstruction techniques are researched. This paper proposes that super resolution processing should be introduced into post-processing of reconstructed image sequences and designs the methods of super resolution processing for single image and multiple images. Experimentations of the proposed algorithms in this paper are carried out and better results are achieved.更多还原