【作者】 姜斌；

【导师】 杨敬钰； 狄明；

【作者基本信息】 天津大学 ， 电子与通信工程（专业学位）， 2016， 硕士

【摘要】 由于散焦而导致的图像模糊是光学成像的难题之一,并且至今仍然无法彻底解决。从传统的光学成像原理上来考虑,透镜引起的光线变向特性使不同景深的物体在传感器上的成像清晰度存在着很大的差异。作为计算摄像学的一个部分,基于编码光圈技术来进行图像散焦模糊的处理有很多方式,本文将在综合国内外研究成果基础上,对如何优化编码光圈形状进行深入拓展,以期得到更好的去模糊效果。同时根据不同深度处的模糊程度来进行深度估计也是编码光圈技术的一个重要组成部分,即通常所说的DFD技术(Depth from defocus)。在经典的通过散焦图像估计深度信息(DFD)的方法中,通常使用单独的光圈将图像信息进行编码,也曾有研究者使用一对互补的编码光圈以期得到效果更好的深度估计,本文创新性地提出了一种旋转编码光圈进行深度恢复,弥补了单独编码光圈和一对编码光圈在恢复过程中需要借助硬件实现的一些缺陷,为此方向的研究增加了新的可能性。在本课题讨论的实验中,一个经过遗传算法优化生成的编码光圈围绕光学中心轴进行旋转,并在三个不同的角度分别拍摄图像,最终使用相应算法进行解码,得到了准确的深度恢复和全聚焦场景图。在旋转编码光圈的选择过程中,综合了其他研究者的一些理论基础,推导出一个标准来评价不同形状编码光圈在深度恢复中的性能表现,同时也获得更好效果的全聚焦场景图。在旋转角度的选择上,本文通过线搜索对所有可能的角度进行验证,来获取最佳旋转角度,并进行了优化。最后,本文在仿真实验和实际拍摄中对比了使用单一光圈、互补光圈对、旋转编码光圈这三种方法在不同的场景中的深度恢复效果,并证明旋转编码光圈有更好的深度恢复效果。更多还原

【Abstract】 The image blur caused by focal is one of the difficult problems of optical imaging,and it can not be solved completely.Different depth of field object is different: when imaging an object in the focal plane,it can get the clarity of the image;when imaging an object far away from the focal plane of the imaging system,the fuzzy image is obtained.As a part of computational photography,there are many ways to deal with the image processing based on the coded aperture technology.This paper will focus on how to select the shape of the coded aperture to get better results.Depth estimation based on the degree of blur in different depth is an important part of the coded aperture technology,which is usually known by DFD technology.Classical depth from defocus approaches infer depth information from two images captured by either a pair of circular apertures with different sizes or a pair of optimized apertures with different patterns.This paper proposes a rotating coded aperture for depth recovery,complementing the previous depth from defocus methods in both recovery accuracy and friendliness for hardware implementation.In the proposed rotating coded aperture,an optimized coded aperture is rotated around the optical axis,and multiple images are taken at a few locations of the rotated aperture.We derive a criterion to evaluate the depth recovery performance of the rotating coded aperture.With this criterion,the coded pattern of rotating aperture is optimized by a genetic algorithm combined with a multiscale refinement strategy,and the angles of rotations are optimized by a line search around all possible yet reasonable candidates.The proposed rotating coded aperture is evaluated on various scenes with both strong and weak textures,and is demonstrated to have better depth from defocus performance over other multiple coded aperture schemes.更多还原