【作者】 吴晓琴；

【导师】 邵晓鹏； 李江勇；

【作者基本信息】 西安电子科技大学 ， 工程硕士（专业学位）， 2022， 硕士

【摘要】 计算光学系统能够化繁为简、降低成本并在实现小型化、轻量化的同时提高成像分辨率。但由于计算成像系统输出的光学图像存在图像模糊的问题,目前多数复原方法与传统图像复原一样大多都使用空间不变点扩散函数(PSF,Point Spread Function),而计算光学系统受像差影响,PSF随视场而改变;现有的空间变化PSF图像非盲去卷积主要采取分块复原的方式,拼接后易出现明显的振铃效应,对光学模糊图像的复原效果十分有限。针对上述问题,本文提出基于简单计算光学系统的空间变化PSF图像复原技术,将光学设计与数字图像处理相结合,复原出高质量图像。通过研究简单计算光学系统的空间变化PSF图像复原,提出一种改进的分块复原方式弱化子图像拼接时产生的振铃效应;为克服经典滤波算法无法有效保留图像边缘信息的问题,提出一种总变分(TV,Total Variation)正则化空间变化PSF图像复原算法以复原出更高像质的图像。主要研究内容如下:(1)提出一种主客观同时分析的空间变化PSF图像分块复原方法。引入重叠区域像素和重叠区域权重系数,从主观层面和客观层面分析确定简单计算光学系统输出模糊图像的最优子块数量,使用维纳滤波算法和Richardson-Lucy算法对简单计算光学系统模糊图像进行空间变化PSF图像分块复原。结果证明提出的分块复原方法可以有效减弱子图像拼接造成的振铃效应。(2)提出一种TV正则化空间变化PSF图像复原算法。首先对空间变化PSF进行列向分解,再通过TV正则化模型引入两个辅助变量,将约束最优化问题转化为最小化问题,最后利用交替方向乘子法(ADMM,Alternating Direction Method of Multipliers)实现复原图像的迭代求解。对计算光学系统输出的不同尺寸模糊图进行复原,通过对复原效果的分析和对比证明了算法的有效性。本文对计算光学系统输出的空间变化PSF模糊图像复原进行了研究,提出的改进的空间变化PSF图像分块复原方法,通过引入重叠区域像素和重叠区域权重系数的分块方法有效减弱了图像拼接出现的振铃效应;提出的TV正则化空间变化PSF图像复原算法使得简单透镜的峰值信噪比提升约1.4%,清晰度提高约10%,能够在大幅提高图像清晰度的同时保持良好的图像边缘信息。本文方法为简单计算光学系统高分辨率成像奠定了基础,未来低成本、轻量化、高成像质量的计算光学系统在重点区域安防、社会生活监控等领域将发挥巨大潜力。更多还原

【Abstract】 Computational optics systems are advantageous in simplifying structures,reducing costs,and improving imaging resolution while achieving miniaturization and weight reduction at the same time.However,it suffers from the problem of image blurring.Most restoration methods are based on space-invariant point spread function(PSF,Point Spread Function).However,affected by aberrations,the PSF of computational optical systems often changes with the field of view.In addition,the existing non-blind deconvolution of space-variant PSF images mainly adopts the method of block restoration,which is prone to an obvious ringing effect after splicing,and the restoration effect on optically blurred images is very limited.In this thesis,a space-variant PSF image restoration technology based on a simple computational optical system is proposed,which combines optical design and digital image processing to restore high-quality images.An improved sectioned restoration method is proposed to weaken the ringing effect of sub-image stitching by studying the space-variant PSF image restoration of a simple computational optical system.In addition,to overcome the problem that the classical filtering algorithm cannot effectively preserve the image edge information,a Total Variation(TV,Total Variation)regularized space-variant PSF image restoration algorithm is proposed to restore higher-quality images.The major contents of this thesis include the following parts:(1)This thesis proposes a sectioned restoration method of space-variant PSF images that analyzes both subjectively and objectively,in which overlapping region pixels and overlapping region weight coefficients are introduced.The optimal number of sections for the output blurred image of the simple calculation optical system is determined from the subjective and objective levels.Using the Wiener filter and Richardson-Lucy algorithms,the space-variant PSF image sectioned restoration is performed on the blurred image of the simple computational optical system.The results demonstrate that the proposed sectioned restoration method can effectively attenuate the ringing effect caused by sub-image stitching.(2)This thesis proposes a TV-regularized space-variant PSF image restoration algorithm.Firstly,the column-wise decomposition of the space-variant PSF is performed,then two auxiliary variables are introduced through the TV regularization model to transform the constrained optimization problem into a minimization problem,and finally the alternating direction method of multiplier(ADMM,Alternating Direction Method of Multipliers)is used to achieve the iterative solution of the restored image.The effectiveness of the algorithm is demonstrated by analyzing and comparing the restored results of different sizes of blurred images output from the computational optical system.In this thesis,the restoration of space-variant PSF blurred images output by computational optical systems are investigated.The space-variant PSF image sectioned restoration method effectively attenuates the ringing effect of image stitching by introducing the overlapping region pixels and overlapping region weight coefficients.In addition,the TV-regularized space-variant PSF image restoration algorithm improves the peak signal-to-noise ratio of the simple lens by about 1.4% and the sharpness by about 10%,which can significantly improve the image sharpness while maintaining image edge information well.The method in this thesis lays the foundation for high-resolution imaging of simple computational optical systems.In the future,computational optical system with low cost,lightweight,and high imaging quality will play a great potential in the fields of security and social life monitoring in key areas.更多还原