Research On Enhancing The Image Quality Of Imperfect Optical Systems And Its Application Via Computational Optics

Using digital image processing to enhance the image quality of non-perfect optical systems is the forefront discipline of contemporary imaging science.At present,combining computational optics with digital image processing is one of the main technical means of image restoration.Image restoration methods use the prior knowledge of degradation process to restore images,which take image deblurring as the basic purpose.Image restoration has advantages of enhancing the image quality of optical systems,reducing requirement of the hardware,and improving the quality of pictures.Thus,it is widely used in military,medical,measurement,public safety,traffic monitoring,and astronomical exploration and other fields,which means image restoration is important in scientific research and industrial production.The basic methods of image restoration can be classified into two categories: the blind deconvolution method and non–blind deconvolution approach depending on whether the point spread function(PSF)of degradation model is known.Blind deconvolution algorithms can get good images by employing the feature of degraded images to estimate PSFs.However,the minimum of the resulting cost function does not correspond to the true sharp solution.Particularly,there is still no evident enhancement in the degraded images when the optical aberrations are substantial.In contrast,non–blind deconvolution algorithms can significantly enhance the quality of images by using calibrated PSFs of imaging systems.These prior PSFs are typically measured at a single depth,thereby leading to inadequate results or even failures when the objects are outside the calibration plane.In this paper,knowledge of image recovery,optical imaging system,sparse dictionary,convolutional neural network,and spatial filter are applied for correcting optical aberrations.The location of test images corresponding to object space can be accurately predicted using the proposed method of this paper,which makes non-blind deconvolution methods suitable for improving the image quality of imperfect optical imaging systems in practical applications.The research work of this thesis mainly includes the following three parts:1.A large number of prior PSFs are obtained using blind deconvolution algorithms and non-blind deconvolution algorithms respectively.Then,they are used for training coupled dictionaries via sparse representation.The location of object space corresponding to test images can be obtained using coupled dictionaries,which provides a reference solution for solving the spatial positioning problem of non-blind deconvolution.Thus,the number of prior PSFs is not limited by prediction problems.The image restoration platform of imperfect optical imaging systems is completed according to the proposed method.2.The object space of an optical imaging system is divided several parts.In order to optimize the depth of field of prior PSFs,prior PSFs of each part are selected using an adaptive filter SUBDF(Sub-Dictionary Filter).Then the optimized prior PSFs of each part are used for training a coupled dictionary,denoted as sub-dictionary.All samples in the object space are selected using an adaptive filter MDF(Main Dictionary Filter).Then the selected prior PSFs are used for training a coupled dictionary,denoted as main dictionary.The main dictionary and sub-dictionaries make up of multiple dictionaries.Using multiple dictionaries,the cost time of predicting PSFs can be effectively reduced,and the number of prior PSFs used is limited only by hardware memory.The image restoration platform of imperfect optical imaging systems is completed according to the proposed method.3.Divide object space into N categories according to the spatial characteristics of the data distribution of prior PSFs.The convolutional neural network method is used to distinguish the category of the test PSF.Then,the sub-dictionary corresponding to this category is used to accurately predict the prior PSF.The prediction time of prior PSF is reduced by this way.The image restoration platform of imperfect optical imaging systems is completed according to the proposed method.