Study Of The Image Restoration Based On The Simple Optical System

Image is the foundation of human vision,which helps human understand scenes directly and concretely,so it's always important to promote images' quality and acquire more details of images.Modern optical system is always a high-complicated system that consists of up to twenty more individual optical elements.The aim of this kind of design is to make up for the geometric and chromatic aberrations of a single lens,for example,geometric distortion,field curvature,wavelength-dependent blur and color fringing.However,the complexity of the optical system means high budget and large volume,and it is also expensive and inconvenient to carry.In result,it's necessary to use simple optical system to restore and acquire highly sharp images with the help of computer technology.Image restoration is a hotspot issues in image processing,which directly determines the quality of the image.There are two main methods of image restoration.The first is non-blind restoration algorithm,in this algorithm,point spread function(PSF)is known,and the blurred image can be dealt with deconvolution.The second is blind restoration algorithm,in this algorithm,PSF is unknown,the degraded image is restored directly.In non-blind restoration algorithm,first,people choose the known relatively accurate PSF as the initial value,then people iterate over the chosen value to find the optimal solution,by this way,we can make the iterative process faster and more accurately.There are three channels of the image generated by the simple optical system,RGB.Considering there is always one channel clearer than two other channels,the paper adopts the non-blind restoration algorithm to restore image.First,the relatively accurate spatial change point diffusion function should be estimated,and then the blurred image is dealt with the deconvolution processing with the characteristics of simple optical system to obtain a clear image for restoration.The researches and work that have been done are listed as following:(1)In the point spread function estimation(PSF)part,the paper proposes a spatially varying point spread function estimation method based on Poisson white noise.In most studies,for the purpose of solving the problem more conveniently,the imaging system is approximately linearly invariant.Unlike these studies,fully considering the nonlinearity ofthe real imaging system,the spatially-varying of point spreading function that causes image degradation and the complexity of solving point spreading function,the paper adopts a balancing method to reduce computational complexity,that is,using block-simulated nonlinear system in which the linear system restoration method is used in the block.To this end,the paper proposes an edge overlap blocking strategy that can both simulate non-linear systems and smooth the edges of images blocks.The Poisson white noise has a uniform spectral density function,and the nonlinear system has multiplicative noise,considering the two natures,the paper chooses Poisson white noise model to estimate the point spread function.The paper constructs the data fitness term with the maximum posterior probability of Poisson white noise,the prior information constraint term with the spectral density function,besides,uses the point spread function based on the traditional regularization term,finally establishes the spatially varying point spread function based on the Poisson white noise.(2)In the image deconvolution part,the paper proposed a method based on HIS edge region.The paper mentioned above that there is always one RGB channel clearer than two other channels,existing method is using prior relative gradient to process image deconvolution,but the dark area of the image is not well performed.In order to make better use of the characteristics of simple optical systems,the paper proposes a method that can be used to construct a mask matric for extracting the edges of an image.In the extracted edge region,image deconvolution is done with the prior information of the characteristics of simple optical system that as a constraint.For this purpose,the captured image of the RGB color model should be converted to the HSI color space,then the edge region can be extracted with the intensity component,after this,an image deconvolution model can be constructed by using the inter-channel color variation consistency and absolute gradient,then a deconvolution method based on the HSI edge region is formed.Finally,the problem can be solved by the alternating direction multiplier method(ADMM).