Blur Degradation Model Estimation Method Research In Blind Image Restoration

With the wide popularity of the digital electronic products,clear and high quality image plays an important role in many application fields.However,due to the limitation of imaging system itself and the influence of many other factors during the imaging process,the obtained image with blur and noise is a common phenomenon.Image restoration technology is designed to improve the quality of the image,to reduce or eliminate the blurs during the imaging process,thereby to restore the real scene,which has been one of the basic challenging problem and most popular research topic in the image processing fields with important theoretical value and practical significance.In this thesis,we first introduce the research background and development in image restoration techniques,the mathematical modeling of image degradation process,the ill-posed solving processing as an inverse problem to the image degradation model,working principle of the solving methods and image quality evaluation criteria.Then we focus on the image blind restoration method based on a priori blur identification,two common types of blur(motion blur and defocus blur)image restoration during the imaging process are studied,the main research results and innovation points are:Firstly,according to insufficiency of the discrete PSF(Point Spread Function)of the defocus-blurred and motion-blurred image at present,in this paper a kind of discretization method based on the combination of geometric property of degradation models and sub-pixel estimation is put forward,the weights allocation principle related to the distance with the neighboring pixels is defined,so as to realize the accurate PSF discretizaion.Experimental results show that the proposed method is highly effective with good performance.Secondly,for the unknown PSF parameters of uniform linear motion blur,an identification method is proposed based on frequency spectrum correlation with reference image.Specifically,by using the motion direction property of the blurred image's twice Fourier transform,a clear reference image is blurred with a fix blur length and various blur directions of the same model,and then direction estimation is obtained by analyzing the spectrum correlation maximum between the blurred image and the image to be restored.The direction error can be limited in a small range by using an adjustment matrix.Furthermore,the motion blur direction and the motion length are calculated accurately by analyzing the spectrum correlation in Fourier frequency domain.Advantage is reflected in the following two aspects: 1)it can realize the motion blur parameters automatically,and can improve the precision of the parameter estimation through setting of parameter's scope and step.2)the linear composition detection in the spectrum for blurred image is susceptible to noise interference,but the method in this paper is based on the spectrum matching,as long as the linear composition in the spectrums for the blurred images with adjacent parameters are different and can be distinguished,it can achieve the optimal parameter estimation and has stronger noise resistance.The image quality is improved significantly by using the non-blind image restoration algorithm.Then,for the unknown PSF parameters of defocus blur,two parameter estimation methods are proposed.The first identification method is proposed based on the blur spectrum characteristic of image basic edge.Specifically,the blur spectrum image of basic edge is analyzed,and then the edge model of natural images is treated as reference image.Furthermore,the max spectrum correlation is analyzed to obtain the right parameter between the image to be restored and the blurred reference image with defocus parameter in a continuous range,and then a non-blind image restoration algorithm is used to restore the blurred image.The other identification method is proposed based on spectrum correlation analysis with the degradation model after the analysis of the parameter estimation errors of the first method in noisy environment.The max spectrum correlation is analyzed to obtain the right parameter between the image to be restored and the degradation circular function with defocus parameter in a continuous range.Theoretical analysis and experimental results prove that the algorithm is feasible and effective in dealing with parameter estimation for large scale blur and noise environment.At last,a hybrid model combining defocus and motion blur is established to restore image degraded by defocus and motion blur.For the unknown PSF parameters of hybrid blur,two kinds of parameter estimation methods are proposed based on analysis of blur model characteristic component in cepstrum domain and frequency domain respectively.One parameter estimation method is proposed based on cepstrum analysis.Cepstrum matching with reference is employed to extract defocus parameter,and then Radon transformation of the cepstrum was used to extract motion direction after enhancing the direction feature by logarithm transform of the cepstrum.The other parameter estimation method is proposed based on spectrum analysis.First the polar coordinate transform is employed for the spectrum for the blurred image,and the accumulation of multiple directions amplitude is used to identify the defocus parameter,then,spectrum correlation analysis with pre-estimation direction is used to identify the motion blur parameters.The experimental results show the method can provide reliable PSF for image restoration,and improve the quality of the restored image.