Research And Applications On Adaptive Restoration Of Blurred Image

Pursuing high quality images is many researchers' goal along with the booming development of digital imaging technology.However,the images may suffer from many adverse factors such as defocus blur,the relative motion between target object and camera,atmospheric diffraction and random noise.With the help of image restoration technology,scholars can get sharp and clear images from these blurred images,and image restoration technology as a hotspot in image technology that is widely used in many fields.However,it is of great significance in practice to obtain clear images adaptively and rapidly,as well as to achieve the real-time capability.The contributions of this paper be concerned with adaptive restoration of blurred image are mainly as follows.(1)An adaptive knife edge method based on the traditional knife edge method is applied to estimating the parameters of defocused kernel according to the model of optical defocusing.The proposed method origenates from the image gradient criterion,which is established combined with Canny edge detection operation and Hough transform.Then,the kernel and Modulation Transfer Function(MTF)can be estimated effectively via automatically extracting the optimal edge,which can decrease the arbitrary interference and blindness of human factors and achieve the real-time processing.Besides,the paper puts forward another method of kernel estimation based on the traditional autocorrelation of derivative images,by utlizing horizontal gradient operator replacing the Laplacian operator to decline the computation quantity and ensure the accuracy of identification simultaneously.Experimental results demonstrate that,for both simulated and virtual blurred images,by combining adaptive knife edge method with the deconvolution of Hyper-Laplacian priors,the ringing effect is reduced and MTF is improved by more than 0.22 at the Nyquist frequency.Meanwhile,only by adopting the improved autocorrelation of derivative images,can the relative error precision of defocus radius be confirmed below 3% under the condition of high signal to noise ratio.(2)Regularization parameters play a vital role for many Total Variation(TV)-based image deconvolution methods or other deconvolution methods.So far most studies in the literature focusing on the TV-based restoration problems only concern a fixed predetermined regularization parameter that is selected by manual operation and experience.However,the disadventages are time-consuming,labor-intensive,and man-blind,etc.Therefore,a compacter method that can adaptively specify the value of the regularization parameter without any inner iteration is proposed in our approach,which introduces two auxiliary variables by adopting variable splitting technique to both the regularization term and data fidelity term,and combining with alternating direction method with linear constraints,consequently,the nondifferentiability of TV problem is solved.Furthermore,the solution satisfied Morozov's discrepancy principle can be obtained rapidly and efficiently.Experiments illuminate that the presented algorithm is superior to that of classical algorithms in peak signal to noise ratio,structural similarity and ringing inhibition effect.Moreover,the elaborated approach can also be applied to the restoration of vitical blurred images.(3)Considering that the proposed adaptive TV algorithm often needs much time on the MATLAB platform,for the Fast Fourier Transform(FFT),and the periodic and the conjugate symmetry of Fourier transform in the frequency domain has been utilized to reduce the amount of calculation and storage of the redundancy in the 2D-FFT.Experimental results show that the new methodology compared with previous studies,the calculation speed can be increased by about 15%.Furthermore,the study accomplishes the TV method using Split Bregman on the VS2010 platform in order to implement on hardware conveniently.Consequently,for the simulation blurred images,the computation time is reduced by a factor of 16 compared with the most existing TV-based algorithms using MATLAB.