Research On Blind Super Resolution Image Reconstruction Algorithms

As a measurement of image information,image resolution describes details contained in an image.It's decreased by degradations in imaging process.The traditional Super Resolution Reconstruction(SRR)techniques reconstruct high resolution(HR)images from several low resolution(LR)images.But they fail to reconstruct HR images in the case of unknown blurring.So we study blind deconvolution(BD)and SRR to solve the problem of Blind Super Resolution Image Reconstruction(BSRIR).According to the traditional image degradation(ID)models,a new general ID model and a new general BSRIR model are proposed.And a novel Image Quality Assessment system based on Peak Signal-to-Noise Ratio(PSNR)and Sum Square Difference(SSD)is proposed to assess the BSRIR algorithms.The experiments show the system's efficiency and rationality.Most Multiframe BSRIR algorithms are computationally expensive,timeconsuming and susceptible to the LR outlier images.And these shortcomings affect the speed and accuracy of the algorithms for the recovery of HR and blur kernel.So the novel BSRIR algorithms are proposed to solve these problems.First,multiframe BSRIR algorithms based on total variation blind deconvolution(TVBD),which combine TVBD with the traditional SRR model,are proposed to solve the problem of blind deblurring by using Alternating Minimization(AM)and total variation regularization for image or blur kernel.Experimental results show that the algorithms are more accurate and computationally efficient for reconstructing HR image and unknown blur kernel.Second,multiframe BSRIR algorithms based on mediam shift and add(MSAA),which remove the outlier effect for the high accuracy by using MSAA and reduce the amount of computation to improve the algorithmic speed,are proposed.Experimental results show that these algorithms reduce the amount of computation effectively with the least outlier effect,higher speed and accuracy.Third,multiframe BSRIR algorithms based on Lower-Bounded Logarithmic Image Priors,which are more accurate and effective for BSRIR by using AM and logarithm of TV based on majorization minimization(LogTVMM),are proposed.In order to improve the efficiency,accuracy and speed,we combine MSAA with LogTVMM to remove outliers by using the information of image gradient.Experimental results show the good performance with higher accuracy and speed.