Research On Super-resolution Image Reconstruction Algorithm Based On Sparse Representation

Super-resolution image reconstruction technology is a significant digital image processing technology that reconstructs a high-resolution image by utilizing one or more low-resolution images with complementary information.Working within the super-resolution image reconstruction algorithm based on sparse representation,this paper mainly studies the motion registration,multi-frame image super-resolution reconstruction,single-frame image super-resolution reconstruction and super-resolution image reconstruction software,etc.The main research contents are as follows:(1)Due to the motion errors between image sequences,the accuracy of motion registration affects the precision of multi-frame image super-resolution reconstruction.In this paper,we study the motion registration algorithm based on the improved pixel recursion which extracts the pixels of the edge by adopting the modulus maxima method to calculate the motion parameters,in order to reduce the interference of the background region and the smooth adjacent area.To improve the registration accuracy of the moving target with large displacement,a coarse-to-fine motion registration algorithm with block matching and pixel recursion is proposed.The block matching method is used for coarse registration,and then the improved pixel recursive method is used to fine Registration.Simulation experiments and real experiments show that the motion registration algorithm with block matching and pixel recursion has high registration precision.(2)Aiming at overcome the shortcomings of multi-frame image super-resolution reconstruction model based on sparse representation,we study the local structure similarity sparse model.By introducing the sparse coding estimation error term and utilizing the numerous redundancy of image sequences to obtain the good estimation of the sparse coding coefficient according to the local structure similarity,the validity of the sparse representation is enhanced.In order to further improve the reconstruction effect,a local structure adaptive sparse representation image reconstruction algorithm is proposed,which can adaptively set the regularization parameter based on the maximum a posteriori estimation.The sparse coding of the image can be updated adaptively according to the local structure in the iterative process,so that the reconstruction model is universally applicable.Experiments show that the local structure adaptive sparse representation algorithm of multi-frame image reconstruction can preserve the edge detail and smooth the non-edge region well.(3)To solve the problem of insufficient single-frame image information,this paper studies a single-frame image super-reconstruction algorithm based on non-local autoregressive sparse regularization.The algorithm adopts the anisotropic diffusion method to improve the quality of the low-resolution image.Then it extracts the characteristics of the high and low resolution image blocks to form a joint sample pair and trains the over-complete dictionary to represent the image sufficiently sparsely.Experiments show that the single-frame image reconstruction algorithm based on nonlocal autoregressive sparse regularization can reconstruct good high resolution images.(4)Visual Studio 2010 and MFC are used to write the super-resolution image reconstruction software and super-resolution image reconstruction algorithms are realized by C++.The software can achieve functions of motion registration and super-resolution reconstruction of multi-frame image and single-frame image super-resolution reconstruction,and the software is tested.