Image Super-resolution Reconstruction Based On Sparse Representation And Dictionary Learning

The most important source of human information is image,and image resolution is the most important indicator of image quality.In practical engineering,due to the reasons of hardware equipment or external environment,the image resolution is often low,which hinders us to obtain information accurately.Therefore,the method of improving image resolution in technology has become the focus of the majority of scholars.In the image super-resolution reconstruction algorithm,compared with the traditional methods based on interpolation and reconstruction,the methods based on learning can complete high-quality reconstruction and does not depend on any background,which has great advantages and become the current research hotspot.In the algorithm based on learning,the reconstruction model combining sparse representation theory with dictionary learning is one of the best algorithms at present.It uses the data model of compressed perception,trains the image data samples,and then obtains the corresponding sparse dictionary pairs,completes the image super-resolution reconstruction and achieves good reconstruction results.Based on this,this paper mainly studies the image super-resolution reconstruction of sparse representation and dictionary learning.Firstly,the theory of super-resolution reconstruction algorithm based on dictionary learning and sparse representation is investigated and analyzed,including signal sparsity,image degradation and degradation model,image sparse representation model,sparse reconstruction algorithm and reconstruction result evaluation methods.Combined with the existing research results,it is found that there are two key steps that affect the efficiency and effect of image reconstruction in the existing algorithm reconstruction model: the first is the process of image sample training to get sparse dictionary pair,and the other is the sparse representation process of reconstructed image.Therefore,this paper focuse on these two steps.At the same time,in view of the low efficiency of the existing algorithms,the significance and importance of dictionary correlation degree are discussed.It is found that dictionary correlation degree index can measure the performance of a dictionary,and it is directly related to the efficiency and effect of reconstruction.Then,in view of the above analysis,this paper proposes an uncorrelated dictionary training method and an efficient sparse representation method from two parts: Dictionary training and image sparse representation.Among the current algorithms of dictionary learning and sparse representation,the dictionary atom correlation is ignored in the process of dictionary training,and the dictionary atoms selection efficiency is low in the process of sparse representation.The training method of uncorrelated dictionary,the sparse representation algorithm based on the combination of kernel function method and efficient dictionary atom correlation selection method are proposed.Finally,in the framework of the classical reconstruction model,based on the structure similarity(SSIM),peak signal-to-noise ratio(PSNR)and reconstruction time of the reconstructed image,the two methods proposed in this paper and similar algorithms are simulated.The results show that this paper improves the quality of atoms in the process of dictionary training and the efficiency of atoms selection in the process of sparse represen tation by introducing uncorrelated processing and kernel method,so as to reduce the time-consuming in the process of dictionary training and sparse representation,improve the precision of sparse representation,get better reconstruction effect,and overcome the correlation between atoms selected by other methods and the original image Low degree results in slow reconstruction speed and poor reconstruction effect due to low precision of atom selection.In addition,this method also has good reconstruction efficiency and effect in the case of less training samples.The trained dictionary has more extensive expression ability,which is suitable for the image reconstruction application scene with less samples in practice.In this paper,there are 10 pictures,7 tables,and 90 references.