Research On Image Super-Resolution Reconstruction Based On Sparse Representation

Super-resolution reconstruction,as a software-based technology,does not require changing the imaging system of the image itself,which is low-cost and easy to use,and therefore has attracted widespread attention.Sparse representation,as a mainstream method of super-resolution reconstruction,has achieved good reconstruction results by characterizing the essential structure of the image through the construction of an overcomplete dictionary and the sparse representation coefficients.In order to solve the problem of poor image reconstruction performance and provide new ideas for the application of super-resolution reconstruction,this thesis has studied the image superresolution reconstruction method based on sparse representation,and the main contents are as follows:(1)A sparse regularization super resolution method based on structural similarity is proposed.This method introduces structural similarity,a perceptual image quality evaluation index,into the sparse representation model,and uses an improved orthogonal matching tracking algorithm to calculate the sparse coefficients.At the same time,sparse reconstruction is performed from multiple angles using a rotation strategy.In addition,global reconstruction constraints combined with non local similarity are used to further regularize sparse reconstructed images.Experimental results show that this method can obtain better visual quality and preserve the local structure and detail information of the image.(2)A dual dictionary local regularization anchored neighborhood regression method is proposed.This method starts with an anchored neighborhood regression model and reconstructs the image in two stages.In the first stage,local regularization and non local similarity are introduced to constrain the reconstruction results of the image.In the second stage,the trained residual dictionary is used to reconstruct the missing high-frequency details,further refining the reconstruction results.Experimental results show that this method has certain improvements in visual effects and performance indicators.