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

With the rapid development of mobile Internet and other industries,as the main carriers of information transmission,images are increasingly used in social activities.The resolution of the image is an important index for evaluating the image quality and capability of containing information.Therefore,the research and exploration of the image super-resolution reconstruction technology have important practical significance.With in-depth study of sparse representation and compressive sensing and other disciplines,image sparse coding super-resolution reconstruction(SCSR)algorithm has developed rapidly.Because its algorithm is not affected by the imaging system model and other factors,SCSR algorithm is widely used.The thesis elaborates the theoretical basis of image super-resolution reconstruction,compares the advantages and disadvantages of multiple reconstruction methods,and summarizes the research status of the algorithm.Based on the SCSR algorithm,detailed sparse representation theory,image feature extraction method,training dictionary acquisition algorithm and sparse coding method are described.However,the classical SCSR algorithm cannot detailed describe the image information,and the effect of the algorithm is greatly affected by the dictionary,and it has poor algorithm stability.Because the feature extraction algorithm of the classic SCSR algorithm cannot obtain rich image detail information,in order to ensure that the feature extraction algorithm preserves the edge detail information of the image to the greatest degree,a nonsubsampled contourlet transform method with multi-directional and multi-scale features is used to replace the original gradient feature extraction method.Considering the KSVD algorithm is a classical and effective dictionary training method,which can effectively reduce the complexity of the algorithm,the thesis applies the KSVD algorithm as the core algorithm in the dictionary training stage to establish a pair of overcomplete dictionaries.According to the problem of the poor stability of the classical SCSR algorithm,the thesis uses the kernel norm to replace the original constraint as the constraint method of the model through analyzing the sparse representation model of the image.This constraint method combines the dictionary with the sparse representation coefficients,it takes into account the sparseness and correlation of the coefficients,at the same time,it cares the characteristics of the dictionary,which makes the sparse coding process of the image adaptable to the dictionary and improves the stability and anti-jamming of the algorithm.In the sparse coding process,the Alternating Direction Method of Multipliers(ADMM)is a typical distributed computing framework.It integrates multiple optimization ideas and is widely used to solve constrained optimization problems.The thesis uses the ADMM algorithm as the sparse coding method.Finally,three kinds of simulation experiments are used to verify the reconstruction effect of the algorithm?the adaptability to the dictionary and the anti-noise performance of the algorithm.The experimental results show that the reconstruction effect of this algorithm is improved compared with the SCSR algorithm,and the reconstruction process is less affected by the dictionary and the algorithm has a certain ability to resist noise.