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

With the development and improvement of digital imaging and display devices,the requirement of image quality is gradually improved,how to effectively improve the image resolution of low resolution images become one of the hot issues of digital image processing research in recent years.Super-resolution image reconstruction technology uses the same scene of one or more low resolution images,through software processing methods to obtain high resolution images.Because super-resolution technology can overcome the inherent resolution limit of imaging equipment,it can effectively improve the quality of images at lower cost.Therefore,super-resolution reconstruction technology has important application prospects and research value in digital high-definition television,medical imaging,public safety,satellite remote sensing and other fields.At present,many algorithms are mainly based on the reconstruction of the algorithm and the learning based on the algorithm,while the super-resolution reconstruction algorithm based on the sparse representation is the research focus of the learning based reconstruction method.This dissertation mainly focuses on the research of single image,the algorithm based on sparse representation.The main work can be divided into several aspects:1.For the super-resolution reconstruction algorithm based on sparse representation of single dictionary,the details of the reconstructed image are not detailed and the time complexity is high,the algorithm of super resolution reconstruction of single dictionary based on L1/2 regularization is proposed.The algorithm is mainly improved in feature extraction and image reconstruction.In order to improve the accuracy of image matching,a new feature extraction algorithm named wavelet single branch reconstruction algorithm is proposed.In order to obtain a more sparse solution and increase the ability of detail representation,a new fast algorithm for L1/2 regularization is proposed.The simulation results show that the reconstruction quality of the algorithm is improved compared with the traditional regularization algorithm.2.Aiming at the super-resolution reconstruction method based on single dictionary sparse model,some problems such as blurred edges of the reconstructed image and visual effects in the process of image reconstruction from a single dictionary,A new algorithm based on multi dictionary L1/2 regularization is proposed.The algorithm mainly considers the complexity of the natural image structure,This dissertation uses the MCA method to decompose the natural image,and gets the image of the structure layer and the texture layer.Then,different feature extraction methods are used to improving the matching accuracy.In the dictionary learning and image reconstruction stage,the L1/2 regularization algorithm is adopted to improve the quality of image reconstruction.In the end,this dissertation aims at reconstructing the local image,such as Inhomogeneous,edge blur and so on.The combination of global constraints and non-local similarity constraints is used to reconstruct the high resolution image.Experimental results show that the method proposed in this dissertation can improve the visual effect of reconstructed image and the objective evaluation index of reconstructed image.