Single Image Super Resolution Reconstruction Based On Sparse Representation

Digital image is regarded as the major carrier of information dissemination,whose resolution level can affect the acquirement to the information of image directly.Due to the constraints of imaging equipment and the environment,it is difficult to obtain high-definition images in some scenes.Image super-resolution reconstruction refers to the technology of reconstructing high-definition images from one or more low-resolution images by using software,and which has become the current research hotspot as the image super-resolution reconstruction is not affected by the factors such as the hardware and environment.This paper studies single image super-resolution reconstruction based on sparse representation.The main contents are as follows:(1)In the process of studying the single image super-resolution reconstruction based on the sparse representation,we discover that plenty of methods based on the sparse representation are through extracting the high-frequency information in low-resolution images to reconstruct the image.While these methods can avoid the adverse effects of smooth areas,the information in low-resolution images is not fully utilized.Therefore,the method of reconstructing the image by combining all of the image block information is proposed,which analyzes the image block structure by using the block structure sparseness to reconstruct the meaningful image block.This method not only makes full use of the information in low-resolution images,but also avoids the adverse effects of image smoothing in the process of reconstructin.Finally,the experiments prove the validity of the proposed method.(2)The sparse domain adaptive model adopts the K-means clustering result as the sub-training set and does not consider the anomaly data of training set and global search characteristics of K-means clustering algorithm,which will cause the existence of the anomaly of clustered data and edge data.Hence,this paper proposes using the support vector data description to optimize the clustered data and updating the cluster center.This method not only improves the accuracy of the sub-dictionary,but also makes the sub-dictionary of the image block match more accurate.Finally,the experiment proves the validity of the method.(3)In the procedure of studying the sparse representation of the image super-resolution reconstruction regularization,we find non-local self-similarity regularization can reconstruct the texture information of the image very well.But it cannot deal with the details of the edge.Meanwhile,the total variational regularization can handle the image details at the image edge,but the problem of the texture area connot deal very well.Therefore,a regularized adaptive method is proposed which analyzes the image block structure by block structure sparseness and then adds different regularization constraints according to different image structures.The advantages of these two regularization can utilize fully and the quality of reconstruction can improve.Finally,the experiment proves our proposed validity of the method.