Research On Image Super-Resolution Reconstruction

With the development of science and technology,there are more and more ways to obtain information through visual images.However,most of the acquired images are low-resolution due to the constraints of various factors.Therefore,it is necessary to process these low-resolution images to obtain expected high-resolution images.The idea of image super-resolution reconstruction is to estimate the corresponding high-resolution images from the initial low-resolution images,where the reconstructed images are required to be as clear as possible,and keep as less artifacts as possible.Recently,image super-resolution reconstruction is widely used in image processing such as semantic image segmentation,digital recognition,and scene recognition.Machine learning method is currently an excellent method to solve the problem of super-resolution,among which the dictionary learning based image super-resolution reconstruction is an important one.Image super-resolution reconstruction is an ill-posed inverse problem,so there are still many problems to be solved.Therefore,in this paper,two improved methods on the basis of the existing algorithms,anchor neighborhood regression,and blur kernel estimation are proposed to solve the problem of super-resolution reconstruction.The main work of this paper is as follows:1.Since most of the classic image super-resolution algorithms has the deficiency of ignoring the residual image structure,this paper considers the structure of residual images combined with anchored neighborhood regression,trains the dictionary,and proposes a Muiti Dictionary Learning-A+(MDL-A+)algorithm.Firstly,cluster all the training image blocks according to the main direction angle of each image block and train the dictionary,then,the training samples are reconstructed by using the trained dictionaries and the residual image blocks are calculated,in addition,the residual is clustered and the residual dictionary is trained.Finally,the anchored neighborhood regression is used in this paper to reduce the time of image reconstruction and facilitate the cluster processing of image blocks.Experimental results show that the proposed MDL-A+ algorithm is superior to the classical algorithms both from the subjective and objective evaluation point of view.2.Blur degradation process plays an important role in the image degradation model.Combining the blur kernel estimation with dictionary learning,a Dictionary Learning based on Blur Kernel Estimation(DL-BKE)algorithm is proposed in this paper.The idea is to perform blur kernel estimation on the image before the reconstruction of the test image,then,use the obtained blur kernel to replace the blur process in image preprocessing,furthermore,use the dictionary learning method to reconstruct the high resolution image.Experimental results show that the proposed DL-BKE algorithm can consistently reconstruct high quality images with better quality robustly compare with the classical algorithms.