A Study On Adaptive Algorithms Of Single Image Super Resolution Reconstruction

Image resolution is one of the basic attributes of the image.It represents the pixel's density of the image and determines the sharpness of the image.Because high-resolution images can provide richer details than lowresolution images,the resolution of images in many areas are required.By now,image super-resolution reconstruction has become a widely used topic.This dissertation focuses on learning-based single-frame image superresolution reconstruction algorithm.Based on the sparse representation theory,the original attributes of the image and the superiority of the learningbased image super-resolution reconstruction is developed by adaptive method for super-resolution with adaptive parameters,and the method,on which is related to super-resolution reconstruction based on non-local similarity and operator learning is proposed.The specific research content is as follows:1.An approach for image super-resolution with adaptive parameters is proposed under the framework of classical image super-resolution based on sparse representation.Instead of setting parameters by experience,the parameters in the proposed approach can be determined by the property of patches.It overcomes the shortage of that all the patches share the same parameter and improves the performance.The results of the experiments demonstrate that the proposed approach performs better than that without adaptive parameters in different magnification factors and noisy environment.The three valuating indicators prove the effectiveness of the proposed approach.2.Image super-resolution based on non-local similarity and operator learning.Considering that image super-resolution reconstruction usually uses interpolation algorithm to perform preliminary reconstruction of the image,we propose a fast learning-based reconstruction algorithm instead of interpolation algorithm so that low-resolution image patches can be obtained by multiplying only two small matrices and the reconstruction effect is stronger than the interpolation algorithm.According to the non-local similarity property of the natural image patch,we use non-local similarity algorithm to post-process the image,so that the reconstruction effect of the image is further improved.The experimental results show that the proposed method is effective.