Research On Image Super-Resolution Algorithm Fused With Frequency Domain Information

As multimedia technology becomes more and more widely used in daily life,digital images have become the main medium for people to receive external information,so people have higher requirements for the quality of digital images.The image super-resolution algorithm aims to reconstruct a higher quality image from the existing low-resolution image,and because of its low cost of implementation and excellent application performance,it has become an important way to improve image quality.In this thesis,an image super-resolution algorithm combining frequency domain and space domain and an attention-based feature fusion algorithm are proposed to solve the problem that conventional super-resolution algorithms cannot effectively reconstruct the image structure information and improve the quality of super-resolution images.The main research content of this thesis is as follows:(1)Image super resolution algorithm combining frequency domain and space domain.This algorithm proposes a dual-branch network for feature extraction and feature reconstruction in both the frequency domain and the space domain according to the characteristics of images,and properly fuses the frequency domain features and the space domain features through systematic reconstruction,which effectively preserves the structural and texture information of images and generates photo-realistic super-resolution images.The experimental results demonstrate that this algorithm can effectively reconstruct the structural information in the image and improve the visual quality of the super-resolution image.(2)Feature fusion algorithm based on attention mechanism.Based on the idea of attention,this algorithm designs and implements a new attention network layer and a feature fusion module based on the attention network layer,which performs weighted fusion of frequency domain features and space domain features in the dual-branch network,thereby suppressing redundant and repetitive features in the network and improving the feature fusion performance.The experimental results show that the algorithm can effectively enhance the representation ability of the network,and further improve the accuracy and visual quality of super-resolution images.