Research On Image Super-resolution Based On Decoupling Thought And Attention Mechanism

Image super-resolution usually refers to a low-level vision enhancement technology that uses computer algorithms to increase image resolution to improve visual effects.In recent years,with the rapid development of deep learning methods,deep learning has become more and more widely used in the low-level vision field.At present,it has shown good performance in image super-resolution,image deblurring and other fields.This paper focuses on the underlying visual tasks of deep learning images,combines decoupling ideas,image depth priors and attention mechanisms to study image super-resolution depth models,and proposes a new image super-resolution method.The experimental results show that the proposed method is effective.This article discusses several issues concerning the superresolution and motion deblurring of the low-level visual image:First,Joint image super-resolution and motion deblurring based on decoupling learning.A deep super-resolution model based on the idea of decoupling learning is studied.It aims to solve the problem of simultaneous super-resolution and motion deblurring of a single image when multiple degradation factors are involved.It is compatible with various existing image super-resolution and motion.We verify its performance by Comparing with the combined deblurring method.Next,An attempt of blind image super-resolution based on the idea of serial decoupling of resolution reduction and blur degradation.Explore a blind-based image super-resolution model which aims to decouple the lowresolution degradation factors of the image from the blur kernel and introduce a degradation network to realize the image blind super-resolution,we also verify its performance by comparing it with various existing blind super-resolution models.Finally,Super-resolution of video images combined with channel attention and position attention.The difference between video super-resolution and image super-resolution is that video super-resolution needs to consider the time consistency and frame alignment between frames.Directly applying the image super-resolution algorithm to video super-resolution often does not get the best results.We will pay attention to applying the attention mechanism is to the video super-resolution reconstruction network.At the same time,channel attention and position attention are used to mine the characteristics of multi-frame video images and introduce non-local residual blocks to achieve video super-resolution.The experiments on Youku's public data set shows that the proposed channel attention and position attention networks can achieve better results.