Image super-resolution method based on recursive attention mechanism

Abstract

An image super-resolution method based on a recursive attention mechanism belongs to the technical field of image processing and comprises the steps of image preprocessing, super-resolution network model construction, network training objective function determination, super-resolution network training and super-resolution image reconstruction connection. Due to the fact that the super-resolution network model is adopted, different areas of input information are subjected to different weight processing, and attention feature graphs are mined for multiple times in a recursive mode on image features by using the same set of parameters, the network parameter quantity is reduced, the network operation speed and efficiency are improved, source image information is reserved, and the super-resolution precision and the quality of a super-resolution reconstructed image are improved. The invention has the advantages of being good in image super-resolution effect, small in memory required by calculation, high in calculation speed and the like, and can be used for image super-resolution.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an image super-resolution method based on a recursive attention mechanism. [0002] technical background [0003] The process of reconstructing a corresponding high-resolution image from a single low-resolution image is called single-image super-resolution reconstruction. Single image super-resolution reconstruction has a wide range of applications in medical, military, security and other fields. [0004] In the image super-resolution reconstruction task, early methods mainly used interpolation methods based on sampling theory. These methods have fewer parameters and are generally faster than learning-based methods. However, these methods have the disadvantages of low image reconstruction accuracy and poor visual effects. . Early learning-based methods mainly established the mapping function between low-resolution images and high-resolution images through sp...

AI Technical Summary

Problems solved by technology

Although the iterative neural network-based super-resolution algorithm has significantly improved the accuracy of image super-resolution, the increase in the number of network layers and channels has also increased the time-consuming calculation of super-resolution.

At the same time, the sharp increase in the number of parameters increases the memory required for the algorithm to run proportionally, which makes the application conditions of these neural network-based super-resolution algorithms harsh.

Images