Image super-resolution method based on pyramid attention mechanism and symmetric network

Abstract

The invention discloses an image super-resolution method based on a pyramid attention mechanism and a symmetric network, and the method comprises the steps of processing a to-be-trained data set, and making a low-resolution image corresponding to a real image; secondly, performing feature extraction on real image input, performing calculation through a pyramid attention module after convolution-pooling operation, and obtaining multi-level information distribution and detail features of the image; obtaining error loss through calculation for updating network parameters and gradients; carrying out reconstruction to obtain a reconstructed result image, carrying out loss calculation on the reconstructed result image and a real image, and reversely updating parameters and gradients of the network; and finally, inputting the low-resolution image into a second network, amplifying the low-resolution image in a resolution-by-resolution manner, finally obtaining a generated high-resolution image, calculating loss of a real image, and updating parameters of the network again. According to the method, parameter updating and iteration are performed on the network through the attention mechanism and the loss function, so that the generation capability and generalization capability of the network are greatly improved.

Description

technical field [0001] The invention belongs to the field of computer vision and visual image reconstruction, and mainly relates to an image super-resolution method based on a pyramid attention mechanism and a symmetrical network. Background technique [0002] Image super-resolution algorithm is a branch research direction in the field of computer vision, and it is widely used in urban monitoring, medical image, remote sensing and other fields. Its main task is to input the collected low-resolution images into the network, which is trained and deduced by the network, and finally generates high-resolution images with more details and texture features. Although image super-resolution algorithms have been extensively studied, the high-resolution images obtained by most traditional methods have problems such as smoothness, ringing, and low definition, and the processing of details still cannot meet the expected standards. [0003] The current image super-resolution algorithms c...

AI Technical Summary

Problems solved by technology

However, the image super-resolution algorithm based on the deep learning network has made a qualitative leap in image accuracy, but the speed of model training and deduction and the size of the model have also increased, and more and more convolutional layers are stacked into the super-resolution network. Among them, it is difficult to apply the algorithm on a platform with limited resources, and most of the image super-resolution algorithms based on the deep convolutional network structure still have problems such as fuzzy details, smoothness, and pseudo-details. Quality still needs to improve

Images