Super-resolution algorithm based on double attention mechanism

Abstract

The invention discloses a super-resolution algorithm based on a double attention mechanism. The super-resolution algorithm comprises the following steps: performing rough feature extraction on a low-resolution image through a shallow feature extraction part; a shallow feature extraction module is used for obtaining a feature map of the low-resolution image, and a deep feature extraction part is used for further extracting high-frequency feature information of the image from the feature map; pooling the feature map along a channel axis, effectively enhancing an area carrying high-frequency information by using a space attention mechanism, and improving the reconstruction quality of the image; the high-frequency information of the image is reconstructed by using double attention, and the low-frequency information of the image can be correctly and stably reconstructed by using a Charbonnier loss function. According to the method, super-resolution detection is carried out on the image by constructing the residual network model, the input image is directly detected, the method is suitable for image detection of any size and various different super-resolution technologies, and the universality is high; the detection time is short, detection can be implemented, and the detection efficiency is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a super-resolution algorithm based on a double attention mechanism. Background technique [0002] Under certain conditions, super-resolution reconstruction technology can overcome the limitation of the inherent resolution of the image system and improve the resolution of the processed image. Therefore, the application of super-resolution technology is growing rapidly, and it has a wide range of practical applications, such as medical imaging , face recognition, remote sensing analysis, public security, etc. Currently, image super-resolution reconstruction methods are mainly divided into three categories: interpolation-based methods, reconstruction-based methods, and learning-based methods. [0003] The interpolation-based method regards the pixel in the image as a point on the image, and the super-resolution image reconstruction can be regarded as fitting the unknown pi...

AI Technical Summary

Problems solved by technology

[0007] The above scheme has a certain effect on image reconstruction, but does not take into account the weight of low-frequency information and high-frequency information of the image, and the texture details of the image are not restored clearly enough, especially in the edge part of the image

Using MSE as a loss function increases the amount of calculation of the model while consuming a lot of space, and the model reduces the robustness of the model

Images