Image super-resolution reconstruction method based on dynamic grouping convolution

Abstract

The invention discloses an image super-resolution reconstruction method based on dynamic grouping convolution. The method comprises the steps: 1) carrying out the feature embedding of a low-resolution image ILR, and obtaining an embedded feature graph; 3) performing size amplification on the extracted information graph and the embedded feature graph by using a dynamic grouping convolution up-sampling module to obtain an extracted amplified information graph and an embedded amplified information graph; 4) performing information fusion on the extracted amplified information graph and the embedded amplified information graph to obtain a fused information graph; and a high-resolution image IHR is obtained. The calculation amount and parameter amount of the network model can be effectively reduced, the model performance is improved, and the lightweight performance extension of the model is realized.

Description

technical field [0001] The present invention relates to the technical field of image super-resolution reconstruction, in particular to an image super-resolution reconstruction method based on dynamic group convolution. Background technique [0002] With the development of society, people have higher and higher requirements for image quality, because high-resolution images can provide more texture details, and these details play an extremely important role in many practical applications, for example, in In the medical field, high-resolution medical images allow doctors to judge patients' conditions more accurately; in the field of geographic detection, high-resolution remote sensing satellite images allow researchers to more easily identify similar objects; in the field of computer vision, High-resolution images allow computers to perform better at pattern recognition. [0003] There are two methods for obtaining high-resolution images at present: 1) use high-definition came...

AI Technical Summary

Problems solved by technology

Although with the introduction of deep learning into the field of image super-resolution, the performance of the model is getting better and better, but there are still the following problems: 1) The demand for computing resources is large and the migration is difficult

Most deep learning models only focus on performance, and the number of network layers is extremely deep, resulting in a large number of model parameters and high computational complexity, making it difficult to implement practical applications

2) Weak scalability

However, the usual method of model lightweighting only relies on reducing the number of network layers, which makes it difficult to expand the performance of the lightweight model and the model architecture is not flexible.

Images