Single-image super-resolution reconstruction method based on three-channel convolutional neural network

Abstract

The invention provides a single-image super-resolution reconstruction method based on a three-channel convolutional neural network in the field of image processing. The method comprises the followingsteps: S10, obtaining a data set of an image, and building a plurality of high-resolution images and low-resolution images corresponding to the high-resolution images based on the data set; s20, creating a three-channel convolutional neural network model, training each high-resolution image and each low-resolution image by using the three-channel convolutional neural network model, and generatinga mapping relationship between the low-resolution images and the high-resolution images; s30, optimizing the mapping relationship by using a mean square error loss function; and S40, based on the optimized mapping relationship, inputting a to-be-reconstructed low-resolution image into the three-channel convolutional neural network model, and outputting a reconstructed high-resolution image. The method has the advantages that the quality of the reconstructed image is greatly improved on the premise that the network depth and model parameters are not increased.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a method for super-resolution reconstruction of a single image based on a three-channel convolutional neural network. Background technique [0002] Image is one of the most convenient and fast carriers for transmitting information. It has the characteristics of large amount of information, intuitive and storable, so it is widely used in medicine, public safety, remote sensing, national defense and other fields. However, due to the limitations of imaging equipment and environmental factors, the quality of acquired images is reduced, which is not conducive to the accuracy and integrity of information transmission, so how to improve image quality is very important. Image resolution is an important indicator to measure image quality, which is manifested subjectively as the clarity and richness of image edges and texture details; objectively manifested as the total number of pixels in ...

AI Technical Summary

Problems solved by technology

[0008] Compared with reconstruction-based methods, Markov network-based super-resolution reconstruction methods can obtain rich high-frequency information, and can still reconstruct higher-quality images at higher magnifications, but there is a problem with training samples The selection requirements are high and the shortcomings of being sensitive to noise

Compared with the super-resolution reconstruction method based on the Markov network, the super-resolution reconstruction method based on domain embedding requires fewer training samples and is less sensitive to noise, but has the disadvantage of being difficult to choose the size of the domain

The method based on sparse representation solves the problem of domain size in the super-resolution reconstruction method based on domain embedding, but it has the disadvantage of being difficult to select an over-complete dictionary; randomly selecting an over-complete dictionary can better reconstruct images in a specific domain, but for The reconstruction effect of ordinary images is not ideal

[0009] The super-resolution reconstruction method based on deep learning has the advantages of high reconstruction quality, fast reconstruction speed, and even real-time super-resolution reconstruction. The disadvantage is that the calculation is huge and requires high computer hardware configuration.

Images