Single image super-resolution reconstruction method based on meta-learning

Abstract

The invention relates to a single image super-resolution reconstruction method based on meta-learning. The method comprises the following steps: pre-training; constructing a meta-learning data set; adopting a meta learning method to learn a meta model from external data; quickly generating an image super-resolution reconstruction model in a specific environment by using the meta-model in an actual environment; and generating a high-resolution image corresponding to the low-resolution image in the specific environment. The invention is reasonable in design, external big data and a small amount of data in a specific environment are fully utilized, the applicability problem of a super-resolution reconstruction model is solved, and the invention can be effectively applied to a real environment.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, in particular to a method for realizing super-resolution reconstruction of a single image. Background technique [0002] From a high-resolution image to a low-resolution image is an image degradation process. Single image super-resolution reconstruction refers to reconstructing a corresponding high-resolution image from a low-resolution image, and its essence is to restore the original image from the degraded image. This is an incomplete inverse problem, and a lot of information is lost during the degradation process, so multiple different high-resolution images can be generated from a degraded low-resolution image, and there is no unique solution. [0003] After the emergence of deep neural networks, using big data, based on convolutional neural networks and generative adversarial networks to model the inverse process of the degradation process from high-resolution images to lo...

AI Technical Summary

Problems solved by technology

[0004] However, there are two problems in the application of the above method in the real environment: on the one hand, the low-resolution image in reality may be obtained by various unknowable degradation methods and their combinations, once the degradation method corresponds to the data used in model training Different degradation methods, the learned model is not applicable, and the effect of super-resolution reconstruction will be greatly reduced; on the other hand, after the learning is completed, because the images in different degradation environments are fixed models, Therefore, the data information contained in the low-resolution images in the new degraded environment has not been fully utilized.

Images