No-reference low-illumination image enhancement method and system based on generative adversarial network

Abstract

The invention relates to a no-reference low-illumination image enhancement method and system based on a generative adversarial network. The method comprises the following steps of: respectively preprocessing a low-illumination image and a normal-illumination image with original resolutions to obtain unpaired low-illumination image blocks and normal-illumination image blocks for training; constructing a generator network for low illumination enhancement and a discriminator network for adversarial training; alternately training the generator network and the discriminator network to converge to Nash equilibrium by using the low-illumination image blocks and the normal-illumination image blocks; and inputting an original low-illumination image for testing into the trained generator network topredict the enhanced image. The method and the system are beneficial to improving the enhancement quality of the low-illumination image.

Description

technical field [0001] The invention belongs to the technical field of image and video processing, in particular to a non-reference low-light image enhancement method and system based on a generative confrontation network. Background technique [0002] As photographing equipment becomes more and more popular and portable, people can capture images with good visual effects. However, it is still very difficult to obtain photographic images with good visual effects in low-light scenes, such as night scenes or dim interiors. Because the scene is poorly lit or shot against a backlight, the captured image will appear underexposed, and this type of image is called a low light image. Image detail is barely visible in some areas of low-light images. Low-light images not only present lower visual effects, but also affect the performance of many fundamental computer vision and image processing methods, such as image classification, image saliency detection, object detection and recog...

AI Technical Summary

Problems solved by technology

However, it is difficult to obtain paired datasets that can be used for deep learning model training on the problem of low-light image enhancement.

Most of the existing low-light image enhancement methods use synthetic paired low-light image datasets. Due to the differences between synthetic datasets and real images, the image enhancement ability of methods trained using synthetic datasets in real scenes cannot be guaranteed.

Images