Unpaired strong-light image augmentation network with illumination curve

Abstract

Images are a way for humans to get information, and they are ubiquitous in our daily lives. Compared to other media, we tend to get information through images. Most current image enhancement methods are based on Retinex theory. However, this method can easily lead to the loss of image information. More importantly, the theory is based on the assumption that light changes smoothly. And this assumption is very unfriendly to bright images. To solve this problem, an unpaired enhancement network for strong light image is proposed. The network consists of illumination curve, illumination loss function and illumination attenuation network. Experiments show that our network can obviously improve the image contrast caused by strong light environment, and significantly improve the quality of strong light image.