62 / 2021-07-15 11:34:36

BEN: Brightness Enhancement Network for Low- Light Image Enhancement in Complex Environment

Special Session > SS7 - Data-driven Engineering Practices in The Fields of Energy, Environment, and Biomedicine

In recent years, a large number of methods have been used for Low-Light (LL) image processing. Convolutional Neural Networks (CNNs) are widely used in the field of low-light image enhancement because they avoid complicated image preprocessing. Although many studies are devoted to the application of LL image processing in industrial environments, intelligent transportation and other fields, there are few studies on structures suitable for this task. Based on the idea of fusing image features of different illumination layers, the Brightness Enhancement Network (BEN) is proposed. We verified the performance of the structure on public datasets and compared it with some existing advanced methods for processing LL images. Experiments show that our method is superior to other methods in SSIM, PSNR, and NIQE, and it can enhance the illumination while retaining the original details.