Unsupervised Extremely Low-Light Image Enhancement with a Laplacian Pyramid Network

Former unsupervised extremely low-light image enhancement methods suffer from two issues: semantic information loss and insufficient noise suppression. To overcome these two problems, we propose an unsupervised Extremely Low-light image enhancement via a Laplacian Pyramid Network (ELLPN). Concretely, concerning the first quandary, we propose to enforce semantic content and style constraints in the low-frequency components of the image’s Laplacian pyramid after histogram equalization, therefore realizing image enhancement. As for the second issue, a generalized denoising module is introduced to process the high-frequency components of the image’s Laplacian pyramid after histogram equalization, thus further restoring the details of the image. Extensive analytical experiments substantiate the efficacy of our approach.