With the development of intelligent era, information captured in low-light environments has become increasingly vital. Low-light enhancement technology is now a significant research topic in the domain of machine vision. Designing a robust low-light enhancement algorithm can not only improve the contrast of images, but also restore color and texture details, so as to obtain more distinct and accurate low-light scene information.

The team led by Prof. Danhua Cao from Huazhong University of Science and Technology (HUST), China, is committed to the research of low-light enhancement technology. They have developed a size-controllable low-light enhancement algorithm based on neural networks that effectively balances enhancement performance with inference speed. Inspired by the signal processing approach of digital cameras, the algorithm first brightens low-light images and then corrects degradation factors through a two-stage network. Experiments demonstrate that their scheme can enhance images with superior noise suppression and color cast correction while maintaining a smaller model size. The work entitled “Low-light enhancement method with dual branch feature fusion and learnable regularized attention” was published on Frontiers of Optoelectronics (published on Aug. 14, 2024).
DOI: 10.1007/s12200-024-00129-z