Learning to Adapt to Light

Light adaptation or brightness correction is a key step in improving the contrast and visual appeal of an image. There are multiple light-related tasks (for example, low-light enhancement and exposure correction) and previous studies have mainly investigated these tasks individually. It is interesting to consider whether the common light adaptation sub-problem in these light-related tasks can be executed by a unified model, especially considering that our visual system adapts to external light in such way. In this study, we propose a biologically inspired method to handle light-related image enhancement tasks with a unified network (called LA-Net). First, we proposed a new goal-oriented task decomposition perspective to solve general image enhancement problems, and specifically decouple light adaptation from multiple light-related tasks with frequency based decomposition. Then, a unified module is built inspired by biological visual adaptation to achieve light adaptation in the low-frequency pathway. Combined with the proper noise suppression and detail enhancement along the high-frequency pathway, the proposed network performs unified light adaptation across various scenes. Extensive experiments on three tasks- low-light enhancement, exposure correction, and tone mapping-demonstrate that the proposed method obtains reasonable performance simultaneously for all of these three tasks compared with recent methods designed for these individual tasks.