Low-light image enhancement with a refined illumination map

It has become very popular to take photographs in everyone's daily life. However, the visual quality of a photograph is not always guaranteed due to various factors. One common factor is the low-light imaging condition, which conceals visual information and degenerates the quality of a photograph. It is preferable for a low-light image enhancement model to complete the following tasks: improving contrast, preserving details, and keeping robust to noise. To this end, we propose a simple but effective enhancing model based on the simplified Retinex theory, of which the key is to estimate a good illumination map. In our model, we apply an iterative self-guided filter to refine the initial estimation of an illumination map, making it aware of local structure of image contents. In experiments, we validate the effectiveness of our method in various aspects, and compare our model with several state-of-the-art ones. The results show that our method effectively adjusts the global image contrast, recovers the concealed details and keeps the robustness against noise.