Deep Multi-path Low-Light Image Enhancement

Noise and color shifts are prone to occur in low-light conditions, which can significantly affect imaging quality. Simply stretching the pixel intensity to improve the illumination level is not able to eliminate noise and color shifts, and it may even lead to artifacts amplification. Various methods have been developed to avoid noise and color shifts in low-light conditions over the past few years. However, most of them may fail to handle different exposure conditions at the same time. In this paper, we propose a novel multi-path convolutional neural network architecture for this task. Luminance and chroma were impacted in two different ways in low-light conditions. By using different well-designed custom networks and loss functions, luminance and chroma can be well restored. Extensive experiments have been performed to demonstrate the superiority of our approach to the state-of-Art methods using multiple metrics under different exposure conditions.