Low-Illumination Image Enhancement Based on End-to-End Network Using Attention Module

Images are always susceptible to variations in light which makes the low-illumination image enhancement an important task. Conventional low-illumination image enhancement methods are typically implemented by improving image brightness and contrast, while suppressing image noise simultaneously. Recently, the deep learning-based methods have also been applied to image enhancement. However, the restoration of the original brightness and detailed textures in dark images remains challenging. In this paper, an end-to-end neural network is proposed. The coordinate attention (CA) module and the squeeze excitation(SE) module are introduced to refme and highlight key features. A perceptual loss function is also proposed to enhance the texture of the details and restore the visual distortion. The effectiveness of the proposed network is demonstrated in experiments on popular datasets.