A Low-Light Image Enhancement Method Combining Attention Mechanism and Retinex Model

Low-Light Image Enhancement (LLIE) faces difficulties arising from issues like color distortion and noise in low-light environments. Specifically, the underexposed areas of an image often contain more noise than the normally exposed regions. In this paper, we introduce an enhancement algorithm that leverages a channel attention mechanism, enabling effective extraction of key features from low-light images while filtering out irrelevant information to improve overall image quality. Furthermore, to boost image recovery and assist with the Retinex decomposition process, we propose a light smoothing loss function, which enhances the image's structural boundaries. Through extensive experimental assessments, it is evident that the approach we propose exceeds the performance of current leading unsupervised methods.