Unsupervised Low Light Image Enhancement via SNR-Aware Swin Transformer

Image captured under low-light conditions presents unpleasant artifacts, which debilitate feature extraction performance in upstream visual tasks. Low-light image enhancement (LLIE) aims to improve brightness and contrast, further reducing noise that corrupts the visual quality. Recently, many image restoration methods based on Swin Transformer have been proposed, and impressive performance has been achieved. However, on one hand, trivially employing Swin Transformer for LLIE would expose several artifacts, including over-exposure, brightness imbalance, noise corruption, etc. On the other hand, it is impractical to capture image pairs of low-light images and corresponding ground truth, i.e., well-exposed images in the same visual scene, for model training. In this paper, we propose a dual-branch network based on Swin Transformer, guided by a signal-to-noise ratio prior map which provides the spatial-varying information for LLIE. Moreover, we leverage unsupervised learning to construct the optimization objective based on the Retinex model, to guide the training of the proposed network. Experimental results demonstrate that the proposed model is competitive with the baseline models.