Low-light Image Enhancement Algorithm Based on Transformer

Low light image enhancement is an important task in computer vision. The Low-light images can adversely affect more advanced computer vision tasks such as target tracking, image segmentation, and target recognition due to the presence of large amounts of noise and low contrast. Therefore, research on low-light image enhancement is crucial. To address these issues, this paper delves into the comparison between traditional and non-traditional low-light image enhancement algorithms and has achieved the following results:We abandon the traditional approach based on a priori knowledge of images and instead redesign a new network structure for low-light image enhancement with the existing theoretical foundation for Transformer. The network performs information extraction in the channel dimension of the image through self-attention mechanism and then extracts the shallow feature information of the image through Shallow layer. After fusing these two types of information, more advanced information is extracted to guide the final image enhancement. Experimental results show that the redesigned network structure not only achieves better results but also has improved execution efficiency.