Research on Low-Light Image Enhancement Algorithm Based on Dual-Branch Structure

This study addresses the prevalent issues of color bias and detail loss in low-light images, which significantly affect image quality. A low-light enhancement algorithm based on the combination of convolutional neural networks (CNN) and a dual-branch structure is proposed to address the problems of detail loss and color shift that existing single-stage CNN models cannot effectively handle. This module is trained with a contrastive regularization method on the basis of the dual-branch structure to ensure the consistency of detail distribution between the generated images and reference images. In terms of the loss function, a color loss function Lcolor is added to balance the low-light enhancement and color bias issues. Experimental results show that the pre-trained model achieved a peak signal-to-noise ratio (PSNR) of 22.133 dB and a structural similarity (SSIM) of 0.873 on the LOL dataset. This indicates that the proposed algorithm significantly improves the balance between image brightness and detail retention, and performs well in enhancing the brightness of natural low-light scene images.