Low-Light Image Enhancement Method Based on Optimized Generative Adversarial Network

This paper proposes a low-light image brightness restoration method based on an improved GAN network. The approach leverages CycleGAN, which does not require paired datasets, simplifying implementation. PatchGAN is used as the discriminator to capture fine-grained image details and accelerate network convergence. A brightness equalization loss function is introduced to ensure balanced brightness in the enhanced images, while a CBAM (Convolutional Block Attention Module) is incorporated to improve feature extraction and pooling effects. Experimental results demonstrate the effectiveness of the proposed algorithm, achieving superior image enhancement in terms of both visual quality and objective metrics. Future work will explore integrating denoising modules to further enhance image quality.