Two-Stage Low-light Image Enhancement Based on Wavelet Transform

In low-light environments, images often suffer from insufficient brightness, decreased contrast, and blurred details, leading to significant degradation of visual quality. To address these issues, this paper proposes a dual-stage wavelet transform-based method for low-light image enhancement. The method builds upon wavelet transform theory and employs a U-Net architecture to progressively achieve feature encoding, decoding, and enhancement through two sequential stages: preliminary restoration and fine-grained enhancement. For effective noise suppression and detail enhancement, we design an enhanced wavelet-domain feature fusion module that integrates discrete wavelet transform, inverse discrete wavelet transform, and dual attention mechanisms. Meanwhile, the proposed dynamic gated spatial attention and lightweight fusion-curve attention mechanisms collaborate within this feature fusion module to process image features with refined adaptability. Additionally, a fusion perceptual loss function is developed to guide the model in generating visually natural enhanced images with authentic details by jointly optimizing pixel-level errors and perceptual quality metrics. Experimental results demonstrate that our method achieves state-of-the-art performance on key metrics (e.g., PSNR, SSIM) across multiple public low-light datasets, exhibiting superior capabilities in both noise suppression and detail recovery