Structure-Aware Wavelet Diffusion for Low-Light Image Enhancement with YCbCr Guidance

Restoring visibility in low-light conditions without compromising structural integrity remains a critical challenge in computer vision. Existing diffusion-based enhancement techniques are frequently hindered by color shifts stemming from entangled RGB conditioning, as well as structural distortions caused by isolated frequency processing. To address these limitations, a novel Structure-Aware Wavelet Diffusion Framework is proposed. Diverging from conventional RGB conditioning, a YCbCr-Guided Spatial Feature Transform (SFT) is introduced to explicitly decouple and modulate luminance and chrominance, ensuring high color fidelity. Furthermore, an Inter-Subband Interaction (ISIA) module is developed. By leveraging cross-attention mechanisms, it transfers essential structural cues from low-frequency components to high-frequency textures, effectively suppressing visual artifacts. Comprehensive evaluations on the benchmark LOL-v1 dataset demonstrate that the presented framework achieves state-of-the-art structural preservation (SSIM of 0.87) and perceptual quality (LPIPS of 0.13), significantly outperforming recent regression and generative models.