Image Enhancement Method Based on Multi-Scale Guided Filtering and Attention Mechanism

Due to issues such as insufficient brightness, low contrast, and severe noise interference in images captured under low-light conditions in scenarios like nighttime surveillance, autonomous driving, and security monitoring, this paper proposes a low-light image enhancement method based on multiscale gradient domain guided filtering and attention mechanisms. The proposed algorithm operates in the HSV color space and incorporates a multi-scale strategy into guided filtering to improve the model's ability to extract features at different scales. By integrating attention mechanisms with the Retinex theory, an illumination-reflection enhancement network is constructed to strengthen the model's capability to process regions with varying brightness. This enables effective enhancement of images captured under different lighting conditions and at multiple scales. Experimental results demonstrate that the proposed method outperforms several classical low-light enhancement algorithms by approximately 1.62% in Peak Signal-to-Noise Ratio (PSNR), 2% in Structural Similarity Index (SSIM), and 3.41% in naturalness evaluation metrics. The enhanced images exhibit significantly improved brightness and natural appearance, while preserving fine texture details and color information.