Low-Light Image Enhancement Algorithm Integrating Retinex Illumination Estimation and Multi-Attention

To address the challenges of poor visibility, blurred textures, and distortions in low-light conditions, this paper presents an image enhancement algorithm that integrates Retinex-based illumination estimation with multi-attention mechanisms. The algorithm introduces a perturbation component into the Retinex model to simulate image degradation, enabling more accurate illumination estimation. Additionally, multi-attention mechanisms are incorporated to strengthen the model’s ability to capture both local and global features. An Illumination Estimation Module is proposed to predict illumination information of the image. A Degradation Restoration Module is designed to boost the visibility of the illumination map and eliminate artifacts. A Multi-Scale Color Correction Module is introduced for color correction on illumination feature maps. The preliminary enhanced image constructed by the Illumination Estimation Module and the restored map from Degradation Restoration Module are residually fused, and the color-corrected map from Multi-Scale Color Correction Module is integrated in the channel dimension with a scaling factor α, completing the enhancement of low-light images. Comprehensive quantitative and qualitative evaluations across seven benchmark datasets validate the proposed algorithm effectively removes artifacts, significantly improves image quality, and enhances visual perception.