Illumination enhancement using adaptive regularization and an improved joint model

Abstract

The Retinex method, inspired by human visual perception, has been widely used for image illumination enhancement due to its ability to effectively separate illumination and reflectance components. However, the problem of enhancing illumination while preserving details for images from different scenes remains a difficult task. This paper enhances the method proposed by Cai et al, referred to as ‘JIEP’, by using adaptive weighted regularization parameters based on brightness, achieving stable results with PSNR at 18, SSIM at 0.6, LOE (light order error) at 171, and ARIS at 2.02. The results indicate that the proposed method, a joint intrinsic-extrinsic prior model based on adaptive brightness weights (JIEP_ABW), preserves structural and detailed information while achieving high illumination authenticity.