This paper proposes a novel method for shadow removal, which includes a shadow-aware decomposition network to derive the shadow reflectance and illumination layers. A novel bilateral correction network is introduced with a local lighting correction (LLC) module and an illumination-guided texture restoration (IGTR) module to recast degraded lighting and restore degraded textures in shadow regions conditionally. The method also annotates shadow masks for the SRD benchmark to enable fair evaluation with existing shadow removal methods. The proposed method outperforms existing state-of-the-art shadow removal methods on three benchmarks, achieving superior performance in terms of RMSE and PSNR. The method is robust to different types of shadow annotations and is effective in reducing shadow ghosting, maintaining color consistency, and correcting textures. The method is implemented using PyTorch on a single NVIDIA TESLA V100 GPU. The method has slightly larger parameters and extended inference times due to the use of diffusion models. The method is evaluated on three shadow removal datasets: SRD, ISTD, and ISTD+. The results show that the proposed method achieves state-of-the-art performance in shadow removal. The method is based on the retinex theory and uses a two-step approach to estimate and rectify the illumination in shadow regions first, followed by the restoration of degraded textures in these regions, conditioned on the recovered illumination. The method is evaluated using quantitative metrics such as RMSE, PSNR, and SSIM. The method is also evaluated using qualitative comparisons with existing shadow removal methods. The method is robust to different types of shadow annotations and is effective in reducing shadow ghosting, maintaining color consistency, and correcting textures. The method is implemented using PyTorch on a single NVIDIA TESLA V100 GPU. The method has slightly larger parameters and extended inference times due to the use of diffusion models. The method is evaluated on three shadow removal datasets: SRD, ISTD, and ISTD+. The results show that the proposed method achieves state-of-the-art performance in shadow removal.This paper proposes a novel method for shadow removal, which includes a shadow-aware decomposition network to derive the shadow reflectance and illumination layers. A novel bilateral correction network is introduced with a local lighting correction (LLC) module and an illumination-guided texture restoration (IGTR) module to recast degraded lighting and restore degraded textures in shadow regions conditionally. The method also annotates shadow masks for the SRD benchmark to enable fair evaluation with existing shadow removal methods. The proposed method outperforms existing state-of-the-art shadow removal methods on three benchmarks, achieving superior performance in terms of RMSE and PSNR. The method is robust to different types of shadow annotations and is effective in reducing shadow ghosting, maintaining color consistency, and correcting textures. The method is implemented using PyTorch on a single NVIDIA TESLA V100 GPU. The method has slightly larger parameters and extended inference times due to the use of diffusion models. The method is evaluated on three shadow removal datasets: SRD, ISTD, and ISTD+. The results show that the proposed method achieves state-of-the-art performance in shadow removal. The method is based on the retinex theory and uses a two-step approach to estimate and rectify the illumination in shadow regions first, followed by the restoration of degraded textures in these regions, conditioned on the recovered illumination. The method is evaluated using quantitative metrics such as RMSE, PSNR, and SSIM. The method is also evaluated using qualitative comparisons with existing shadow removal methods. The method is robust to different types of shadow annotations and is effective in reducing shadow ghosting, maintaining color consistency, and correcting textures. The method is implemented using PyTorch on a single NVIDIA TESLA V100 GPU. The method has slightly larger parameters and extended inference times due to the use of diffusion models. The method is evaluated on three shadow removal datasets: SRD, ISTD, and ISTD+. The results show that the proposed method achieves state-of-the-art performance in shadow removal.