<p>Low-light imaging poses a significant challenge in various applications, including medical imaging, photography, and surveillance, due to the limited availability of illumination. The paper presents a deep decomposer and refiner for low-light image enhancement. The proposed method develops the mf-block to focus on multi-level features using variation of the receptive field. The decomposer sub-module contains down-blocks and up-blocks that leverage channel attention to process the features of mf-block for Retinex-based image decomposition. Moreover, we introduce a hybrid method for adjustment of the estimated illumination combining deep learning-based techniques with non-linear statistical methods to adjust non-uniform illumination. Additionally, we developed a refiner sub-module utilizing a similar model to reduce noise and enhance contrast in the reflectance component. The exhaustive experimentation on dataset with various low-light images demonstrates the effectiveness of our approach in significantly improving image details and reducing noise. The proposed method shows results superior to those of state-of-the-art techniques, highlighting its potential for real-world low-light imaging applications.</p>