Non-Uniform Illumination Image Enhancement via a Retinal Mechanism Inspired Decomposition

For many consumer image processing applications, it is highly required to maintain scene details faithfully at low computational cost. Existing enhancement methods often fail to meet this requirement. In this paper, we present a straightforward yet effective method to enhance non-uniformly illuminated images. Unlike existing methods based on physical illumination models and deep learning based approaches, our method does not make any explicit physical assumptions about the image formation process, nor does it require any learning procedures. This is achieved through a novel contrast/residual decomposition perspective, which separates an image into two components: a contrast containing the details in the scene, and a residual containing the variations in illumination. This decomposition is performed at multiple scales, and the enhanced image at each scale is produced by modifying both components and adding them back. The output image is achieved by fusing the multi-scale results using a residual based weighting scheme. Subjective and objective experiments on multiple datasets demonstrate that the proposed method produces competitive results compared to state-of-the-art methods. Our method is easy to use and is well-suited for image processing applications in consumer electronics. The project website is https://github.com/retinexpt/CRIE.