Improved MSR Model for Image Enhancement in Different Color Spaces

During the process of low-light image enhancement, the classic Multi-Scale Retinex (MSR) suffers from halo effects and insufficient color protection. Therefore, this paper proposes an improved MSR model. The core idea of this paper is to replace the Gaussian filter in MSR with gradient-domain guided filtering. Meanwhile, in the process of image enhancement, the differences among various color spaces are taken into account. Experiments have proven that the improved method proposed in this paper can effectively improve the quality of low-light images. However, when applying the proposed method to different color spaces, the experimental results do not confirm the existence of a universal optimal color space model. In the field of image enhancement, it is still necessary to choose the appropriate color space model according to the characteristics of the specific application scenario.