An Automated Multi-scale Retinex for Dim Image Enhancement
Shuozhe Zhang, Minling Zhu, Kun Meng · 2022 IEEE 2nd International Conference on Power, Electronics and Computer Applications (ICPECA) · 2022
Multi-scale Retinex with Color Restoration (MSRCR) suffers from problems such as noise amplified in the original images, especially dim and dark images, and the hue of the output images becoming greyish. Firstly, we take Singular Value Decomposition (SVD) to perform initial noise reduction to solve the problems. Secondly, images convert from RGB to HSV color space model that is more suitable for human vision, and then luminance and saturation are enhanced by automated MSRCR and Gamma stretching, respectively. Besides, to further denoise, enhanced parts are re-combined into RGB color space, and Guided Filtering denoises them once again. Experiments indicate that compared with MSRCR, our method’s denoising effect, color restoration, and image clarity improve by 2.3%, 11%, and 83.3%, respectively. Not only is our method objectively better in terms of evaluation metrics, but also subjectively, it is observed that the images processed by our method are more appropriate to the human visual system. What is more, our processing time is the most optimal among all comparison algorithms.