Perceptive low-light image enhancement via multi-layer illumination decomposition model

The images captured under low light conditions generally have less than satisfactory visual quality. To address this issue, many low-light image enhancement methods have been studied. However, these existing algorithms mostly suffer from unnaturalness, over-enhancement and artifacts. In this paper, a perceptive low-light image enhancement via multi-layer illumination decomposition model is proposed, to preserve the naturalness and improve the contrast for low-light images. First, the contrast of the target image is defined from global, local and the effect of noise aspects. Then, inspired by the human visual system, the perceptive contrast is designed by combining the defined contrast with just-noticeable-difference transformation. Last and most importantly, the target image is decomposed in a multi-layer way based on the multi-scale adaptive filter, which utilizes the perceptive contrast to decide the variance adaptively. This step can effectively obtain multiple illumination and reflectance layers. Combining these reflectance with adjusted illumination components can generate the final enhanced result. The proposed method has better no-reference quantitative measurement results than other compared methods. Experimental results on several public challenging low-light image datasets demonstrate that the proposed method can achieve great performance in balancing the contrast, brightness and naturalness.