Noisy Low-Light Image Enhancement using Reflectance Similarity Prior

Abstract

Images captured in low-light conditions, such as night vision, shadow or backlit, often suffer from low contrast and color distortion, which lose details and naturalness. In this paper, a noisy low-light image enhancement algorithm is presented using reflectance similarity prior. In order to preserve the hue and saturation, the proposed method is performed on the value channel of the HSV color space. First, we propose the reflectance similarity prior by exploring the characteristic of the reflectance component. Then, according to the proposed prior, the image is decomposed into reflectance and illumination components simultaneously based on the Retinex model. This step constrains the reflectance component from color, structure and texture. Finally, the illumination is adjusted by the Gamma correction. Compared with the existing state-of-the-art methods, the proposed method produces high-quality enhanced images, which achieve better subjective and objective assessments.