A structure and texture revealing retinex model for low-light image enhancement

Low-light image enhancement is a crucial yet challenging task in computer vision and mul- timedia applications. Retinex-based approaches have been continuously explored in this domain. However, the Retinex decomposition is an ill-posed problem, as the proper con- straints of illumination and reflectance should be considered to regularize the solution space. Aiming at a faithful enhancement, we develop a Structure and Texture Revealing Retinex (STR2 ) model to accurately estimate the illumination and reflectance components. The pro- posed STR 2 model utilizes an exponential relative total variation method to draw structure and texture maps by analyzing the difference in gradient distribution between the illumi- nation and reflectance components. The resulting structure and texture maps are used to regularize the illumination and reflectance components. With a tailored alternating opti- mization algorithm, the STR 2 model can jointly update the illumination and reflectance efficiently to produce a faithful enhanced image. Experimental results on several public datasets verify the effectiveness of the proposed model in low-light image enhancement.