Abstract: Traditional low-light image enhancement methods cannot simultaneously preserve texture details and suppress the noises. In view of this problem, we propose an algorithm of low-light image enhancement based on the unified variational model. The classic Retinex model was modified by adding the noise term. We constructed a unified variational model based on Gaussian total variation (GTV) and L2 norm regularizations to constrain the illumination term, the reflectance term and the noise term. Subsequently, the alternating direction minimization technique was adopted to iteratively solve the unified variational model for simultaneously obtaining the illumination component, the reflectance component and the noise component. The illumination was adjusted using gamma correction and then it was multiplied with the reflectance to acquire the final resultant image. Experimental results and comparative data show that the proposed algorithm performs better than most other methods on natural image quality evaluator (NIQE) and information entropy (IE) and can suppress the noises of dark regions while effectively keeping the texture details.