“…Because of monotonicity of the logarithm operation, it easily follows that r ≤ 0 and s ≤ l . By introducing variables u , v such that the logarithmic illumination l and the logarithmic reflectance r can be reformulated as Usually, the logarithmic reflectance r = − u is close to the difference between the logarithmic observed image s and the logarithmic illumination l ; see, for instance, the work of Wang et al By assuming that both the logarithmic reflectance r and the logarithmic illumination l are spatially smooth (see, for instance, the works of Wang et al and Kimmel et al), the following constrained optimization model is proposed: where β , ω , μ , ν are positive regularization parameters and ‖·‖ 2 denotes the Euclidean norm. In addition, is the fidelity term, the regularization terms and concur with the smoothness assumption on both r and l .…”