We present a new lattice Boltzmann method for vector image nonlinear diffusion by embedding vector image’s edge feature into the relaxation parameter of the Lattice Boltzmann equation and designing high efficient equilibrium distributions. This overcomes the problems of poor efficiency and the difficulty in parallelization of the traditional numerical methods using nonlinear diffusion model for vector image denoising. To keep the algorithm stable, the method can iterate with large step to improve computation efficiency. Influence of equilibrium distribution on computation accuracy and efficiency is evaluated in the numerical simulation of second dimensional diffusion problem. To illustrate effectiveness of the method, color images contaminated by Gaussian noise and impulsive noise are tested. Experimental results and the analysis show that computation accuracy and efficiency of the proposed method are better than that of the traditional numerical method such as additive operator splitting (AOS) algorithm