To overcome poor efficiency and difficulty in parallelization of traditional numerical method for image denoising with a nonlinear diffusion model, we propose a new lattice Boltzmann method to embed the image edge features into the relaxation parameters of the lattice Boltzmann equation for image nonlinear diffusion. While keeping stability in calculation, the algorithm allows large steps in the iteration so that the computation efficiency is improved. Additionally, the algorithm is easy to be parallelized. Experiment results show that the lattice Boltzmann method has better quality, accuracy and algorithmic efficiency in image denoising as compared to the additive operator splitting (AOS) algorithm.