To augment precision and computing efficiency of traditional Lattice Boltzmann (LB) methods in speckle reduction, this paper proposes a multi-grid LB (M-LB) method. The method automatically realizes anisotropic image diffusion of LB by applying different scales of grid: fine-grids are applied to regions with
obvious gradient change, and coarse-grids applied to regions with tiny gradient change. Experiments were carried out to test the method for speckle noise reduction. The proposed M-LB method was compared to an existing multi-grid method and two traditional LB methods. Natural images, composite images and medical ultrasound images were used. The experiments show that the M-LB method can achieve better denoising results and higher computation efficiency.