Abstract: A novel local-structure-adapted image diffusion algorithm is proposed. During the image diffusion, the conduction coefficient is determined not only by its gradient but also by its local structure. First, pixels are classified into three basic types, that is, isolated, image structure, and flap region, according to their local structure. Different conduction functions are defined for different pixel types to calculate their diffusion flux based on gradients. With local image structure taken into account, fine image details can be preserved while non-structured noises are eliminated quickly and efficiently. Experimental results on both synthetic and real images show that the proposed algorithm outperforms many typical image diffusion algorithms.