Lattice Boltzmann model of anisotropic diffusion for image denoising

To overcome the inefficiency of the traditional numerical methods that implement the anisotropic diffusion model for image denoising, a novel lattice Boltzmann model of anisotropic diffusion is presented in this paper. In the model, the diffusion rate is adapted to the image itself and independently set for each direction of diffusion. The mass accumulation is calculated through a weighted summation of the particle distribution functions. Our method is stable with large iteration steps thus reducing the iteration steps greatly. The experiment results showed that compared to others, our approach performs better in terms of the resulting images as well as computing efficiency. In addition, our approach is easy for parallel implementation.

1. INTRODUCTION

Image denoising is a fundamental problem in the field of image processing and computer vision. Numerous image denoising algorithms exist in the computer vision and image processing literatures. The nonlinear diffusion approach was first proposed by Perona and Malik [1]. Later their original model was improved by Catt!e et al [2] from both theoretical and practical viewpoints. An anisotropic extension with a diffusion tensor was presented by Weickert [3] to improve the processing quality. Nonlinear anisotropic diffusion has been proposed and applied to various fields in image processing in recent years,