Nonlinear Diffusion Driven by Local Features for Image Denoising

We propose a nonlinear diffusion algorithm that takes into account the local features in an extended neighborhood for the image denoising. In the conventional linear or nonlinear diffusion algorithms, the change of intensity value of a pixel is considered only in a small neighborhood, and the relationship between pixels in larger region is neglected. Our proposed algorithm overcomes this limitation. Moreover, it is not simply generalization of the conventional diffusion algorithms. In order to remove the noise, simultaneously, preserve edges in an image, the local central moment in an extended neighborhood is extracted, and the appropriate diffusion coefficient is established, such that the diffusion speed is properly controlled according to the characteristic of each image local region. The divergence of the new flow function is derived, and it has a compact expression format. The relationship between the size of the extended neighborhood and the performance of this proposed method is discussed. The experimental results show the effectiveness of the proposed method for image denoising