Adaptive Non-linear Diffusion Based Local Binary Pattern for Image Denoising

Abstract

Non-linear diffusion approaches are an effective way to reduce noise and preserve the edge information. In this paper, the work is extended to integrate the non-linear diffusion and local binary pattern (LBP) textons, where the diffusivity function is adapted to pixels type after LBP classification. This allows smoothing on homogenous and noisy regions but not on edges. The proposed method preserves edges better because the diffusion is controlled taking into account the difference of diagonal neighbors in addition to four nearest neighbors. Experimental results on synthetic and real images illustrate the effective performance of the proposed method.