This paper proposes a new approach of anisotropic diffusion used to remove several kind of noise in order to detect sharp boundaries. Based on Perona-Malik (PM) equation, we propose to adapt the edge magnitude k parameter of the diffusivity function according to the local intensity variance, which is computed for each image pixel. We will show that this technique removes several kind of noise (additive, multiplicative and blur) while preserving sharper boundaries. The results show that the proposed approach outperforms classical P-M technique.
展开▼