Texture preserving Perona-Malik model

Abstract

The Perona-Malik (PM) model, an effective anisotropic diffusion, can preserve edges while removing the noise. However, the disadvantage of the PM model is tending to impair textures and details so that denoising is not sufficient in the whole process. For this reason, we present a novel texture preserving Perona-Malik (TPPM) models based on the local directional variance. In the TPPM model, the diffusion coefficient of the PM model is adaptively determined to be low diffusion in large variance domain and be high diffusion in low variance domain. The related parameters are studied. Comparative results on real image denoising demonstrate that our model outforms the PM model, classical total variational (TV) method, a wavelet-based method and a nonlocal means filter in signal-to-noise ratio. The proposed model is also competitive with other methods visually. Furthermore, the execution times are very fast.