Adaptive kernel for triangular meshes smoothing

In this paper we propose a new anisotropic smoothing method that mimic the statistical noise distribution. First, we estimate the probability density function (pdf) of the noise data, then we incorporate the estimated pdf into a convolution formulate that, when expressed on mesh gives arise to an updating formulae that allows to reduce iteratively the noise. To preserve the edges and corners during the smoothing operation, we identify the covariance matrix of the pdf to the structure tensor, which allows to perform anisotropic smoothing. Experiments on noisy object with artificial noise and acquisition noise show that such this smoothing method allows to achieve a good smoothing results in comparison with existing standard smoothing techniques. The proposed method is stable, fast and easy to implement.