Reweighted Global Bilateral Filtering Based on Normal Regularization

One of the problems existed in mesh denoising is the facial normal's sensitivity to noise especially when the level of noise is high.It's difficult to give proper variance parameters of the two Gaussian functions in bilateral filtering, which can distinguish the change of normals around noise and feature adaptively.This will lead to the residual of noises or the damage of the structures'features.To deal with the problem, we propose an improved global bilateral filtering method based on face normals.Firstly, we enhance the difference of normal change around noise and feature by regularizing the original noisy face normal.And then we use the regularized face normals to compute the bilateral weights.This can effectively overcome the problem of parameter choosing.We also use the idea of reweighting to punish noise more seriously, and this helps improve the global denoising method a lot. We demonstrate that our method produces better denoising result than the present method both numerically and visually.