Normal Domain Mesh Modeling Using Non-Linear Guided Filtering

It is a basic research problem to model 3D mesh in the field of digital geometry processing. To improve the simplicity and robustness of mesh modeling, we first propose a non-linear guided filtering method, which is performed in normal domain. The filtered normal is a quadratic transformation of guidance mesh’s normal. Then, the above method is applied to deal with mesh denoising and mesh smoothing. The main problem is the construction of guidance mesh. For mesh denoising, we get guidance mesh by bilateral normal filtering at each iteration. For mesh smoothing, we use the result of Gaussian filtering as the initial guidance mesh, and then iteratively update it with the original mesh. Finally, we perform denoising and smoothing on mesh models with complex shape or rich features. Experimental results show that the proposed algorithm is simple, effective, and robust. Specifically, the denoising method can effectively remove strong noises and retain geometric features, and the smoothing method can extract features of small and medium scales and maintain features of large scales.