International Journal of Multimedia and Ubiquitous Engineering
Vol. 9, No. 11 (2014), pp. 11-20
http://dx.doi.org/10.14257/ijmue.2014.9.11.02
Copyright ⓒ 2014 SERSC 11
Feature-preserving, Adaptive and Anisotropic Smoothing Algorithm
for Triangular Mesh Models
Liu Xu-min
1
, Yang Li-xin
2
and Li Cai-ling
1
1
College of Information Engineering, Capital Normal University, Beijing 100048,
China
2
College of Computer, Beijing University of Post&Telecommunication, Beijing
100876, China
liuxumin@126.com
Abstract
Despite the great success in smoothing of triangular mesh, the classical methods mostly
require human interactions. In order to reduce the parameter settings, an adaptive and
anisotropic smoothing algorithm for removing noise and preserving features of triangular
mesh model was proposed. First calculate the expected normal vector of the model triangles,
the expected normal vector should be an accurate representation of the desired movement
direction and be similar to the original surface normal vector, and then we compute the value
of the offset value. We develop an adaptive coefficient scheme which can avoid the parameter
settings to obtain the coefficient value for each vertex. Finally, we updated every vertex’s
position of the model by formula. Experimental results show that the algorithm can be
adaptive to preserve sharp features and avoid shrinkages by comparing with classical
methods.
Keywords: smoothing; anisotropic; adaptive; triangular mesh; feature-preserving
1. Introduction
With the development of the theory and technology of computer-aided geometric design,
reverse engineering has become an important approach to represent three-dimensional model.
The mock-ups are usually obtained by 3D scanner, the point cloud are collected from the
three-dimensional data measurement tools, while on account of human factors and other
reasons, often lead to data noise , data redundancy and other issues. The noises should be
removed before 3D data processing. Mesh smoothing is to eliminate the noise, to maintain its
topology and geometry, to avoid oversmoothing and volume shrinkage.
Existing triangular mesh smoothing algorithms have two main categories: local iterative
method and global optimization method. Field [1] proposed a classical Laplace algorithm that
mesh vertices move towards its neighboring's center direction. The algorithm is simple and
fast, yet the model is prone to be shrinking and oversmoothing with the increasing number of
iterations. Taubin [2] applied the knowledge of signal processing to remove noise, and
presented a weighted Laplacian algorithm with two alternative scale factors of opposite signs.
Such smoothing can suppress high frequency and enhance low frequency slightly without
volume shrinkage. Mingqiang Wei et al., [3] present a feature-preserving optimization for
noisy mesh algorithm by using joint bilateral filter and constrained Laplacian smoothing, yet
a large computation is needed when the mesh is very large. Desbrun et al., [4] proposed the
mean curvature flow algorithm, each mesh vertex moves towards its normal vector direction
with the speed of mean curvature. While in some special cases, the grid is easy to cause