(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
3D-Mesh denoising using an improved vertex basedanisotropic diffusion
Mohammed El Hassouni
DESTECFLSHR, University of Mohammed V-Agdal-Rabat, MoroccoMohamed.Elhassouni@gmail.com
Driss Aboutajdine
LRIT, UA CNRSTFSR, University of Mohammed V-Agdal-Rabat, Moroccoaboutaj@fsr.ac.ma
Abstract
—This paper deals with an improvement of vertex basednonlinear diffusion for mesh denoising. This method directlyfilters the position of the vertices using Laplace, reduced centeredGaussian and Rayleigh probability density functions asdiffusivities. The use of these PDFs improves the performance of a vertex-based diffusion method which are adapted to theunderlying mesh structure. We also compare the proposedmethod to other mesh denoising methods such as Laplacian flow,mean, median, min and the adaptive MMSE filtering. Toevaluate these methods of filtering, we use two error metrics. Thefirst is based on the vertices and the second is based on thenormals. Experimental results demonstrate the effectiveness of our proposed method in comparison with the existing methods.
Keywords-
Mesh denoising, diffusion, vertex.
I.
I
NTRODUCTION
The current graphic data processing tools allow the design andthe visualization of realistic and precise 3D models. These 3Dmodels are digital representations of either the real world or animaginary world. The techniques of acquisition or design of the 3D models (modellers, scanners, sensors) generallyproduce sets of very dense data containing both geometricaland appearance attributes. The geometrical attributes describethe shape and dimensions of the object and include the datarelating to a unit of points on the surface of the modelledobject. The attributes of appearance contain information whichdescribes the appearance of the object such as colours andtextures.These 3D models can be applied in various fields such as themedical imaging, the video games, the cultural heritage... etc[1]. These 3D data are generally represented by polygonalmeshes defined by a unit of vertex and faces. The most meshesused for the representation of objects in 3D space are thetriangular surface meshes.The presence of noise in surfaces of 3D objects is a problemthat should not be ignored. The noise affecting these surfacescan be topological, therefore it would be created by algorithmsused to extract the meshes starting from groups of vertices; orgeometrical, and in this case it would be due to the errors of measurements and sampling of the data in the varioustreatments [2].To eliminate this noise, a first study was made by Taubin [3]by applying signal processing methods to surfaces of 3Dobjects. This study has encouraged many researchers todevelop extensions of image processing methods in order toapply them to 3D objects. Among these methods, there arethose based on Wiener filter [4], Laplacian flow [5] whichadjusts simultaneously the place of each vertex of mesh on thegeometrical center of its neighboring vertex, median filter [5],and Alpha-Trimming filter [6] which is similar to thenonlinear diffusion of the normals with an automatic choice of threshold.The only difference is that instead of using the nonlinearaverage, it uses the linear average and the non iterative methodbased on robust statistics and local predictive factors of firstorder of the surface to preserve the geometric structure of thedata [7].There are other approaches for denoising 3D objects such asadaptive filtering MMSE [8]. This filter depends on the form[9] which can be considered in a special case as an averagefilter [5], a min filter [9], or a filter arranged between the two.Other approaches are based on bilateral filtering byidentification of the characteristics [10], the non local average[11] and adaptive filtering by a transform in volumetricdistance for the conservation of the characteristics [12].Recently, a new method of diffusion based on the vertices [13]was proposed by Zhang and Ben Hamza. It consists in solvinga nonlinear discrete partial differential equation by entirelypreserving the geometrical structure of the data.In this article, we propose an improvement of the vertex baseddiffusion proposed by Zhang and Ben Hamza. The onlydifference is to use of different diffusivities such as thefunctions of Laplace, reduced centred Gaussian and Rayleighinstead of the function of Cauchy. To estimate these variousmethods of filtering, two error metric
L
2
[13] are used.This article is organized as follows: Section 2 presents theproblem formulation. In Section 3, we review some 3D mesh
330http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
denoising techniques; Section 4 presents the proposedapproaches; Section 5 presents the used error metrics. InSection 6, we provide experimental results to demonstrate amuch improved performance of the proposed methods in 3Dmesh smoothing. Section 7 deals with some concludingremarks.II.
P
ROBLEM FORMULATION
3D objects are usually represented as polygonal or trianglemeshes. A triangle mesh is a triple
M= (P,
ε
, T)
where
P ={P
1
, ……,P
n
}
is the set of vertices,
ε
= {e
ij
}
is the set of edgesand
T = {T
1
, ……,T
n
}
is the set of triangles. Each edgeconnects a pair of vertices
(P
i
, P
j
).
The neighbouring of avertex is the set
P
*
= {P
j
∈
P: P
i
~ P
j
}
. The degree
d
i
of avertex
P
i
is the number of the neighbours
P
j
.
N(P
i
)
is the set of the neighbouring vertices of
P
i
. N(T
i
)
is the set of theneighbouring triangles of
T
i
.We denote by
A(T
i
)
and
n(T
i
)
the area and the unit normal of
T
i
, respectively. The normal
n
at a vertex
P
i
is obtained byaveraging the normals of its neighbouring triangles and isgiven by
( )
( )
∑
∈
=
*
5
1
i j
PT T jii
T nd n
(1)The mean edge length
l
of the mesh is given by
∑
∈
=
ε
ε
ij
eij
el
1
(2)During acquisition of a 3D model, the measurements areperturbed by an additive noise:
η
+=
'
PP
(3)Where the vertex
P
includes the original vertex
P
’
and therandom noise process
η
. This noise is generally considered asa Gaussian additive noise.For that, several methods of filtering of the meshes wereproposed to filter and decrease the noise contaminating the 3Dmodels.III.
RELATED WORK
In this section, we present the methods based on the normalssuch as the mean, the median, the min and the adaptiveMMSE filters and the methods based on the vertices such asthe laplacien flow and the vertex-based diffusion using thefunctions of Cauchy, Laplace, Gaussian and Rayleigh.
A.
Normal-based methods
Consider an oriented triangle mesh. Let
T
and
U
i
be a meshtriangles,
n(T)
and
n(U
i
)
be the unit normal of
T
and
U
i
respectively,
A(T)
be the area of
T
, and
C(T)
be the centroid of
T
. Denote by
N(T)
the set of all mesh triangles that have acommon edge or vertex with
T
(see Fig. 1)
T
Fig. 1 Left: Triangular mesh.Right: updating mesh vertex position.
1)
Mean Filter:
The mesh mean filtering scheme includesthree steps [5]:
Step 1.
For each mesh triangle
T
, compute the averagednormal
m(T)
:
( )( )( ) ( )
( )
∑∑
∈
=
T N U iii
i
U nU A
U AT m
1
(4)
Step 2. N
ormalize the averaged normal
m(T)
:
( )( )( )
T mT mT m
←
(5)
Step 3.
Update each vertex in the mesh:
( )( ) ( )
∑∑
+←
T vT A
T APP
old new
1
(6)With
( ) ( ) ( )
T mT mC PT v
..
=
(7)
( )
T v
is the projection of the vector
..
C P
onto the directionof
m(T)
, as shown by the right image of Fig. 1.
2)
Min filter :
The process of min filtering differs from theaverage filtering only at step1. Instead of making the averageof the normals, we determine the narrowest normal,
η
i, foreach face, by using the following steps [9]:-
Compute of angle
Φ
between
n(T)
and
n(U
i
).
-
Research of the minimal angle: If
Φ
is the minimalangle in
N(T)
then
n(T)
is replaced by
n(U
i
).
331http://sites.google.com/site/ijcsis/ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 2, 2010
3)
Angle Median Filter:
This method is similar to minfiltering; the only difference is that instead of seeking thenarrowest normal we determine the median normal byapplying the angle median filter [5]:
( ) ( )( )
ii
U nT n
,
∠=
θ
(8)If
θ
i
is the median angle in
N(T)
then
n(T)
is replaced by
n(U
i
).
4)
Adaptive MMSE Filter:
This filter differs from theaverage filter only at step1. The new normal m(T) for eachtriangle T is calculated by [8]:
( )( )( ) ( )
≠≤+
−=>=
010
2222222222
ljljnlj
ljn jljnljljnlj
j
et T M T n
ouT M T m
σ σ σ σ σ σ σ σ σ σ
(9)
( )( ) ( )( )
∑∑
−=−=
=
1010
N ii N ii jilj
U AU nU A
T M
(10)
__ _
σ
n2
is the variance of additive noise and
σ
lj2
is the variance of neighbouring mesh normals which is changed according toelements of normal vector. Thus,
σ
lj2
is calculated as follows:
( ) ( )( )( )
T M U AU nU A
lj N ii N ii jilj
2101022
−=
∑∑
−=−=
σ
(11)
B.
Vertex-based methods1)
Laplacian Flow
Fig 2. Updating vertex position by umbrella operator.Considering the following expression which allows the updateof the mesh vertices [5]
( )
old old new
P DPP
λ
+←
(12)Where
D(P)
is a displacement vector, and
λ
is a step-sizeparameter.The Laplacian smoothing flow is obtained if the displacementvector
D(P)
is defined by the so-called umbrella operator [14](see Fig. 2) :
( )
( )
∑
∈
−=
P N ji ji
PPnPU
1
(13)
N(P)
is the 1-ring of mesh vertices neighbouring on
P
i
P
2)
Vertex-Based Diffusion using the Function of Cauchy:
This method [13] consists in updating the mesh vertices bysolving a nonlinear discrete partial differential equation usingthe function of Cauchy.Fig 3. Illustration of two neighbouring rings.The update of the vertices of mesh (see Fig. 3) is given by
( )
( )( )
jiPPii j jiii
PgPg
d Pd Pd PP
i j
∇+∇
−+←
∑
∈
*
1
(14)Where
g
is Cauchy weight function given by
( )
22
11
c x xg
+=
(15)
and
c
is a constant tuning parameter that needs to be estimated.
332http://sites.google.com/site/ijcsis/ISSN 1947-5500
Search History:
Result 00 of 00
00 results for result for
p.
3D-Mesh denoising using an improved vertex based anisotropic diffusion
This paper deals with an improvement of vertex based nonlinear diffusion for mesh denoising. This method directly filters the position of the vertices using Laplace, reduced centered Gaussian and Rayleigh probability density functions as diffusivities. The…
(More)
911
Reads
2
Readcasts
1
Embed Views
More From This User
Notes
Load more
