Proc. of the Intl. Conf.

on Advances In Engineering And Technology - ICAET-2014

Copyright Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-01-113

3D Reconstruction of Face: A Comparison of

Marching Cube and Improved Marching Cube
Algorithm
T. Senthil Kumar, Anupa Vijai

Abstract3D reconstruction of face is one of the

advancements in physical modeling techniques which uses
II. Related Works
engineering methods in the field of medicine.The systems
The well-known method in iso-surface rendering is classic
in development propose a software tool that will help in
Marching Cubes (MC) algorithm that was first published by
craniofacial surgery. The existing approaches for 3D Lorensen and Cline in 1987[2].
reconstruction has different applications from real scenary
to human parts of body. The analysis of the different A. Marching Cube Algorithm
algorithms allow developers to make vital decisions in
The input volume data consist of samples organized into a
understanding the modelling of the face. The human face regular 3D Cartesian grid. From such a grid, it is easy to
has different regions including the tissue and hard bones. obtain a set of cells. The cell in this case has a cube shape and
The paper presents a comparison of two surface rendering consists of eight corresponding samples from two adjacent
techniques, Marching Cube(MC) and Improved Marching sample planes. Four samples are from the first plane and four
Cube(IMC) algorithms, and draws conclusions for samples are from the second plane. MC method processes
analysing the suitable approach for a specific range of sequentially all the cells that can be found in volume data. The
application. iso-surface is specified by a threshold value.

Keywords MC, IMC, CT, DICOM, Voxel Applying the marching cubes algorithm to reconstruct the
surface, the volume data needs to be partitioned into cubes.
The algorithm determines how the surface intersects the cube,
I. Introduction then marches to the next cube. To find the surface intersection
Surface rendering is a method to generate surfaces from a set in a cube, we assign one to a cubes vertices if their data value
of data points. It is a way to visualize the object by means of exceeds or equals to the value of the surface we are
the image data as a set of certain basic elements, such as reconstructing, and consider them as inside vertices. Cube
voxels, their faces, other polygons, line segments, and points vertices with values below the surface receive a zero and are
which represents the boundary of the structure. An iso-surface considered as outside vertices. Therefore a logical cube is
is a three-dimensional surface that represents points of a obtained to configure the surface. Since there are eight vertices
constant value within a volume of space or it can be a level set in each cube, there are only 28 = 256 ways a surface can
of a continuous function whose domain is the 3D-space. intersect the cube. Recurred to the reverse and symmetric
properties of the cube, we can reduce those 256 cases to 15
In medical imaging, iso-surfaces may be used to represent patterns. Each cell face is shared by another cell. Due to such
regions of a particular density in a three-dimensional sharing, the iso-surface is continuous among adjacent cells.
Computed Tomography(CT) scan, allowing the visualization There are few approaches how to avoid the holes. Ambiguous
of internal organs, bones, or other structures. One popular cases can be detected and a special triangulation can be
method of constructing an iso-surface from a data volume like applied. The algorithm complexity of MC method is O(N),
CT scan of human face is iso-surface extraction algorithms where N is the number of all cells. The algorithm[2] provides a
like marching cubes or marching tetrahedral. 3D surface reconstruction by giving physicians a 3D view of
anatomy. It creates triangle models of constant density
surfaces from 3D medical data. The algorithm processes the
3D medical data in scan-line order and calculates triangle
T.Senthil Kumar, Anupa Vijai vertices using linear interpolation.
Amrita Vishwa Vidyapeetham
Coimbatore, India

529
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-01-113

B. Method Based on Modularization

2) Standard MC algorithm generates a huge quantities of
A 3D surface reconstruction method of medical image based triangle patches, which slows the reconstruction down greatly.
on modularization has been proposed by Bin Lee, Lian-Fang For example, a set of 128 128 256(8bit) skull CT data,
Tian, Chen Ping, Hongqiang Mo and Zong-Yuan [3] .This after being processed using MC algorithm, will become more
divides the whole process of three dimensional reconstruction than 500,000 triangle patches.
into different modules and parts based on functions.
3) Standard MC algorithm can probably generate ambiguous
The whole process of 3D reconstruction is divided into surface in certain situation.
different steps. First an appropriate segmentation method is
used to given slices to get relevant information from given This means sometimes the topological structure may be
slices of CT images. The output of segmentation is used as the wrong. If one voxels plane has 4 intersection points made by
input of classical MC algorithm which will generate proper the arrises and the iso-surface, there will be 2 probable
iso-surface which is required. Based on the features of the situation, and this can probably lead to a wrong topological
voxels, a novel voxel-tracing algorithm is developed. In structure. If the adjacent voxels which share the same plane
normal method the whole voxels created will be searched one chose different situation, there will be a hole in the topological
after the other On the other hand, the voxels that actually structure.To overcome the above drawbacks of the MC
intersect with the iso-surface are only a small part. So in order algorithm, as well as to deal with the special nature of medical
to enhance searching efficiency, a favorable data structure is images, several approaches are adopted, such as adding a filter
needed this will only make the process to search lesser number module and a image segmentation module, changing the single
of voxels. loop traversing method into small quantities parallel traversing
method, adding a mesh simplifying module to reduce the
A large amount of triangle meshes will be generated using MC number of iso-surface patches, and avoiding producing the
algorithm. So, measures should be taken to merge some wrong surface by use of the asymptote theory.
triangle meshes before applying rendering to them. Here a
mesh optimization method proposed is applied. Here two Another approach is applying Improved Marching Cube
vertex of a triangle mesh are merged according with the two Algorithm(IMC)[14] for facial reconstruction. Based on the
following constraints: normal constraint and space constraint. drawbacks of traditional MC algorithm, the following three
The condition for merging two vertexes is that the two above improvements are made.
constraints are met, and then the triangle mesh is deleted.The
algorithmic modularization for 3D reconstruction will improve 1. Replace cube edge linear interpolation with midpoint
the execution time for rendering. The voxel can be traced selection.
more faster than usual algorithm which will reduce the error
surfaces and holes in the surface rendering. In the case of increasingly high-resolution of medical images,
the thickness of CT slices has become increasingly small. So,
taking the midpoint of the cube edge as the intersection point
C. Improved MC Algorithm instead of calculating it using linear interpolation almost make
An improvement is suggested to basic iso-surface no differences in reconstruction
reconstruction by Jun Xiao, Miao Yu, Ningyu Jia by adding a results. At the same time, the triangular patches generated in
filter module and segmentation module in 3D reconstruction this way could form a smoother isosurface in local area, which
algorithm. This method proposed discusses the basic will benefit mesh simplifying.
principle of MC algorithm and both the advantages and
disadvantages of it. The MC algorithm is improved, and a 2. Index of the intersection point is used to avoid
filter module, a image segmentation module and a mesh repeated calculation.
simplifying module are added. Marching cubes algorithm is a
widely used surface rendering method. But MC algorithm also In order to avoid the repeated calculation between the adjacent
has its disadvantages, such as ambiguous surface and the huge cubes, create a three-dimensional array to store the pointer
quantities of the triangle patches generated. This method which points the information of the intersection point of the
improves the MC algorithm by changing traversing method cube edge and isosurface. Calculate normal of the intersection
into multi-line traversing method to take advantage of multi- point, query the array using the coordinate of the intersection.
pipeline technology. And this method also resolves the If the current value is NULL, it indicates that the information
ambiguity by use of asymptotic decider. of the intersection point has not been calculated. Repeat the
calculation, and after that store the pointer of the vertex
Although MC algorithm is simple in principle, and it is easy to information in the array Otherwise, the information of the
implement, but it has following drawbacks: intersection has been calculated, and use the pointer in the
array directly.
1) Standard MC algorithm traverses all the voxels in a single
loop way. It is dull and inefficient.

530
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-01-113
3. Contract edge to reduce the number of triangular
patches.
Then calculate the normals to each slices using the
equation
Simplifying the grid method can be divided into four
categories: sampling, adaptive subdivision, decimation, vertex V=V1+f(V1)(V1-V2)/f(V2)-f(V1). (2)
merging. As the simplicity and robustness of vertex merging,
in this paper, the edge contraction method is selected. When Then by determining the surface values for different tissue
two vertices satisfy the following three constraint conditions, layers we can construct the face. The values of different layers
the two vertices will merge into one, and the edge consisting ranges from 0-255.
of the two vertices will contract into a point. If the conditions
are not met, the triangle will be kept. A. 3D Model of a Face
3D Face is created from 415 axial slices of CT scan images of
III. Comparison and Analysis head of a patient by applying the above steps.
The visualization of such data may be done in two-
dimensional manner, where each slide can be viewed
separately on the screen. Often this will require a bit of
processing, because computers typically use 8 bits per color
channel to represent an image - 8 bits for red, 8 bits for green
and 8 bits for blue color. If CT slice is more than 8 bits
resolution, the programmer needs to scale down CTs values
to make them fit into 8 bits resolution range. The Digital
Imaging and Communications in Medicine(DICOM) will
contain all the details of the patient taken during the CT scan.
This information might be sensitive if exposed to public for
medical and non medical applications. DICOM is a binary
protocol and data format. DICOM has many versions of the
standard, with support for 8 and 16 bit images.

Figure 2: 3D Model of a face

B. Observations
The dataset is a series of 415 axial CT slices of size
512*512 of the head of a patient with 3mm thickness
between consecutive slices. The surface rendering
algorithms for 3D reconstruction, MC and IMC with
transformation is compared based on time and number
of vertices generated. From the result it is clear that
IMC with transformation takes less time for the
reconstruction process without compromising the
Figure 1: Image viewer created to view the dataset
accuracy.

Perform translation on each slide. Then do a resampling

operation on each slide by applying the equation

F ( i 1, j , k 1 ) F ( i , j , k )
(1)
k

531
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-01-113

[9] Philippe Lacroute, Marc Levoy, Fast Volume Rendering

Using a Shear-Warp Factorization of the Viewing
TABLE I. COMPARISON OF MC AND IMC WITH TRANSFORMATION Transformation, Computer Graphics Proceedings, Annual
Conference, July 1994.
Features
Algorithms [10] Jurgen P. Schulze , Ulrich Lang, The Parallelized
Time(Seconds)
Vertex(number) Perspective Shear-Warp Algorithm For Volume Rendering,
MC 7.72 13148 Fourth Eurographics Workshop on Computer Graphics and
IMC(with Visualization,2002.
3.45 5569
transformation)
[11] Kamil Toszek, Support System For Skull Trepanation,
Based On The Space Positioning Devices, Politechnika
Acknowledgment Lodzka, Instytut Informatyki ,number 120384,2008.
We would like to thank Dr. Sherry peter of Amrita Institute of
Medical Sciences, Kochi, for providing the dataset for this [12] Bin Lee, Lian-Fang Tian, Chen Ping, Hong-Qiang Mo,
work. Zong-Yuan Mao, A Fast Accurate 3d Surface
Reconstruction Method Of Medical Image Based On
Modularization, Proceedings of the Fourth International
References Conference on Machine Learning and Cybernetics, August
2005.
[1] H.Fuchs, Z.M. Kedem, and S.P. Uselton, Optimal
Surface Reconstruction from Planar Contours, [13] Zhusheng Xu, Chuangbai Xiao, Xiao Xu, An Improved
Communications on ACM,volume 20,number 10,October Marching Cubes Algorithm Based on Edge Contraction,
1977. Proceedings of International Conference on Signal
Processing, 24-28 Oct. 2010 ,pp. 944 - 947, Beijing, China.
[2] William E. Lorensen , Harvey E. Cline, Marching
Cubes: A High Resolution 3d Surface Construction [14] J. P Schulz, M. Kraus, U. Lang, T. Ertl, Integrating pre-
Algorithm, Computer Graphics, Volume 21, Number 4,July integration into the shear-warp algorithm, VG '03
1987. Proceedings of the 2003 Eurographics/IEEE TVCG Workshop
on Volume graphics ,Pages 109 - 118, Newyork, USA.
[3] Choi, Jae- Jeong, Efficient multidimensional volume
rendering, Proceedings on SPIE Vol. 3658, p. 9-16, Medical
Imaging 1999: Image Display, Seong K. Mun; Yongmin Kim;
Eds.

[4] Jun Xiao, Miao Yu, Ningyu Jia, An Improved MC

Algorithm Applied in Medical ImageReconstruction, Control
and Decision Conference, 2008 July.

[5] Un-Hong Wong ,Hon-Cheng Wong,Zesheng Tang, An

Interactive System for Visualizing 3D Human Organ Models,
Ninth International Conference on Computer Aided Design
and Computer Graphics,(CAD/CG) 2005.

[6] Robert A. Drebin, Loren Carpenter, Pat Hanrahan,Volume
Rendering, SIGGRAPH '88, Atlanta, 1988 August.

[7] Marc Levoy, Display of Surfaces from Volume Data,

IEEE Computer Graphics and Applications,volume 8, Number
3, May 1988.

[8] Yun Tian, Ming-Quan Zhou, Zhong-Ke Wu, A

Rendering Algorithm Based On Ray-Casting For Medical
Images, Proceedings of the Seventh International Conference
on Machine Learning and Cybernetics, Kunming,July 2008.

532