APPLICATIONS OF 3D SHAPE ANALYSIS AND RETRIEVAL

Afzal Godil
National Institute of Standards and Technology, Gaithersburg, U.S.A.
godil@nist.gov
ABSTRACT
With recent advances in 3D modeling and scanning
technologies, large number of 3D models are created and stored in
different databases. This has created an impetus to develop
effective 3D shape analysis and 3D shape retrieval algorithms for
these domains. This has made the field of 3D shape analysis and
retrieval become an active area of research in the 3D community.
In this paper we will survey few applications where 3D shape
analysis and retrieval has been applied effectively. The main
applications we have discussed are: 3D human shape analysis;
CAD/CAM applications; structural bioinformatics; and other
applications.
Index Terms --- 3D Shape Analysis, 3D shape retrieval, 3D
applications
1. INTRODUCTION
With advances in 3D modeling and scanning technologies, large
numbers of 3D models are created and store in different databases.
For these databases to be most useful for scientific and design
applications require methods for storage, indexing, analysis,
searching, recognition, clustering and retrieval of the content under
study are essential. 3D shape analysis and retrieval has applications
in various disciples and widely used in computer graphics and
visualization, archeology, bio-imaging and medical imaging
applications, defense and security, industrial design, and games,
entertainment and film industry, and so on. This has created an
impetus to develop effective 3D shape analysis and retrieval
algorithms for these domains. This has made the field of 3D shape
analysis and retrieval has become an active area of research in the
3D community. With the increasing number of 3D models created
and available on the Internet, many domains have their own 3D
repositories such as the National Design Repository for CAD
models [3], Protein Data Bank for structural data of biological
macromolecules [2], CAESAR for Anthropometry [1] and so on.
Laser scanning of cultural heritage sites have created digital
archives like the Digital Michelangelo Project [4], and the Stanford
Digital Formae Urbis Romae Project [5].
One of the main areas of research in 3D Shape analysis and
retrieval fields is the shape descriptor (feature descriptor). The
shape descriptors can be classified into four main categories: view-
based, graph-based, statistics-based and transformed-based, or as
local and global shape descriptors. Recent work shows that view-
based approaches perform better than other approaches for 3D
shape retrieval. The view-based approaches can also be used for
query searching interface, based on depth images, binary images
and 2D sketches.
One of the main drawback of view-based methods is that there an
infinite number of possible view. One very effective solution is to
apply pose normalization to the object via Principal Component
Analysis (PCA) then uniformly sample the view direction on a unit
sphere.
In this paper we will survey few applications where 3D shape
analysis and retrieval has been applied effectively. The main
applications are: 3D human shape; CAD/CAM applications;
structural bioinformatics; and other applications
A typical 3D shape retrieval task can be defined as: given a
3D shape query, to obtain a list of 3D shapes ordered by the
similarity between the query object and the one on the list. Several
methods are proposed to solve the problem, such as Light Field
descriptors [7], spherical harmonics descriptor [11], D2 shape
distribution [14], and local feature based methods [12, 15] among
others.
2. RELATED WORK
In this section we will only describe some of the more relevant
methods with respect to our work. For more information about
different 3D shape retrieval methods we refer the reader to the
survey paper of Tangelder et al. [8] and for review of different
application the paper of Bustos et. al. [9].
In [7] and [13], the authors developed a descriptor based on the
silhouettes from multiple directions for visual similarity based
comparison. Initial methods proposed for 3D object searching were
based on the histograms [14]. Although these methods are very
simple and fast to compute, the results were not that promising. In
[18] Laga et al, discussed the generation of a descriptor for 3D
shape comparison using spherical wavelet transforms. Recently in
[28] some work has been done to define shape features using
Krawtchouk moments. In [6] Shilane et al discussed the creation of
the Princeton Shape Benchmark, and analyzed some of the
contemporary algorithms for 3D shape retrieval. In [27] authors
discussed methods to compare images based on the Zernike
moments and compared the performance of Zernike moments with
other type of image similarity measures.
Previous work on human body retrieval based on body shape was
performed by Paquet and Rioux [19]. They performed content
based anthropometric data mining of three dimensional scanned
human by representing them with compact support feature vectors.
They showed a virtual environment to perform visual data mining
on the clusters and to characterize the population by defining
archetypes. Paquet [20] introduced cluster analysis as a method to
explore 3D body scans together with the relational anthropometric
data as contained in the CAESAR anthropometric database. Azouz
et al. [21, 22] analyzed human shape variability using a volumetric
representation of 3D human bodies and applied a principal
components analysis (PCA) to the volumetric data to extract
dominant components of shape variability for a target population.
Through visualization, they also showed the main modes of human
shape variation. The work of Allen [26] demonstrated a system of
synthesizing 3D human body shapes, according to user specified
parameters; They used 250 CAESAR body scans for training.
Retrieval based on head shape was performed by Ip and Wong [23].
Their similarity measure was based on Extended Gaussian Images
of the polygon normal. Godil et al. in [24] used body shape for
human identification and in [25] for 3D face recognition.
2. HUMAN BODY AND HEAD SHAPE
The understanding of shape and size of the human body, head and
faces is vital for design and important in many applications, such
as clothing design, machine design, transportation sector,
medical/healthcare field, aircraft cockpit design, space suit design
for astronauts, safety, biometric, criminology, interface design for
household/industrial products, and for facial wear products, such as
respirators, helmets, eyeglasses. The human shape and size are also
some of the basic tools used for analysis and design requirements
by human factors, ergonomics professionals, architects, interior
designer and industrial engineers. With the emergence of 3D laser
scanners, there have been large scale surveys of humans around the
world, such as the CAESAR anthropometric database. The 3D
scans of human bodies contain over hundreds of thousand grid
points. To be used effectively for design, analysis, indexing,
searching, clustering and retrieval, these human bodies requires a
compact shape representation. We have developed two such
compact representations based on the human body and two
compact representations for the human head shape by applying
Principal Component Analysis on the facial surface and in the
second method the whole head is transformed to a spherical
coordinate system expanded in a basis of Spherical Harmonics. We
have then used these shape descriptors for 3D shape retrieval from
the CAESAR database [29] and human identification from body
shape in [24] and for 3D face recognition in [25].
Figure 1. Similarity based retrieval for CAESAR body 00014
based on PCA facial shape
Then we use cluster analysis on these head/facial shape descriptors
along with the number of cluster to come up with the sizing system
for different product; such as, facial respirators, eyeglasses,
helmets, and so on. Cluster analysis is a technique for extracting
implicit relationship or patterns by grouping related shape
descriptors. A cluster is a collection of objects that are similar to
one another and are dissimilar to the objects in other clusters.
There are a number of clustering techniques, but we have only tried
k-mean and k-median techniques. In the paper on Facial Shape
Analysis and Sizing System [30], we have shown how to come up
with a sizing system for facial respirators using 3D facial data.
Facial respirators are used by millions of people around the world
to reduce their risk to diseases, harmful and hazardous airborne
agents. At the heart of the effectiveness is the seal of the respirator
which mainly depends on the fit, which prevents harmful gases and
particulates from enter into the wearers respiratory system.
The Los Alamos National Laboratory (LANL) fit test panel
developed in the 1970s is based on an anthropometric survey.
Although the fit test panel is in the process of being updated, the
core of these fit tests is still traditional anthropometric measures,
which simplify the complexity of the shape of the human face.
Today most manufactures supply half and full facial mask
respirators based on the facial grouping that are based on the above
surveys. However, many researchers have shown that there are
little or no correlations between facial dimensions and the fit of
half mask respirator by Yang et al. [31]. Hence the best respirator
shape for the best seal fits can only be archived by using the full
3D facial data.
3. CAD/CAM PARTS
Large number of 3D CAD models are created by engineering and
manufacturing companies and stored in local repositories. The 3D
CAD models provide the high fidelity description as well as the
corporate memory of the engineering part. Searching,
classification, clustering and analysis for parts across the
manufacturers supply chain efficiently are a very important issue.
According to AutoDesk there are over 20 Billion CAD models,
compared to 6 Billion people. Different estimates by experts put
the number of unique designs of parts at around 60 to 800 Billion.
Even a single Boeing Aircraft 787 has more than 3 million unique
parts from different part suppliers. Using 3D shape searching early
in the design cycle can detect duplicate parts and can locate similar
parts manufactured across your supply chain. Hence, there will be
cost saving associated with parts reuse by identifying and reuse of
the existing designs and manufacturing processes. Shape similarity
for 3D parts can also recommend the optimum manufacturing
process and also provide the cost estimation of manufacturing of
parts.
While there have been advances in the retrieval of information
from text data, these methods simply cant be extended to 3D
objects. Search of 3D object models require surface-based or
volume-based features to effectively characterize the shape,
semantics, content and topology. We have developed different
types of shape descriptors for 3D shape retrieval from a CAD
database. In Figure 3 and Figure 4, we show the retrieval results
visually from the Purdue Engineering Benchmark [10] for a
Bearing block and a screw like part.
Figure 3. Bearing Block searching results from Purdue Engineering
Benchmark [4]
Figure 4. Screw like part searching results from Purdue
Engineering Benchmark [10]
5. STRUCTURAL BIOINFORMATICS
It is widely believed that the 3D shapes of macro molecules and
their active sites provide a discriminating role in bio-molecular
recognition and function. Geometrical shapes determine their
ability to bind to their targets. Characterization of geometrical
shape may thus provide information to classify and retrieve related
and functionally relevant macromolecules for purposes such as
drug targeting. There are over 46,000 protein structures in the
Protein Data Bank (PDB). These 3D structural databases pose
challenges for storing, indexing, searching, clustering, retrieval of
shape based structural information. Techniques used in text based
retrieval of structural information may not be easily extended to
shape based 3D or 2D searches that require surface-based and
volume-based descriptors to effectively characterize the shape,
semantics and geometric topology. Hence there is a need for an
automated rule-based 3D retrieval and classification system to
efficiently manage Structural Bioinformatics Databases. Some of
these requirements may be met by the existing technology used in
3D shape searching for instance in Computer Aided Design (CAD)
as illustrated in figure 5, the Web interface shows the visual query
results for a images. We have developed a shape based retrieval
and classification method for a few of the structures taken from the
PDB. The method involves developing 3D shape descriptors to
describe the 3D shape of each structure. The shape descriptors that
we have developed are based on the surface of the molecules. Then
depth images are rendered uniformly on a unit circle and different
types of features are calculated. We have used this method to
develop proximity measures for structures that may be used for
assessing their similarity.
Figure 5. Shape Searching results for the Structural Bioinformatics
6. OTHER APPLICATIONS
3D Shape analysis and 3D shape retrieval is used in many other
application areas, such as:
x Defense and security applications such as for target
recognition and face recognition
x Archeology for example, to find similar objects or
missing parts or for classification and clustering of
objects
x Architecture for example, to identify objects that
spatially fit into a specific space or how to fit a set of
objects in a fixed space
x Medical imaging to understand shape changes related to
illness/diseases and for aid in surgical planning
x In games, movies, animation and virtual environments or
in a 3D model database to identify similar models or for
finding copyright issues
We also have applied 3D shape retrieval to the field of
Generic 3D database. In Figure 6 and Figure 7, shows visual results
for the human and fighter plane as a query object. Only the top 36
results are listed here on the Princeton Shape Benchmark [5].
Figure 6. Retrieval results for human from the Princeton Shape
Benchmark [6]
Figure 7. Retrieval results for fighter plane from the Princeton
Shape Benchmark [6]
7. CONCLUSION AND DISCUSSION
In this paper we have survey few applications where 3D shape
analysis and retrieval has been applied effectively. The main
applications we have discussed are: 3D human shape analysis;
CAD/CAM applications; structural bioinformatics; and other
applications. We have presented results on the 3D human shape
analysis, 3D CAD/CAM shape applications and retrieval, structural
bioinformatics application and other applications.
REFERENCES
[1] .http://store.sae.org/ CAESAR /
[2] http://www.rcsb.org/pdb/home
[3] http://www.designrepository.org/
[4] http://www.graphics.stanford.edu/data/mich
[5] http://www.formaurbis.stanford.edu/index.html
[6] Princeton Shape Benchmark. Shape Modeling Applications,
2004. Proceedings (2004), 167178
[7] D. Y. Chen, M. Ouhyoung, X.P. Tian, Y.T. Shen. On visual
similarity based 3D model retrieval. Eurographics 03, pp.
223-232
[8] Tangelder, J.W.H., Veltkamp, R.C.: A survey of content
based 3d shape retrieval methods. In: Proceedings of the
Shape Modeling International, pp. 145156 (2004)
[9] Bustos, B., Keim, A.D., Saupe, D., Schreck, T., Vranic, V.D.:
Feature-based similarity search in 3d object databases. ACM
Computing Surveys 37(4), 345387
[10] S. Jayanti, Y. Kalyanaraman, N. Iyer, K. Ramani. Developing
an Engineering Shape Benchmark for CAD Models.
Computer-Aided Design, Vol. 38, No. 9, 2006, pp. 939-953.
[11] M. Kazhdan, T. Funkhouser, S. Rusinkiewicz. Rotation
Invariant Spherical Harmonic Representation of 3D Shape
Descriptors. Symposium on Geometry Processing. June 2003.
[12] X. Li, A. Godil, A. Wagan. 3D Part Identification Based on
Local Shape Descriptors. Permis 2008.
[13] D. Vranic, 3D model retrieval, Ph.D. dissertation,
University of Leipzig, Germany, 2004.
[14] R. Osada, T. Funkhouser, B. Chazelle, D. Dobkin. Shape
Distributions. ACM Transactions on Graphics, 21(4):807-832,
2002
[15] R. Ohbuchi, K. Osada, T. Furuya, T. Banno. Salient Local
visual features for shape-based 3D model retrieval, Proc.
IEEE Shape Modeling International 2008, pp. 93-102
[16] R. Ohbuchi, K. Osada, T. Furuya, T. Banno. Salient Local
visual features for shape-based 3D model retrieval, Proc.
IEEE Shape Modeling International 2008, pp. 93-102
[17] OHBUCHI R., FURUYA T.: Accelerating Bag-of-Features
SIFT algorithm for 3D model retrieval.
[18] H. Laga, H. Takahashi, and M. Nakajima, Spherical Wavelet
Descriptors for Content-based 3D Model Retrieval,
Proceedings of the IEEE International Conference on Shape
Modeling and Applications 2006 (SMI06)-Volume 00, 2006.
[19] Paquet, E., Rioux, M. 2003 "Anthropometric Visual Data
Mining: A Content Based Approach," Submitted to IEA, 2003
International Ergonomics Association XVth Triennial
Congress. Seoul, Korea. NRC 44977.
[20] Paquet, E., "Exploring Anthropometric Data ThroughCluster
Analysis," Digital Human Modeling for Design and
Engineering (DHM). Oakland University, Rochester,
Michigan, USA. June 1517, 2004. NRC 46564
[21] Ben Azouz, Z., Rioux, M., Lepage, R. "3D Description of the
Human Body Shape: Application of Karhunen Love
Expansion to the CAESAR Database," Proceedings of the
16th International Congress Exhibition of Computer Assisted
Radiology Surgery. Paris, France. June 2629, 2002
[22] Ben Azouz, Z., Rioux, M., Shu, C., Lepage, R., Analysis of
Human Shape Variation using Volumetric Techniques, The
17
th
Annual Conference on Computer Animation and Social
Agents (CASA2004). Geneva, Switzerland. July 79, 2004.
[23] Ip, H. H. S. and Wong W. 2002 3D Head Model Retrieval
Based on Hierarchical Facial Region Similarity, Proc. of
15th International Conference on Visual Interface (VI2002),
Canada
[24] Afzal Godil, Patrick Grother, Sandy Ressler, Human
Identification from Body Shape, proceedings of 4
th
IEEE
International Conference on 3D Digital Imaging and
Modeling, 2003, Banff, Canada.
[25] Afzal, Godil, Sandy Ressler and Patrick Grother, "Face
Recognition using 3D surface and color map information:
Comparison and Combination", the SPIEs symposium on
Biometrics Technology for Human Identification, April
1213, 2004, Orlando, FL
[26] Allen, B., Curless, B., and Popovic, Z. 2004. Exploring the
space of human body shapes: data driven synthesis under
anthropometric control. Proc. Digital Human Modeling for
Design and Engineering Conference, Rochester, MI, June
1517.SAE International
[27] A. Khotanzad and Y. Hong, Invariant image recognition by
Zernike moments, Pattern Analysis and Machine
Intelligence, IEEE Transactions on, vol. 12, no. 5, pp. 489
497, 1990.
[28] A. Mademlis, A. Axenopoulos, P. Daras, D. Tzovaras, and M.
Strintzis, 3D Content-Based Search Based on 3D
Krawtchouk Moments, Proceedings of the Third
International Symposium on 3D Data Processing,
Visualization, and Transmission (3DPVT06), pp. 743749,
2006.
[29] Afzal Godil and Sandy Ressler, "Retrieval and Clustering
from a 3D Human Database based on Body and Head Shape ",
SAE Digital Human Modeling conference, 02 July to 4th
August, 2006, Lyon, France
[30] Lei Yang, and Henggen Shen, A pilot study on facial
anthropometric dimensions of the Chinese population for half-
mask respirator design and sizing, International Journal of
Industrial Ergonomics, Volume 38, Issues 11-12, November-
December 2008, Pages 921-926