1. TECHNOLOGY NEWS
3D Searching Starts
The voxel
The key to the way computer pro-
grams look for 3D objects is the voxel
(volume pixel). A voxel is a set of graph-
to Take Shape
ical data—such as position, color, and
density—that defines the smallest cube-
shaped building block of a 3D image.
Computers can display 3D images
only in two dimensions. To do this, 3D
Sixto Ortiz Jr.
rendering software takes an object and
slices it into 2D cross sections. The
cross sections consist of pixels (picture
T
oday, both the Internet and elements), which are single points in a
corporate data warehouses 2D image.
are full of all types of digital To render the 3D image on a 2D
information, from simple text screen, the computer determines how
documents to complex appli- to display the 2D cross sections stacked
cations. One type of information gain- on top of each other, using the applic-
ing in prominence is the three-dimen- able interpixel and interslice distances
sional object. From computer-aided to position them properly. The com-
design (CAD) drawings of complex puter interpolates data to fill in inter-
engineering parts to digital represen- slice gaps and create a solid image.
tations of proteins and complex mol- item in a collection of objects.
ecules, an increasing amount of 3D 3D search engines could also help big Query formulation
information is making its way onto the companies quickly find whether they True 3D search systems offer two
Web and into corporate databases. have certain parts in their inventories. principal ways to formulate a query:
Because of this, users need ways to For example, many big engineering and Users can select objects from a catalog
store, index, and search this infor- manufacturing companies have huge of images based on product groupings,
mation. Typical Web-searching ap- databases of parts they search for com- such as gears or sofas; or they can uti-
proaches, such as Google’s, can’t do ponents to use in new products. lize a drawing program to create a pic-
this. Even for 2D images, they generally Frequently, engineers spend a lot of ture of the object they are looking for.
search only the textual parts of a file, time looking for such parts either man- For example, Princeton’s 3D search
noted Greg Notess, editor of the online ually or by older automated methods. If engine uses an application to let users
Search Engine Showdown newsletter. a part exists but engineers can’t find it, draw a 2D or 3D representation of the
However, researchers at universities the company must make it again, wast- object they want to find.
such as Purdue and Princeton have ing valuable time and money.
begun developing search engines that 3D search engines could eliminate The 3D search process
can mine catalogs of 3D objects, such these problems and also let users mine The 3D-search system uses algo-
as airplane parts, by looking for physi- for other important parts-related infor- rithms to convert the selected or drawn
cal, not textual, attributes. Users for- mation, such as cost or manufacturing image-based query into a mathemati-
mulate a query by using a drawing ap- method. cal model that describes the features of
plication to sketch what they are look- However, the technology faces sev- the object being sought. This converts
ing for or by selecting a similar object eral technical and marketplace hurdles drawings and objects into a form that
from a catalog of images. The search before it can become popular and com- computers can work with.
engine then finds the items they want. mercially successful. The search system then compares the
Susan Feldman, research vice presi- mathematical description of the drawn
dent of content technologies at IDC, a SEARCHING IN THREE DIMENSIONS or selected object to those of 3D objects
market research firm, said 3D image Advances in computing power com- stored in a database, looking for simi-
recognition and searching would be bined with interactive modeling soft- larities in the described features.
important for intelligence agencies ware, which lets users create images as
examining photos, corporate market- queries for searches, have made 3D- 3D SEARCH ENGINES
ing departments looking for one of search technology possible. Purdue and Princeton researchers
many product images, or any individ- 3D searching has several important have developed two major examples of
ual or organization trying to find an elements. 3D search engines.
24 Computer
2. Purdue University
Scientists at the Purdue Research and
Loop. Loop.
Education Center for Information L1 L2
Systems in Engineering, led by Pro-
fessor Karthik Ramani, created a 3D
Edge. Edge.
shape search technology called 3DESS
E1 E2
(3D Engineering Search System; http://
tools.ecn.purdue.edu/~cise/dess.html). Loop.
(a) (b) (c) L3
The engine is designed primarily to find
computer-designed industrial parts. Source: Purdue University
For example, said Ramani, the sys-
tem would be useful for any design and Figure 1. Purdue University’s 3DESS 3D search technology starts with (a) a drawing of an
manufacturing company that deals object that a user wants to find in a database. The system uses algorithms to extract the
with many CAD models. most important parts of the shape to create (b) a skeleton, a reduced graphical represen-
3DESS gives users three options for tation of the object. 3DESS then develops (c) a skeletal graph, using an algorithm that
searching industry databases: using a renders the skeleton in terms of three common topological constructs: loops, edges, and
drawing application to sketch a part nodes. This further reduces the amount of data in the object’s representation and makes
from scratch, penciling in modifica- it easier to store, index, and search for descriptions within the database.
tions to an existing part to create a new
search object, or selecting a part from or its skeletal graph with those of base and to describe drawings that
a catalog of choices. objects stored in a database. users submit as queries. The system
3DESS begins by converting query When the system retrieves models in compares queries’ shape representa-
drawings into voxels via an algorithm response to a query, users can input tions with those stored in the database
that uses mathematical representations which models more closely resemble to find the ones that are most similar.
describing the solid shape. the object they’re seeking. 3DESS then Princeton’s 3D search engine first
Another algorithm uses thinning, uses neural-network technology to renders a 3D object or drawing as a
which extracts only the voxels that rep- analyze users’ feedback, learn more voxel grid and then maps this to a
resent the most important parts of the about what they want, and fine-tune sphere, yielding spherical-based math-
shape, to create a skeleton of the both the current search and future ematical functions. The system then
object. This is a reduced graphical rep- similar searches. breaks the spherical functions into
resentation of the object that displays component parts that make up the ulti-
its basic outline and topology, as Figure Princeton University mate shape descriptor, a graphlike
1 shows. Professor Thomas Funkhouser and abstraction that represents the item’s
3DESS then develops a skeletal his colleagues at the Princeton Shape overall shape.
graph, using an algorithm that ana- Retrieval and Analysis Group (www. An advantage of the spherical
lyzes and renders the skeleton in terms cs.princeton.edu/gfx/proj/shape) de- approach, Kahzdan explained, is that it
of three common topological con- veloped their Web-based 3D Model lets the search engine work with shape
structs: loops; edges; and nodes, which Search Engine. The engine works with descriptors regardless of the object’s
are connecting points between loops a Java-based application that lets users position. Thus, the system will match a
and/or edges. Expressing an object in produce, via their mouse, 2D render- query with the appropriate result even
terms of these common constructs, ings of objects to search for in a data- if the query drawing is positioned dif-
plotted on a graph, reduces the overall base. ferently than the object in the database.
amount of data in the object represen- Funkhouser and his team were not Searchers can refine their queries by
tation and makes it easier to store and available for comment. However, adding a text description.
index descriptions in the database. Computer was able to speak with The Web-based 3D Model Search
According to Ramani, 3DESS also Michael M. Kahzdan, an assistant pro- Engine has not yet been released com-
describes objects in terms of feature fessor at Johns Hopkins University mercially, according to Kahzdan.
vectors, a set of mathematical repre- who, while a Princeton PhD student,
sentations of various aspects of the worked on the 3D search project with CONCERNS
item’s shape, such as its volume, sur- Funkhouser. 3D search technology faces several
face area, and the number of loops, The Princeton search engine, like its potential obstacles. For example,
edges, and nodes. Purdue counterpart, uses mathemati- researchers are trying to improve accu-
The system can analyze a query by cal representations to store the salient racy so that search results more closely
comparing either its feature-vector set characteristics of 3D shapes in its data- match queries.
August 2004 25
3. Te c h n o l o g y N e w s
In addition, researchers must problem if users can select predrawn ing. For example, a physician could
improve search speed, said Nainesh objects from catalogs, he added. search a hospital’s database of already-
Rathod, president and CEO of soft- “Advances in systems allowing users diagnosed magnetic-resonance scans to
ware vendor Imaginestics, which is to efficiently and effectively generate accurately diagnose a vexing abnor-
licensing Purdue’s engine for use in its models would play an important role,” mality found in a patient’s scan.
products this fall. He noted that Johns Hopkins’ Kahzdan said. Most of today’s individual and cor-
researchers are currently trying to porate computer users don’t conduct
improve speed by developing new the type of searches that would justify
search algorithms. urrently, said Kahzdan, 3D paying for and learning to use sophis-
Ease of use is another important
issue. For 3D search engines, ease of use
will boil down to the manner in which
C search focuses on submitting an
image of an entire object as a
query and receiving an entire object as
ticated 3D search engines. Thus, said
IDC’s Feldman, the technology is likely
to find only small, niche markets.
people can enter queries, according to a matching result. He said it would be However, Notess said, within those
Search Engine Showdown’s Notess. important if, for example, researchers markets, 3D search could prove useful
Drawing shapes is more difficult than could give 3D search engines the capa- and successful. I
entering text queries. Thus, Notess bility to take a query that is a part of an
said, 3D search engines will be most object or scene and either return the Sixto Ortiz Jr. is a freelance technology
helpful for users with computer-aided entire object or scene as a result or find writer based in Spring, Texas. Contact
drawing skills, such as those who work the part within a larger object or scene. him at sortiz1965@charter.net.
with CAD and graphic design. Developing this approach, he noted,
Also, he noted, the quality of mod- is a much more challenging technical
els drawn by users can be a factor if problem but could be useful for such Editor: Lee Garber, Computer,
drawing is the main or only query purposes as recognizing health prob- l.garber@computer.org
approach. This would not be such a lems in the results of 3D medical imag-
C omputer
Wants
DON’T
BE SECURE.
RUN THE RISK.
You
Computer is always looking for interesting
editorial content. In addition to our theme
articles, we have other feature sections such as Ensure that your networks operate
safely and provide critical services even
Perspectives, Computing Practices, and Research in the face of attacks. Develop lasting
security solutions, with this peer-reviewed
Features as well as numerous columns to which you publication.
can contribute. Check out our author guidelines at Top security professionals in the field share
information you can rely on:
www.computer.org/computer/author.htm Wireless Security • Securing the Enterprise •
Designing for Security Infrastructure Security
for more information about how to contribute to • Privacy Issues • Legal Issues • Cybercrime •
Digital Rights Management • Intellectual Property
your magazine. Protection and Piracy • The Security Profession •
Education
Order your charter subscription today.
www.computer.org/security/