Call for participation
GRAPHICS RECOGNITION CONTEST
to be held in conjunction with
Second IAPR Workshop on Graphics Recognition (GREC'97)
Nancy, France, August 22-23, 1997
This contest is the second in the series that started with the Dashed
Line Detection Contest held in conjunction with GREC'95 [1,2,3]. The
proposed contest will test the participants' ability to segment text
from graphics and to recognize graphical primitives such as lines,
arcs, circles, and arrowheads. The specifications and the evaluation
protocol for the contest will be intentionally kept simple in order to
encourage participation.
All prospective participants must respond to this announcement by
contacting the contest chair indicating their intent to take part in
the contest (please see the instructions for prospective participants
at the end of this message).
As in the previous contest, select participants may be invited to
submit an article on their system and algorithms for the post workshop
book.
DETAILS OF THE CONTEST
**********************
o The uptodate version of this document is available at the URL
http://graphics.nynexst.com/iapr-tc10/contest.html
o The source code for generating synthetic images with varying degrees
of complexity and varying amount of noise will soon be made available
as a compressed tar file at
http://graphics.nynexst.com/iapr-tc10/contest97/synth-src.tar.gz
This code will also generate a ground truth file for every image.
o To assist those who have difficulty compiling the software on their
particular hardware/software platform, some sample images and ground
truth data generated by the above code will be made available at
http://graphics.nynexst.com/iapr-tc10/contest97/synth-img.tar.gz
o Some real scanned images with ground truth data will soon be made
available at
http://graphics.nynexst.com/iapr-tc10/contest97/real-img.tar.gz
o Evaluation software will be made available as a compressed tar file
at
http://graphics.nynexst.com/iapr-tc10/contest97/eval-src.tar.gz
o After people have had some time to fine tune their software to work
with the synthetic and real images, some test images will be
provided for the participants to gauge their success for themselves.
The test images will be located at
http://graphics.nynexst.com/iapr-tc10/contest97/test-img.tar.gz
The participants must return their text extraction and graphics
recognition results to the organizers by a specified date. These
results will not be used for ranking. They will only be used by the
organizers to make sure that the participants are on the right track.
o The README file in src directory will contain information about the
image generation parameters such as line thickness range, font size,
etc.
o Ftp access will be provided soon. Details will be made available in
the updates of this document.
Image File Format
*****************
o Only bi-level images will be used in this contest.
o The images will be in TIFF 6.0 CCITT Group 4 format. The software for
generating synthetic images will be linked to the public domain TIFF
library available by anonymous ftp at
ftp.sgi.com
in the directory
graphics/tiff
Hardware Platforms and Operating Systems
****************************************
o Participants in the contest can run their software on any of the
following platforms:
Sun SPARC with Solaris or SunOS
Silicon Graphics
PC's running Windows 3.1 or Windows 95
o The code for generating synthetic images and for evaluation will run
only on
Sun SPARC with Solaris or SunOS
PC's running Windows 3.1 or Windows 95
Live Contest
************
o Contestants will have to bring their software to the contest. They
will be required to run a live demonstration of their software on
several TIFF CCITT Group 4 images that will be provided at the
contest. These images will include a mix of synthetic and real images.
FTP access will be made available at the contest to assist
participants in downloading their software via the internet. However,
participants are strongly urged to still bring a copy of their
software on tapes or diskettes.
Evaluation
**********
Programs will be judged on the basis of the following performance metrics :
o Ability to extract text strings from mixed text and graphics where
the text could be
+ sparse or dense
+ isolated or touching/overlapping graphics
Characters within text strings could be isolated or touching.
No OCR is required. Only the bounding boxes of detected text
strings need to be provided to the evaluation software.
Fairly loose error tolerance will be allowed in these
bounding boxes.
All text will be horizontal or vertical.
o Detect the following graphical primitives
+ lines
+ arcs
+ circles
+ arrowheads
The lines, arcs, and circles could be solid or dashed (only simple
dash dash pattern will be used). The arrowheads could be hollow or
filled.
o The following attributes of each graphical primitive must be
identified
+ line: start point, end point, thickness, solid or dashed
+ arc: center, radius, start angle, end angle (start and end angles
must be in counter-clockwise order), solid or dashed
+ circle: center, radius, solid or dashed
+ arrowheads: tip, direction, hollow or filled
For lines that intersect each other, line segments created by the
intersections should be reported as separate lines.
For the above coordinates, the x-y coordinate system must have its
origin at the bottom left-hand corner of the image with the y axis
pointing upward and the x axis pointing to the right.
The units of the x and y coordinates should be in pixels.
NOTE : The exact evaluation methods are not yet complete and will be
posted as soon as they are ready.
* For further correspondence, contact Atul Chhabra at atul@nynexst.com.
Please use the subject heading "GREC Contest Inquiry".
References
**********
[1] R. Kasturi and K. Tombre (eds.), "Graphics Recognition: Methods
and Applications," First International Workshop, University Park, PA,
USA, August 1995, Selected papers published as Lecture Notes in
Computer Science volume 1072, Springer, 1996
[2] B. Kong, et al., "A Benchmark: Performance Evaluation of Dashed
Line Detection Algorithms," In "Graphics Recognition: Methods and
Applications," Lecture Notes in Computer Science volume 1072,
Springer, 1996.
[3] D. Dori, L. Wenyin, and M. Peleg, "How to win a dashed line
detection contest," In "Graphics Recognition: Methods and
Applications," Lecture Notes in Computer Science volume 1072,
Springer, 1996.
Instructions for Prospective Participants
*****************************************
All prospective participants must send a formal application to
participate in the contest. In addition to the intent to participate,
the application must include a brief description of your organization
or research group.
Along with the application, please enclose your answer to the
following questions:
1. Which hardware platform (e.g., Sun SPARC, Intel Pentium, ...)
would you like to run your software on for the live demo at the
contest?
2. Which operating system (e.g., Sun Solaris 2.5, Windows 3.1, ...)
would you like to run your software on for the live demo at the
contest?
3. Do you have access to a Sun SPARC machine or a PC running Windows
3.1 or higher for the purpose of running the synthetic image
generation software and the performance evaluation software? If
you have access to both, which one would you prefer to use?
The application may be sent in any of the following forms:
Email (plain ascii) to
atul@nynexst.com
Please use the subject heading "GREC Contest Application"
Regular or express mail to
Atul Chhabra
NYNEX Science & Technology
500 Westchester Avenue
White Plains, NY 10604, USA
Phone: (914) 644-2786
Fax to
Atul Chhabra at (914) 644-2561
For e-mail, fax, and express mail applications, if you do not receive
a response within a week of your submission, please try to reach me to
find out if I received your application. For applications sent through
regular mail, please allow two weeks to hear back.
Atul K. Chhabra Phone: (914)644-2786
NYNEX Science & Technology Fax: (914)644-2561
500 Westchester Avenue Internet: atul@nynexst.com
White Plains, NY 10604, USA