May 7, 1998

CENPARMI's reading skills rise to a new level

Bogus cheques perform a genuine service

by Jordan Zivitz

Concordia's Centre for Pattern Recognition and Machine Intelligence (CENPARMI) needs your name in writing.

A CENPARMI research team directed by Professor Ching Suen and assistants Nicholas Strathy, Christine Nadal and Ke Liu has developed a computer program with the potential to automatically process cheques and payment slips, and handwriting samples are needed in order to train the system.

Suen and his research associates were turned down by a number of banks in their requests for handwriting samples and financial backing.

So the research team recently distributed 27,000 blank cheques to the Concordia community in an effort to collect samples. As an incentive for people to fill out and return their blank (and uncashable) cheques, CENPARMI held a draw on
Friday, April 24.

Everyone who submitted handwriting samples to the research centre was automatically entered into the contest. At the time of the draw, 6,000 cheques had been returned to CENPARMI.

The winners of the draw were Dawn Johnson (Human Resources), who received a dinner for two at Le Petit Szechuan restaurant, Gabrielle Korn (Alumni Affairs), who got a multi-purpose watch; and Computer Science student Rachel Bissonnette, recipient of a laser pointer with multiple heads.

CENPARMI's unorthodox method of collecting handwriting samples was necessary to adequately train its computer system, which works by scanning and digitizing cheques.

"The number [of samples] had to be large enough so that the computer would be familiar with different people's handwriting, and would perform better in recognizing the handwritten material," Suen explained.

The automatic cheque reader developed by CENPARMI has been in the works for more than 15 years. The system had been fed small amounts of handwriting samples before, but it was only in the last year or so that Suen's research team felt it was ready to be trained on a larger scale.

A $500,000, three-year contract from Bell Canada provided the support needed to get the main component of the cheque-reading computer system up and running.

After CENPARMI's computer program digitizes a cheque, it uses several pattern recognition methods to read the cheque's date and numerical and legal monetary amounts.

"[The computer program] learns how the shapes of different numerals are written," Suen said. "[It recognizes] structural and geometrical features like straight lines, loops, curvatures, and densities, and [is learning] new networks and classifiers to decipher handwritten numbers and words."

"The main component of the system is recognizing digits. That's what the [financial] industry uses now to process their cheques," added Strathy, who is the principal designer of the system. He admitted that the program's word recognition is still in the research stage.

The computer system currently recognizes 50 per cent of cheques correctly, with a 1-per-cent margin of error. If a cheque cannot be confidently read, it must be rejected; hence the need for CENPARMI to collect more handwriting samples in order to raise the system's confidence ratings in recognizing patterns.

When they sent the cheques, Strathy said, research assistant Christine Nadal got calls asking, "Is this safe?" and "What are you guys doing with the data?"

"This is serious research, and Concordia stands to be able to produce something that is really useful. Reading handwriting is a problem that's been around for years without really getting cracked. We human beings tend to think that it's obvious how to interpret this kind of information, but a machine has to be told every detail of what to do."

Even though the draw has already taken place, Strathy said there isn't any deadline for people to return their cheques to CENPARMI. "We'll take them any time." To submit handwriting samples to CENPARMI, please contact Christine Nadal, at 848-7952

Copyright 1998 Thursday Report. For technical comments please e-mail   Webmaster