by Frank Kuin
A modest computer room in one of Concordia's most renowned research centres offered a revealing glimpse into the future on November 20 -- a future in which computers will be better able to interpret such wide-ranging images as handwritten addresses, printed documents in different languages, and samples of biogenetic material.
During an open house at the Centre for Pattern Recognition and Machine Intelligence (CENPARMI), held to mark its 10th anniversary, remarkable demonstrations were given of pioneering projects to teach computers how to read. On a series of screens, digitalized images of handwritten city names and DNA samples were processed -- material that is still read and interpreted best by humans.
However, such demand exists for computer technology that would flawlessly recognize human handwriting and other irregular patterns that CENPARMI is conducting research in co-operation with a variety of institutions, such as the Jewish General Hospital, the postal service of France, and the Department of National Defence in Ottawa. They are co-funding research projects by Concordia scientists and graduate students to improve existing technologies.
"Computers are able to deal with larger quantities of data than people," said Xiangyun Ye, a post-doctoral fellow who is working with technology for the digital processing of DNA samples. "They are also more objective than people," she said, while demonstrating ways in which the computer corrects distortions in the images of six DNA segments, in order to make them directly comparable.
Her project in Biogenetic Image Analysis, carried out in collaboration with the Jewish General Hospital and l'École de Technologie Supérieure, is aimed at giving biologists an opportunity to contrast and compare great numbers of DNA samples electronically. That should help them, among other things, identify segments that are most important to human aging, the researcher explained.
Seated next to her, Frédéric Grandidier, a PhD student from France, was demonstrating an equally practical application of pattern recognition technology. In a project funded by the French postal service, he is trying to improve a system for recognition of handwritten city names that is already in experimental use in France.
"We try to remove all variability," Grandidier said, as he showed how a handwritten image on the screen reading 'Saint-Gilles' was made comprehensible to the computer. "The information has to be reduced to the most pertinent features. He ordered the machine to put the writing on a horizontal line, remove the skewed angle and make all the letters of equal height. After submitting it to a database of French city names, the computer then matched it with the right one.
Professor Ching Suen, Director of CENPARMI, elaborated on the challenges of handwriting recognition. "We hope to be able to recognize more casual handwriting in the future," he said. "Right now, the system still makes some mistakes here and there. We're trying to find out more about the knowledge of human readers, and how to implement it into the computer."
Once that is done, "people will be able to use the system with more confidence," said Suen, who has done more than 25 years of research in handwriting recognition. That, he said, would pave the way toward errorless electronic processing of handwritten cheques, one of the biggest challenges of the science.
But CENPARMI is not alone in that quest. "Internationally, there are several very active groups competing on this subject," Suen said, citing groups in the United States, France, Italy, Switzerland and Russia. "I think we are doing very well because of our long history and very extensive research in this area. We are leading in the field."
Suen is the driving force behind the centre, which has about 12 professors from different universities and whose industrial collaborators include Bell, AT&T, IBM and research institutes in Japan, China and Taiwan. In 2002, the centre will hold the largest international pattern recognition conference, in Quebec City, to be chaired by Suen.