Beauty is in the eye of the beholder, the popular saying goes, but new Canadian research suggests this is not really true.

University of Windsor undergraduate student Joshua Chauvin has found that a computer can be trained to rate human facial attractiveness the same way that people rate the looks of others.

For his research, Chauvin trained a computer program called a neural network -- which is essentially a pattern recognition program loosely based on the human brain -- to mimic how humans assess attractiveness.

He had 100 people rate images of 100 others on attractiveness and then asked the neural network to rate the attractiveness of 33 images.

The findings show that 85 per cent of the time, the neural network's ratings fell within one point of the human subjects' ratings.

"So what (the neural network is) doing is coming up with its own rating for those images based on some sort of understanding of what it thinks the population finds as attractive," Chauvin told CTV.ca in a telephone interview.

Chauvin was assisted in his research by Dr. Marcello Guarini from the department of philosophy and Dr. Chris Abeare from the department of psychology.

He will present his findings this fall at the International Conference for Neural Computation in Madeira, Portugal.

In his research paper, Chauvin points to myriad research that has found that rating facial beauty is consistent across the globe and that specific features, such as facial symmetry, are important for determining facial attractiveness.

"One of the ideas is that there is some sort of objective basis for assessing facial attractiveness. Some sort of biological inclination to like those things and not like others," Chauvin said. "That a computer can recognize patterns in faces would further suggest that there are objective characteristics."

While computers are unlikely to replace humans as beauty-pageant judges, Chauvin said there are some practical implications for his research.

Marketing and advertising companies could use a neural network to assess how their target audience may respond to their casting choices.

"If you want a population's opinion about what attractiveness is, you can take a one-time poll and have them rate any given number of images on attractiveness features," Chauvin said. "And then if an advertising campaign has a model, they could input the model into the neural network and find out what the population thinks about that individual without having to poll the population over and over again for each advertisement."

Chauvin also said the research may one day lead to neural networks identifying telltale facial characteristics of some diseases.

While that research is in its infancy, the next step in Chauvin's study will be to determine if the neural network can also mimic how the subjects assessed the personality characteristics of the faces in the images.