Facial attractiveness in females has long been measured via proportions, distances and angles of faces. Specifically, facial beauty is positively correlated with things like eye height and eye width, and negatively correlated with nose area, chin length and cheek width - ie: bigger eyes and smaller noses.

All this judging used to be done by humans ticking off numbers on a 7-point Likert scale, but no more. As in the case of pilotless drones, humans can now taken out of the beauty contest loop, thx to the efforts of Amit Kagian and others.

Kagian has applied machine learning techniques to this challenge, and taught a computer to process some 84 facial coordinates and 6972 distance and angle features so as to closely emulate the results of human evaluators. They showed 91 facial images to 28 human raters, asked them to rate the attractiveness of each, and then tinkered with various induction algorithms to mimick the results to a correlation of 0.82 … not bad.

…presumably, Charlize Theron, extolled by femininebeauty.info would be ranked highly by humans and machine learners alike.

A machine learning predictor of facial attractiveness - A.Kagian et al, Vision Research, 48 (2), p.235-243, Jan 2008

This entry was posted on Sunday, April 6th, 2008 at 2:16 am and is filed under General. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.