In the first step of the study, 30 men and women were presented with 100 different faces of Caucasian women, roughly of the same age, and were asked to judge the beauty of each face. The subjects rated the images on a scale of 1 through 7 and did not explain why they chose certain scores. Kagian and his colleagues then went to the computer and processed and mapped the geometric shape of facial features mathematically.Vision Research http://dx.doi.org/10.1016/j.visres.2007.11.007
Additional features such as face symmetry, smoothness of the skin and hair color were fed into the analysis as well. Based on human preferences, the machine "learned" the relation between facial features and attractiveness scores and was then put to the test on a fresh set of faces.
Says Kagian, "The computer produced impressive results -- its rankings were very similar to the rankings people gave." This is considered a remarkable achievement, believes Kagian, because it's as though the computer "learned" implicitly how to interpret beauty through processing previous data it had received.
A machine learning predictor of facial attractiveness revealing human-like psychophysical biases
Amit Kagian et al.
Recent psychological studies have strongly suggested that humans share common visual preferences for facial attractiveness. Here, we present a learning model that automatically extracts measurements of facial features from raw images and obtains human-level performance in predicting facial attractiveness ratings. The machine’s ratings are highly correlated with mean human ratings, markedly improving on recent machine learning studies of this task. Simulated psychophysical experiments with virtually manipulated images reveal preferences in the machine’s judgments that are remarkably similar to those of humans. Thus, a model trained explicitly to capture a specific operational performance criteria, implicitly captures basic human psychophysical characteristics.