Automatic evaluation of facial attractiveness

Abstract

In this paper we present an approach of applying machine learning algorithms to the task of predicting human attractiveness. We have collected human beauty ratings of female facial images. We have chosen eigenfaces and ratio-based features as face representations. Along with k-nearest neighbors, we have used neural network and AdaBoost algorithms, which had not been used for this task before. Our analysis shows that machine learning algorithms have a preference towards facial symmetry, but also that a wider set of features needs to be included. We validate our results with a survey of four participants, which shows that facial attractiveness is a highly subjective judgement.