Computational facial attractiveness prediction by aesthetics-aware features

In this paper we address a novel topic, i.e., facial attractiveness prediction, which has both theoretic interest and numerous potential applications such as online dating and face beautification. Our focus is a computational model for prediction purpose. To this end, we briefly review existing algorithms and accompanying benchmarks. To overcome the drawbacks of prior study, we introduce a new dataset with high-quality landmark annotations and multiple facial attractiveness ratings. Several features (both appearance or geometric based) and baseline algorithms are validated on the data. Especially, an aesthetics sensitivity test is performed by utilizing the statistical Lasso. Other concerns such as prediction accuracies and personalized regression are also discussed in the evaluation.