Evaluation of the Putative Ratio Rules for Facial Beauty Indexing

Abstract

Understanding the rules of facial beauty is important for esthetic plastic surgery. Averageness and ideal proportions are the most investigated rules. In this paper, we integrate the findings on these two aspects to identify race invariant ideal facial proportions. Extensive research on the averageness hypothesis have verified that average faces are beautiful, which provides an objective way to generate representatives of beautiful faces. In order to ensure ethnic variety, 148 average faces from 61 countries/regions around the world have been collected to build the data set. 26 putative ratio rules, including golden ratio, neoclassical canons, etc., are collected to construct a candidate feature set. We first perform k-means clustering and then examine the 26 rules with respect to accuracy and universality on both the entire average face data set and individual clusters. The results show that: 1) the clustering result is consistent with the anthropologic divisions, 2) the top universal ratio features are consistent across different clusters, and 3) the accuracy of putative ratio rules can be improved by using data driven ideal values. The validity of the corrected ideal facial proportions has been verified on both synthesized faces and well-known beautiful faces in the real world.