A study for facial beauty prediction model

Currently, most facial beauty research focuses on geometric features and apparent features by traditional machine learning methods. Geometric features rely heavily on accurate manual landmark localization of facial features. In addition, geometrical features and some apparent features focus on a particular aspect of facial description, such as distance, proportion, and texture, which causes a loss of information about facial beauty. To solve these problems, we present a facial beauty prediction model based on adaptive deconvolutional networks (ADN). ADN can extract multilayer apparent features from input images unsupervised and the feature extraction process is in accordance with the hierarchical visual perception mechanism of human brain. Experimental results show that the facial beauty prediction model presented can achieve high recognition rate based on three-class face image database.