Male and female facial attractiveness prediction: An image-based approach using convolutional neural network-based models

Abstract

In recent years, significant research has been conducted on the use of deep learning for prediction of facial attractiveness. These studies are expected to have various applications such as recommendation systems and face beautification. Therefore, it is crucial to improve the prediction accuracy. In this study, to improve the accuracy of facial attractiveness prediction, several convolutional neural network-based models were built using sex-specific datasets. Then, their accuracies were compared. The results showed that VGG19 and VGG16 had the highest accuracies for the male and female face datasets, respectively. A detailed confirmation of the factors necessary for prediction is expected to contribute to the construction of models based on human perceptual characteristics. These models maybe utilized in various engineering applications.