2D Facial Images Attractiveness Assessment Based on Transfer Learning of Deep Convolutional Neural Networks

While beauty is subjective, it is not easy to quantify. Assessing facial beauty based on a computer perspective is an emerging research area with various applications. Different trainable models have been proposed to identify the attractiveness of facial beauty utilizing different types of features, machine learning techniques and lately, convolutional neural networks (CNNs) have proven their efficiency in image classification. The main objective of recent previous work is to enhance the performance of the existing trainable methods and make them suitable for beauty attractiveness identification. In this study, the accuracy and effectiveness of four affective pre-trained CNNs models (AlexNet, GoogleNet, ResNet-50, and VGG16) in assessing the attractiveness of human facial images using the CelebA dataset have been explored, evaluated, and analyzed. The results demonstrate that GoogleNet surpassed the investigated pre-trained networks with a performance accuracy of 82.8%.