Celebrity: Creation of A Database Towards an Automatic Beauty Evaluation System

In recent days, facial beauty analysis has attracted various beauty-related fields like plastic surgery, the cosmetic industry, photo retouching applications, etc., The facial beauty analysis framework depends on an aesthetic scoring mechanism and needs high quality large-scale facial image database for facial beauty prediction. Beauty analysis can be extended to the ranking of beauty pageants in beauty pageant competitions. Most of the beauty scoring systems have focused on feature fusion and selection mechanisms through hand-crafted features with limited low-quality databases which are inadequate to analyze facial beauty and lead to infancy in the research of facial beauty. This work presents a machinebased automatic evaluation system for beauty by creating and compiling a database that includes beauty pageant images of winners and runners-up of beauty pageants competition using a deep learning algorithm. To evaluate the created database, pre-trained deep-learning models are used where deep-learned facial features are extracted for learning latent representations of beauty from facial images. From experiments performed on the created dataset, the classification accuracy achieved is 76.15%, using a pre-trained CNN model, ResNet 50.