Facial Attractiveness Classification using Deep Learning
Ricky Anderson, Aryo Pradipta Gema, Suharjito Suharjito, Sani Muhamad Isa · 2018
Facial attractiveness classification application has many various usabilities, including photo editing, photo beautification, photo grading and dataset labeling. While face attractiveness classification seems to be related to personal preference, building a robust attractiveness classifier is not impossible. There are several studies that have developed a classification system of facial attractiveness using a convolutional neural network and provide satisfactory results. The use of Image-net pre-trained convolutional neural network has been largely used by face-related research, yet none of them are related to facial attractiveness. This study aims to compare the famous deep learning architecture, such as VGG, Inception, and ResNet models. This research also aims to analyze the effect of using the Viola-Jones algorithm, as a preprocessing method, to the classification result of the built model. Viola-Jones algorithm will detect faces in image data, and 2 types of cropping will be done to extract the face region from the image, namely loose-crop and tight-crop. This research produces the highest accuracy value of 82.52% by using ResNet50 model and loose-crop preprocessing method.