VIScial: Visual Classification of Facial Shape Using Deep Transfer Learning

This study presents a deep learning approach for classifying five different face shape classes: heart, oblong, oval, round, and square. It aims to provide a hassle-free face shape classification system used for anthropometric studies, cosmetology, and as basis for make-up techniques and eyewear selection. A CNN model was initially built with user-defined parameters and was further optimized to enhance its accuracy. In comparison, utilizing transfer learning with VGG-16 notably achieved a classification accuracy of 96.83% and a classification error of 3.07% demonstrating excellent performance for classification tasks. While the device already provides satisfactory results, there remains room for further exploration.