Automatic Physiognomy System using Active Appearance Model and Convolutional Neural Network

Abstract

This research discusses the design and development of an automatic physiognomy system to determine a person’s tendencies based on the features of its face. Physiognomy itself is a method of predicting a person’s characteristics based on their facial features. Each facial feature has its uniqueness and characteristics, such as variations in distance, overall shape, and size. The facial image as input data is processed in every system step. Finally, the system displays the personality of that person. Simulations show that each algorithm can perform its respective functions well. The simulation results show that the combination of extracting facial features using the Active Appearance Model and Convolutional Neural Network for solving classification problems produces a very good number of personality traits predictions with each model accuracy value between 0.8 to 1, or 0.8797 on average. In addition, the model made proved to produce a good performance for the classification process with a true positive rate between 0.8834 to 1, or 0.9417 on average. This method can also detect many personality traits, with 28 personality traits that can be detected.