Leveraging Computer Vision Face Representation to Understand Human Face Representation
Face processing plays a central role in human social life. Humans, including very young babies, readily perform sophisticated computational tasks based on a brief glimpse of a face, such as recognizing individuals, identifying emotional states, and assessing social traits (e.g. attractiveness or trustworthiness). While the accuracy of the last phenomenon, known as physiognomy, is debated, it is consistent among individuals with similar demographic and cultural background and has an impact on real-life decisions such as in dating, employment, education, law enforcement, and criminal justice. A computational understanding of human face perception is of pressing importance not only for psychology but also for engineered systems that need to anticipate or generate faces with the desired percept and for fairness in law enforcement and the criminal justice system. We leverage ideas and tools from computer vision, metric learning, and information theory to derive scientific insights on how humans represent and perceive faces.