Generating Virtual Characters from Personality Traits via Reverse Correlation and Linear Programming
Résumé
This paper presents a system which generates a virtual character defined along three personality traits: Dominance, Trustworthiness, and Agreeableness. From these three traits, 14 surface physical attributes of the target character are automatically inferred. The configuration of our system accounts for an initial training phase, based on a reverse correlation experiment, from which we infer a multivariate linear model explaining the relationship between the perception of the three personality traits and the 14 physical attributes. The inverse model -- solved using linear programming -- allows for the real-time generation of virtual characters from an input personality profile.