Attribute-Driven Automatic Generation of Realistic Face Textures

When creating a video game, every digital character must be created by professional artists. Their work is very labor intensive because the number of created characters are in the thousands, each of which has multiple visual components that must be created for each one. “Scanning” real actors to create a digital version of themselves can help speed up this process, but each scan must be altered to preserve the actor’s anonymity. Given all of the face data Ubisoft has from past projects, we plan to create a system that can generate faces of completely new digital characters automatically with just a few desirable attributes and with high quality and believable results. The faces generated should be good enough to use on player characters and minor characters with little to no modification by artists. This will give them more time to work on major characters in the game.

Faculty Supervisor:

Sudhir Mudur

Student:

Christian Murphy

Partner:

Ubisoft Divertissement

Discipline:

Engineering - computer / electrical

Sector:

University:

Concordia University

Program: