Skip to main content

Three-Dimensional Craniofacial Modeling for Novel Two-Dimensional Pose Synthesis and Age Progression

Researcher: Marios Savvides

Research Area: Security of Cyber-Physical Systems

Abstract

Scope: Sponsored by IARPA, the goal of this project is to develop robust methods to model faces and their underlying 3D structure for the purpose of synthesizing novel 2D views using 2D images. Example scenarios will be to automatically generate novel out-of-plane 3D/2D views (poses) of a person given just a single 2D facial mug-shot image collected from anywhere (web, camera, passport etc..) and be able to match against a face presented at any angle from surveillance footage or other sources. Additionally we can use multiple images to generate more accurate 3D models from a video sequence of 2D images for un-constrained matching. Additionally we will employ computer graphics and vision techniques to synthesize novel illumination views of a face as well as perform age progression for matching against temporally changed biometric data.

Outcomes: A complete system that will be able to analyze a 2D facial image, generate a 3D model that can be rotated to any pose to match against an un-constrained captured face image. 3D face models will be trained using actual 3D scanned data and the effects of ethnicity, lighting and age will be modeled.