Facial expression analysis using 2D+3D facial feature tracking

Abstract

This paper presents a facial expression analysis system using the real-time tracking of a 3D camera. We first applied 2D facial motion tracking based on extended Active Shape Models (ASMs) from 2D texture images corresponding to captured 3D depth information. Three-dimensional facial motions are estimated based on the estimated feature point tracking from extended 2D facial motion tracking. From the estimated 3D facial motion using extended ASMs and depth maps, we extract MPEG-4 facial animation parameters (FAPs) from the 3D facial motion tracking, which provides more accurate estimation of facial motions invariant to view variations. Our experimental results for facial expression recognition using the Bosphorus database show better performance of facial expression recognition by 3D feature points (FPs) compared to 2D FPs after alignment of FPs, whereas direct distance measurement shows better performance by 2D FPs than by 3D FPs without alignment. ©2014 International Information Institute.