Large-pose Face Analysis: Alignment, Reconstruction, and Recognition

This talk focuses on three relevant research problems in analyzing faces with large poses: face alignment, reconstruction, and recognition. We start by presenting two lines of research that focus on estimating the 2D and 3D facial shape from unconstrained photos, respectively. We achieve 2D facial shape estimation by face alignment. Specifically, we develop a pose-invariant face alignment method that can align faces with arbitrary poses, by fitting a 3D face model via Convolutional Neural Network (CNN). We will also present a photometric stereo-based algorithm for unconstrained 3D face reconstruction – 3D facial shape estimation. For large-pose face recognition, we address the fundamental problem of matching a frontal-view face image with a profile-view face image, a task that human still outperforms machine in the era of deep learning. We develop a novel Generative Adversarial Network (GAN) to simultaneously learn pose-invariant identity features and synthesize face images with arbitrary pose. In the end, we will briefly overview other ongoing research efforts in the Computer Vision Lab at Michigan State University, including image super-resolution, pedestrian detection for autonomous driving, etc.
Xiaoming Liu is an Assistant Professor at the Department of Computer Science and Engineering of Michigan State University. He received the Ph.D. degree in Electrical and Computer Engineering from Carnegie Mellon University in 2004. Before joining MSU in Fall 2012, he was a research scientist at General Electric (GE) Global Research. His research interests include computer vision, machine learning, and biometrics. As a co-author, he is a recipient of Best Industry Related Paper Award runner-up at ICPR 2014, Best Student Paper Award at WACV 2012 and 2014, and Best Poster Award at BMVC 2015. He has been the Area Chair for numerous conferences, including FG, ICPR, WACV, ICIP, and CVPR. He is the program chair of WACV 2018 and BTAS 2018. He is an Associate Editor of Neurocomputing journal. He has authored more than 100 scientific publications, and has filed 22 U.S. patents.