Dissertation Proposal
In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
Pengfei Dou
will defend his dissertation proposal
Single/Multi-View 3D Face Reconstruction for Pose-Robust Face Recognition
Abstract
One of the most difficult challenges in automated face recognition is matching facial images acquired at different views. The pose variation in facial images causes not only misalignment between images but also inconsistency in facial appearance. To resolve the issue of misalignment between images, three-dimensional information, being a strong prior invariant to view perspectives, has been demonstrated beneficial in different computer vision applications. In face recognition research, 3D data has been widely employed as an intermediate media for pose normalization or synthesis to amend for the misalignment between facial images caused by pose variation. In these works, one crucial step is acquiring the personalized 3D face which, ideally, can be captured with 3D camera systems. However, the high cost and limited effective sensing range of 3D cameras have constrained their applicability in practical deployment. An alternative approach is reconstructing the 3D facial shape using 2D facial images, which has found wide applications in both the computer vision community and the computer graphics community. To resolve the issue of inconsistent facial appearance, discriminative feature learning and part-based face representation has been demonstrated promising in previous works.
The goal of this research is to improve 3D-2D and 2D-2D face recognition by at least 5% over state-of-the-art with explicit 3D face reconstruction from single/multi-view image(s). The specific objectives are to:
- Develop and evaluate a single-view 3D face reconstruction algorithm that overcomes the challenges of inaccurate facial landmark detection and localization and self-occlusion and erroneous 3D-2D pose estimation in facial images with large pose.
- Develop and evaluate a pose-robust face representation for 3D-2D and 2D-2D face recognition.
- Develop and evaluate a multi-view 3D face reconstruction algorithm.
- Compare single-view and multi-view 3D face reconstruction.
The proposed algorithms were evaluated on multiple public benchmarks for both 3D face reconstruction and 3D-2D/2D-2D face recognition. Compared with state-of-the-art, significant improvements were achieved by the proposed algorithms.
Date: Friday, March 24, 2017
Time: 1:00 PM
Place: HBS 350
Advisor: Dr. Ioannis A. Kakadiaris
Faculty, students, and the general public are invited.