Date of Graduation

2014

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Guodong Guo

Committee Co-Chair

Marvin Cheng

Committee Member

Xin Li

Abstract

Face recognition in unconstrained environments is still a challenge, because of the many variations of the facial appearance due to changes in head pose, lighting conditions, facial expression, age, etc. This work addresses the problem of face recognition in the presence of 2D facial appearance variations caused by 3D head rotations. It explores the advantages of the recently developed consumer-level RGB-D cameras (e.g. Kinect). These cameras provide color and depth images at the same rate. They are affordable and easy to use, but the depth images are noisy and in low resolution, unlike laser scanned depth images. The proposed approach to face recognition is able to deal with large head pose variations using RGB-D face images. The method uses the depth information to correct the pose of the face. It does not need to learn a generic face model or make complex 3D-2D registrations. It is simple and fast, yet able to deal with large pose variations and perform pose-invariant face recognition. Experiments on a public database show that the presented approach is effective and efficient under significant pose changes. Also, the idea is used to develop a face recognition software that is able to achieve real-time face recognition in the presence of large yaw rotations using the Kinect sensor. It is shown in real-time how this method improves recognition accuracy and confidence level. This study demonstrates that RGB-D sensors are a promising tool that can lead to the development of robust pose-invariant face recognition systems under large pose variations.

Share

COinS