THE DISSERTATION DEFENSE FOR THE DEGREE OF Ph.D. IN COMPUTER SCIENCE FOR

                                                                                Huimin Guo

Face recognition has been a long standing problem in computer vision. Among which, the most difficult problems are considered as unconstrained face recognition. Unconstrained conditions include allowing great variability in pose, ambient lighting, expression, size of the face, age, and distance from the camera, etc. In this dissertation work, we study both face identification and verification in unconstrained environments.

In the first part, we propose a face verification framework that combines Partial Least Squares (PLS) and the One-Shot similarity model. The idea is to describe a face with a large feature set combining shape, texture and color information. PLS regression is applied to perform multi-channel feature weighting on this large feature set. Meanwhile the PLS regression is used to compute the similarity score of an image pair by One-Shot learning (using a fixed negative set).

Secondly, we study face recognition with image sets, where the gallery and probe are set of face images from an individual. We model a face set by covariance matrix (COV) which is a natural 2nd-order statistic of a sample set. By exploring an efficient metric for the SPD matrices, i.e., Log-Euclidean Distance (LED), we derive a kernel function that explicitly maps the covariance matrix from the Riemannian manifold to a Euclidean space. Then discriminative learning is performed on the COV manifold: the learning aims to maximize the between-class COV distance and minimize the within-class COV distance.

Sparse representations dictionary learning has been widely used in face recognition, especially when large numbers of samples are available for each face (individual). Sparse coding is promising since it implies more stable and discriminative face representation. In the last part of this dissertation, we explore sparse coding and dictionary learning for face verification application. More specifically, in one work, we apply sparse representations to face verification in two ways via a fix reference set as dictionary. In the other work, we propose a dictionary learning framework with explicit pairwise constraints, which unifies the discriminative dictionary learning for pair matching (face verification) and classification (face recognition) problems.

Examining Committee:

Committee Chair:                       Dr. Larry S. Davis

Dean's Representative:              Dr. Min Wu

Committee Members:                Dr. Rama Chellappa

                                                Dr. Samir Khuller

                                               Dr. David Jacobs