Cunjian Chen

Date of Graduation

2014

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Arun Ross

Committee Co-Chair

Donald Adjeroh

Committee Member

Guodong Guo

Committee Member

Xin Li

Committee Member

Cunquan Zhang

Abstract

The problem of Heterogeneous Face Recognition (HFR) involves comparing and matching face images that look significantly different primarily due to variations in their photometric composition. Examples include matching face images acquired in different spectral bands (e.g., visible versus thermal spectrum), or before and after the application of makeup. In this dissertation, we develop and evaluate a robust face recognition method to address this challenge.;The first part of the thesis deals with the topic of facial cosmetics. In this regard, we demonstrate the negative impact of facial cosmetics on existing face recognition as well as gender and age estimation systems. Next, we design a method that automatically detects makeup in face images. The proposed method extracts a feature vector that captures the shape, texture and color characteristics of the input face image, and employs a learning approach based on SVM/AdaBoost to determine the presence or absence of makeup. Finally, we design a patch-based ensemble learning method to perform makeup-invariant face recognition. In the proposed scheme, each face image is tessellated into patches and each patch is represented by a set of feature descriptors, viz., Local Gradient Gabor Pattern (LGGP), Histogram of Gabor Ordinal Ratio Measures (HGORM) and Densely Sampled Local Binary Pattern (DS-LBP). Then, a novel Semi Random Subspace Linear Discriminant Analysis (SRS-LDA) method is used to perform ensemble learning by sampling patches and constructing multiple common subspaces between before-makeup and after-makeup facial images. Finally, collaborative-based and sparse-based representation classifiers are used to compare feature vectors in this subspace and the resulting scores are combined via the sum-rule. Extensive experimental analysis demonstrates the efficacy of the proposed method.;The second part of the thesis deals with the topic of cross-spectral face recognition. Here, we design a method to compare input face images originating from the visible (VIS) and thermal (THM) spectrum. In the training phase of the proposed method, face images from VIS and THM are filtered and tessellated into patches. Each patch is represented using Pyramid Scale Invariant Feature Transform (PSIFT) or Histograms of Principal Oriented Gradients (HPOG). Then, a cascaded subspace learning process, consisting of whitening transformation, factor analysis, and common discriminant analysis, is used to construct multiple common subspaces between VIS and THM facial images. During the matching phase, the projected feature vectors from individual subspaces are concatenated to form a single feature vector. A Nearest Neighbor (NN) classifier is then used to compare feature vectors and the resulting scores corresponding to three image filters are combined via the sum-rule. The proposed face matching algorithm is evaluated on two multispectral face datasets and is shown to achieve very good results.;In summary, the primary contribution of this dissertation is the design of a novel patch-based ensemble learning method in conjunction with a cascaded subspace learning process to perform effective heterogeneous face recognition.

Download

Share

COinS