Semester

Spring

Date of Graduation

2019

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

Afzel Noore

Committee Co-Chair

Victor Fragoso

Committee Member

Victor Fragoso

Committee Member

Keith Morris

Committee Member

Mayank Vatsa

Committee Member

Richa Singh

Abstract

Developing automatic face recognition algorithms which are robust to intra-subject variations is a challenging research problem in the computer vision research community. Apart from the well-studied covariates such as pose and expression, temporal variations in the facial appearance also lead to a decline in the performance of face recognition systems. This research focuses on analyzing the temporal variations in facial features due to facial aging, facial plastic surgeries, and prolonged illicit drug abuse. The contributions of this dissertation are fivefold: (i) behavioral and neuroimaging studies are conducted to understand the human perception of faces affected by temporal variations, specifically facial aging; (ii) a novel generative adversarial network-based solution is proposed to match age-separated faces; (iii) the influence of temporal variations in faces altered by plastic surgery procedures is examined and a novel framework for detecting and verifying such faces is proposed; (iv) the effect of illicit drug abuse on face images is introduced and a new Illicit Drug Abuse Face (IDAF) database is created for the research community; and (v) a novel algorithm for single-image based detection of faces with plastic surgery and illicit drug abuse is proposed and its utilization as soft biometric in enhancing face recognition performance is demonstrated.

This research attempts to evaluate how humans perceive facial age and their ability to recognize age-separated faces. To accomplish this objective, two human studies (behavioral and neuroimaging) are conducted. The findings from these studies suggest that regular faces are processed differently from age-separated faces, highlighting the need for building specialized face recognition algorithms for processing such faces. Motivated by this observation, we propose a novel deep learning algorithm for matching faces with temporal variations caused by age progression. The proposed algorithm utilizes a unified framework which combines facial age estimation and age-separated face verification using generative adversarial networks. The key idea of this approach is to learn the age variations across time by conditioning the input image on the subject’s gender and the target age group to which the face needs to be progressed. We demonstrate the efficacy of the proposed architecture on different facial age databases for age-separated face recognition.

We also analyze the temporal variations with respect to facial plastic surgeries. A novel solution is proposed to differentiate plastic surgery faces from regular faces by learning representations of local and global plastic surgery faces using multiple projective dictionaries. Experimental results on the plastic surgery database show that the proposed framework is able to detect plastic surgery faces with a high accuracy of around 98%. To verify the identity of a person, the detected plastic surgery faces are divided into local regions of interest that are likely to be altered by a particular plastic surgery followed by distance metric calculation of feature representations. The proposed framework for face verification is combined with two commercial systems to demonstrate an improvement in face verification performance.

In this research, the impact of prolonged illicit drug abuse on face recognition is also introduced. Certain drugs, when taken continuously in large quantities, can cause physiological changes in the skin. For instance, the long-term effects of methamphetamine and heroin can cause severe weight loss and skin sores while addiction to opiates may lead to accelerated aging. We demonstrate that these physiological variations induced due to extensive substance abuse dramatically decrease the performance of current face recognition algorithms by increasing the intra-class distance between the facial appearance of a subject. This research also proposes a novel projective dictionary learning based illicit drug abuse face classification framework to effectively detect and separate faces affected by drug abuse from normal faces.

Lastly, two novel algorithms for single-image based detection of faces with temporal variations, specifically, plastic surgery and illicit drug abuse are proposed. In the proposed formulations, the variations in different local regions of these faces are analyzed by incorporating deep learning based multi-instance learning. The proposed approaches also utilizes multi one-shot metric to encode inter-class and intra-class variations leading to higher face image classification accuracy. Moreover, the classification scores from the proposed algorithms are utilized as soft biometric information to enhance the performance of existing face recognition algorithm.

Download

Share

COinS