Back to search

Aristotle University Of Thessaloniki (AUTH)

Digital signal processing with applications to biometrics and forensics

Abstract

dc:description

This PhD thesis introduces new algorithms and research approaches to the fields of biometrics and forensics. The research focuses on the prediction of biometric features such as gender, age interval and changes in facial appearance as a result of aging. Initially, an approach was developed that utilizes the tensor decomposition method Parallel Factor Analysis 2 in the field of biometrics. More specifically, the proposed method integrates two sources of information, namely, speech utterances and facial images, to predict the age interval and the gender of an individual. The motivation behind the proposed bimodal system stems from the desire to enhance age estimation and gender classification accuracy, when both speech recordings and facial images of the individual are accessible. The experimental results showcased the benefits of utilizing a combination of auditory and visual features for accomplishing both prediction tasks. Subsequently, the research focused on predicting and simulating the effects of aging on human facial features leveraging the dynamics of the Generative Adversarial Networks. More specifically, an approach was developed that addresses face aging as an unsupervised image-to-image translation problem. The proposed framework achieves age progression (i.e., future looks) and regression (i.e., previous looks) of face images that belong to a specific age class by translating them to other (subsequent or precedent) age classes. Two variants are presented. The first one learns a global transformation among age classes, while the second one incorporates a pyramid encoding and decoding scheme to more effectively diffuse age class information. The proposed variants were thoroughly evaluated with respects to both qualitative and quantitative criteria, yielding appealing face age progression and regression results when compared to ground truth images and to the corresponding results of state-of-the-art approaches for face aging. Furthermore, the doctoral research focused on addressing the problem of Cross-lingual Transfer Learning, and more specifically, on Cross-lingual Document Classification and Cross-lingual Authorship Attribution. The proposed approach utilizes the Parallel Factor Analysis 2 method to extract latent features of reduced dimensions from documents that are written in different languages. More specifically, the proposed method involves training two Parallel Factor Analysis 2 models for each pair of target and source languages: an unsupervised and a supervised model. The supervised model uses in its training the pseudo-labels for the training documents written in the target language which are derived from the unsupervised model to achieve a better fit. In addition, an explainable face aging framework was developed based on a well-known face aging approach, namely the Conditional Adversarial Autoencoder. The proposed method incorporates explainable Artificial Intelligence methods, such as Saliency maps and Shapley additive explanations, into the training process of the network. The goal of training that is guided by explanatory techniques is to supplement the feedback given from the discriminator to the generator with explanations that provide a “reason” for the discriminator’s decision. Thorough qualitative and quantitative evaluation demonstrates that the incorporation of the explanatory methods contributed significantly to the generation of more realistic age-progressed and regressed images.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Pantraki, Evangelia
  • Παντράκη, Ευαγγελία

Subjects

dc:subject × 26

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/55367

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Pantraki, Evangelia; Παντράκη, Ευαγγελία. Digital signal processing with applications to biometrics and forensics. Aristotle University Of Thessaloniki (AUTH), 2023. http://hdl.handle.net/10442/hedi/55367