Affine-Invariant Fourier Descriptions for Feature-Based Facial Recognition

Date of Award

2006

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Mullins, David

Keywords

Fourier, Affine, Facial Recognition

Area of Concentration

Computer Science

Abstract

Facial recognition systems have been commercially employed in efforts to halt identity theft, eliminate duplicate voters, catch criminals, and automate the collection of demographic data. Despite the high demand for this technology, no existing software has provided a reliable means of identifying individuals in a scene using a stored database of images. In this paper, I explore one aspect of face recognition, known as image classification. This deals with the way in which images are parameterized and stored prior to performing comparisons. First, I explore the obstacles present in classifying facial images. Next, I examine the various approaches to resolving these issues. In researching the state of the art of facial recognition, I hypothesized that a geometric based feature-oriented method would be the most effective technique. I chose to implement this by using Fourier descriptors modified for affine invariance, as this would yield a substantial amount of information while excluding noisy and irrelevant data. I coded a program called OpenFace which utilizes affine-invariant Fourier descriptors. After performing a number of tests, OpenFace showed favorable results. I believe that this is a promising technology for facial classification.

This document is currently not available here.

Share

COinS