Date of Award

12-1992

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering

Department

Department of Electrical and Computer Engineering

First Advisor

Steven K. Rogers, PhD

Abstract

In this thesis three variations of an end-to-end face recognition prototype system are developed, implemented and tested. Each version includes real-time image collection, automated segmentation, preprocessing, feature extraction, and classification. The first version uses a Karhunen Loeve Transform (KLT) feature extractor and a K-nearest neighbor classifier. Version two uses the same feature set but utilizes a multilayer perception neural network with a back propagation learning rule. Finally the third version uses a Discrete Cosine Transform as the feature extractor and the K-nearest neighbor as the classifier. Only the KLT versions of the system were tested. The tests were based on three image sets, each collected over multiple days to analyze the effect on recognition accuracy of variations in both the image collection environment and the subjects over time. The first set consisted of 23 Subjects and was taken over a two day period. The second set consisted of four users and was taken over a seven day period. Finally, the third set consisted of 100 images of a single subject collected over several weeks.

AFIT Designator

AFIT-GE-ENG-92D-33

DTIC Accession Number

ADA258997

Comments

The author's Vita page is omitted.

Download

Share

COinS