Semester

Fall

Date of Graduation

2011

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

W Scott Wayne

Committee Co-Chair

Alfred Stiller

Abstract

Following an earlier development for fingerprints by Deal (1) and Stoffa (2), it was suggested that this algorithm may work on faces (or more precisely, face images). First, this work transformed a 2-D electronic image file of a human face into a numeric system via a similar random walk process by Deal and Stoffa. Second, the numeric system was analyzed, and the numeric system may then be tested against a database of similarly converted images. The testing determined whether the subject of the image is part of the database. Finally, the efficiency, quickness, and accuracy of such an algorithm were tested and conclusions about the general effectiveness were made.;The algorithm employed a Random Walk analysis of digital photographs of human faces for a fixed number of binary images which were generated from the source photograph using a Boolean conversion scheme. The Random Walk generated a series of transition probabilities for a particular scale. In short, the numeric system used to describe the face will consisted of two dimensions of data---scale and binary image. The numeric system for a particular source photograph was tested against a database of similarly constructed systems to determine whether the subject of the source photograph was in the database.;For the purpose of this work, a database of 400 images was constructed from 167 individual subjects using the FERET database. The 400 images where then analyzed, and tested against the database to determine whether the algorithm could "find" the subjects in the database. The algorithm was able, in its best configuration, to identify correctly the subjects of 168 of the 400 photographs. However, the total time to run an image (after capture by a digital camera) to database comparison was only 62 seconds, which represents a substantial improvement over previous systems.

Download

Share

COinS