Digital Eye Modification A Countermeasure to Automated Face Recognition

Abstract

Digital Eye Modification A Countermeasure to Automated Face Recognition Domenick Poster III This thesis describes and assesses a series of subtle digital eye modification techniques and their impact on automated face detection and recognition. The techniques involve altering the relative positioning of a person's eyes in a photograph using a variety of horizontal and vertical movements local to the eye regions. Testing with Eigenfaces, Fisherfaces, and Circular Local Binary Pattern face recognition algorithms on a database of 40 subjects and over 4000 modified images shows these subtle geometric changes to the eyes can degrade automated face recognition accuracy by 40% or more. Certain modifications even lower the chance a face is detected at all by about 20%. The combined effect of particular eye modifications resulted in subjects being both detected and recognized less than 20% of time. These results indicate that nearly imperceptible modifications made to one or more key facial features may foil face recognition algorithms. iii vii A.2 Vertical Modification Rank-1 Accuracy of Non-Normalized Images (1 Training Image per Subject

An Analysis and Validation of an Online Photographic Identity Exposure Evaluation System

“…An easy to use graphical interface is provided to users allowing images to be modified simply by moving a series of sliders. The underlying algorithms for performing the facial feature modification are Python scripts developed by Wolen [19] and Poster [7] being called from the Java Client providing the interface. The highlights of this feature are as follows:…”

“…Running the client on a machine requires the Java Runtime Environment (JRE) and version 2.7 or higher of Python to be installed. The validation of the Image Modification tool can be found in Wolen [19] and Poster's [7] work respectively as it is their systems that run underneath the UI. The successful interaction of these systems to completion when running the experiment inherently proves that a cohesive system built with readily available technologies can be used to provide fluid automation of photographic PII data gathering into biometric analysis for evaluation of exposure.…”