Continuous Mobile Face Authentication

TitleContinuous Mobile Face Authentication
Publication TypeThesis
Year of Publication2017
AuthorsAltenhofer, C
Date Published12/2017
UniversityDepartment of Mobile Computing, School of Informatics, Communication and Media, University of Applied Sciences Upper Austria
Thesis TypeMaster's thesis

There are plenty of other authentication approaches on the market or in research but most of them are active. Active means that users have to explicitly do something in order to be authenticated by the system which demands the users attention and time. In this work we present our continuous mobile face authentication approach to distinguish between the owner of a smartphone and possible attackers, just by using video sequences or camera streams captured by the front facing camera. Our approach does not demand explicit user actions. The authentication is continuous, which means that the system does not require users to enter credentials at a certain point. Continuous authentication is another way to prevent unauthorized access to the mobile device and works passively in the background of the smartphone. In this work we explain our continuous mobile face authentication approach with our three design goals: unobtrusive, continuous and mobile. The authentication system should have a reasonable authentication performance. We develop and evaluate our approach in three steps in order to evaluate weaknesses and do improvements in the next step. Our first implementation is a prototypical implementation of a face recognition system which is able to recognize faces in images and distinguish between different people. This prototype however is not able to detect people beyond the test database which is the major improvement in the second step of the implementation process. Also a very important part in the second step is the weight function module which weights the importance of the face observation sample depending on the elapsed time between two observations. Another part of this implementation step is the recording of the face-database as source for evaluating our approach. In the third and last step a time decay module is introduced. This is necessary to give the confidence value a decay over time when for a period no new face samples can be detected to avoid an allegedly attack. Concluding, our results indicate that our continuous mobile face authentication is a viable approach for face authentication with a good performance specially when detecting allegedly attackers.