Volume 12 - Issue 3
Earprint touchscreen sensoring comparison between hand-crafted features and transfer learning for smartphone authentication
- Jose-Luis Cabra
Fundacion Universitaria Compensar, Avenida (Calle) 32 No. 17 – 30. Bogota, 111311, Colombia.
jlcabra@ucompensar.edu.co
- Carlos Parra
Pontificia Universidad Javeriana, Ak. 7 #40 - 62. Bogota, 110231, Colombia
carlos.parra@javeriana.edu.co
- Luis Trujillo
Pontificia Universidad Javeriana, Ak. 7 #40 - 62. Bogota, 110231, Colombia
trujillo.luis@javeriana.edu.co
Keywords: Smartphone Authentication, Earprint, Capacitive images, Machine Learning, Touchscreen
Abstract
The smartphone’s lock screen is at a threshold between usability and comfort. For example, some
smartphone users prefer not to use the sliding or acceptance call button, but a more secure and efficient
way of picking up the phone instead. Others prefer the smoothest interaction possible with
their devices for getting quick access to smartphone services. In this paper, from a smartphone authentication
point of view, we propose using the touchscreen as an ear shape detector. This approach
helps verify the right user for incoming calls, supporting user privacy, as well as avoiding any action
approval through a button. In a one-against-all authentication scheme, looking for the best discrimination
model, genuine and impostor data are evaluated with two different authentication engines:
(i.) Transfer Learning (ii.) Different classifiers are fed by fused hand-crafted features like LBP, HoG,
and LIOP. Previous to both authentication approaches execution, the ear shape is extracted by an own
heuristic architecture to remove skin-related noises and highlight the region of interest. The classifier
results of this paper confirm that Earprint guarantees user verification, reaching an accuracy of 97.7.