IEEE Consumer Electronics Magazine - January/February 2022 - 29

Table 1. Smartphones Embedded Sensors.
Sensor
Magnetometer
Accelerometer
Gyroscope
Table 2. Features Extracted From Touch Events and
Sensors.
Sensor description
Measures three axes (x,y,z) of the
system geomagnetic sector.
Measurements of the acceleration
force, including the gravity force.
Measures device rotation through
three axes (x,y,z) with roll, pitch,
and yaw movements of the
smartphone.
consists of one digit of PA, and two random
numbers between 1 and 6. Assume that the PA
letter is B and random numbers are 1 and 3,
where B represents the addition operation. So,
the user will add the digits of his PC with the
random numbers of LP, i.e., (4+1) and (5+3)
resulting in a pair (5,8), which represents the
fifth column and the eighth row of the OP-Grid
image. User will draw a pattern of opposite Lshape
from the fifth column to the eighth row.
This drawn pattern on OP-Grid image will be
verified by the password verifier module.
DATA COLLECTION AND TRAINING
This section explains the OP-Grid method in
terms of obtaining touch dynamics from the
touch interface and sensors, obtaining useful features
from the extracted touch dynamics to identify
users and what types of the classifier should
be included. In current smartphones, numerous
built-in sensors are mounted from which the sensor(s)
that can accurately reflect the behavioral
characteristics of the user was selected, whereas
the OP-Grid pattern is being drawn. OP-Grid considers
the touchscreen and three sensors that
are commonly available in current smartphones:
gyroscope, magnetometer, and accelerometer.
The touchscreen and these sensors represent different
levels of information about the user's
behavior. Table 1 shows a set of smartphone sensors
and their description.
In total, 33 students (14 females and 19
males) volunteered from our university to
engage in this study. Among 10 out of 33 participants
were assigned the role of attackers. Participants
created an account with their unique
usernames, PA, and PC. After choosing the
January/February 2022
Symbols
Tdr
Tp
Stp
Mp
Mnp
Ss
Sa
Ns
Avgx
Avgy
Stdx
Stdy
Tm
Name of feature
The time between touch down and touch
release
Average touch pressure
Standard deviation of touch pressure
Maximum touch pressure
Minimum touch pressure
Slide speed of finger from between points
Slide angle of finger from between points
Number of touch events of each segment
Average x value of each sensor
Average y value of each sensor
Standard deviation of x-axis value of each
sensor
Standard deviation of y-axis value of each
sensor
The maximum threshold time interval
between two moving events to form a
moving sequence.
registration credentials, the participants were
told to get login to the system in practice mode.
They were asked to get login daily with OP-Grid
technique, with their credentials for 20 days.
The participants were told to get login using OPGrid
pattern for no less than 25 times a day.
With this experiment, around 40 000 samples
were obtained from the participants.
Using machine learning techniques, from
these samples, the behavioral features were
extracted, which identify each sequence input
as a valid or imposter user.
When a user successfully logged-in based on
the input of his/her OP-Grid pattern, then the
information of user performing the pattern collected
by the touch sensors would be stored as
training sample; otherwise, the pattern will be
discarded. Authentication of the pattern may be
seen as a problem of classification. Legal users
will be labeled 1 and illegitimate users will be
labeled 0. Many research works9 have implemented
three classification algorithms, such as
SVM, Random Forest, and Neural Networks to
resolve classification issues. Table 2 shows the
29
IEEE Consumer Electronics Magazine - January/February 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2022
