1 Introduction
2 State of the art
2.1 Information fusion in biometrics
2.2 Mouse dynamics
2.3 Eye movement biometrics
2.4 Paper’s contribution
-
Introduces a new idea for biometric identification based on fusion of eye and mouse movements that reduces identity verification time and improves security.
-
Elaborates a new experiment type which can be easily applied in many environments.
-
Applies a dissimilarity space using dynamic time warping for extraction of features from eye movement and mouse dynamics.
3 Experiment
3.1 Scenario
3.2 Collections used
-
C4—24 users, four sessions per user, each containing three trials,
-
C3—28 users, three sessions per user, each containing three trials,
-
C2—32 users, two sessions per user, each containing three trials.
4 Methods
-
Preparation phase—when every trial was processed to extract different signals,
-
Feature extraction phase—when a sample was built on the basis of features derived from signals (there are three different approaches presented below),
-
Training phase—when samples with known identity were used to build a classification model,
-
Testing phase—when the model was used to classify samples with unknown identity,
-
Evaluation phase—when the results of the testing phase were analyzed.
4.1 Preparation phase
Signal | Formula | Description |
---|---|---|
x, y |
X and Y
| The raw coordinates |
vx, vy |
\(V_{x}=\frac{\partial x}{\partial t}, V_{y}=\frac{\partial y}{\partial t}\)
| The first derivative of X and Y (i.e., vertical and horizontal velocities) |
vxy |
\(V=\sqrt{V_{x}^{2}+V_{y}^{2}}\)
| The first derivative for absolute velocity |
ax, ay |
\(V_{x}^{\prime }=\frac{\partial V_{x}}{\partial t}, V_{y}^{\prime }=\frac{\partial V_{y}}{\partial t}\)
| The second derivative of X and Y (i.e., vertical and horizontal accelerations) |
axy |
\(V^{\prime }=\sqrt{V_{x}^{{\prime }{2}}+V_{y}^{{\prime }{2}}}\)
| The derivative of vxy |
jx, jy |
\(V_{x}^{{\prime }{\prime }}=\frac{\partial V_{x}^{\prime }}{\partial t},V_{y}^{{\prime }{\prime }}=\frac{\partial V_{y}^{\prime }}{\partial t}\)
| The third derivative of X and Y (jerk) |
jxy |
\(V^{{\prime }{\prime }}=\sqrt{V_{x}^{{\prime }{\prime }{2}}+V_{y}^{{\prime }{\prime }{2}}}\)
| The derivative of axy |
Diffmgx |
\(x_{\mathrm{mouse}}-x_{\mathrm{gaze}}\)
| The difference between mouse and gaze positions—axis x
|
Diffmgy |
\(y_{\mathrm{mouse}}-y_{\mathrm{gaze}}\)
| The difference between mouse and gaze positions—axis y
|
4.2 Feature extraction phase
-
Statistic values
-
Histograms
-
Distance matrix
4.2.1 Features based on statistic values
4.2.2 Histograms
4.2.3 Distance matrix
4.3 Training and testing phase
Collection | Samples per user | Training samples (N) | Testing samples (M) |
---|---|---|---|
C4 | 12 | 216 | 72 |
C3 | 9 | 168 | 84 |
C2 | 6 | 96 | 96 |
-
One for statistic values.
-
One for histogram values (normalized sum of results for five histograms).
-
One for distance matrix values (normalized sum of results for matrices built for 24 signals).
4.4 Evaluation phase
5 Results
Collection | Set | Accuracy (%) | EER (%) |
---|---|---|---|
C2 | Stat | 25.00 | 31.15 |
C2 | Hist | 21.88 | 34.78 |
C2 | Matrix | 15.62 | 34.59 |
C3 | Stat | 32.14 | 21.28 |
C3 | Hist | 32.14 | 20.68 |
C3 | Matrix | 46.43 | 16.78 |
C4 | Stat | 28.57 | 20.30 |
C4 | Hist | 57.14 | 10.32 |
C4 | Matrix | 92.86 | 6.82 |
5.1 Comparison of mouse and gaze
Set | Accuracy (%) | EER (%) |
---|---|---|
Fusion | 92.86 | 6.82 |
Gaze | 64.29 | 16.79 |
Mouse | 78.57 | 9.05 |
5.2 Examining the learning effect
Set | Accuracy (%) | EER (%) |
---|---|---|
Fusion | 100 | 2.94 |
Gaze | 92.86 | 9.37 |
Mouse | 85.71 | 5.04 |
6 Discussion
References | Testing sample duration (s) | Equal error rate | Training samples duration (s) |
---|---|---|---|
Gamboa et al. [13] | 50 s (200 s) | 2 % (0.2 %) | 200 s |
Hashiaa et al. [16] | 20 s | 15–20 % [HTER\(^{\mathrm{a}}\)] | 400 s |
Zheng et al. [15] | 100 s–37 min | 1.3 % | 166 min–60 h |
Feher et al. [18] | 42 s (139 s) | 10 % (7.5 %) | n/a (15 h per user) |
Shen et al. [17] | 12 s | 8.35 % [HTER\(^{\mathrm{a}}\)] | 885 s |
Our result (mouse) | 20 s | 9.05 % | 60 s |
Our result (fusion) | 20 s | 6.82 % | 60 s |
References | Testing sample duration (s) | Equal error rate (%) | Recording frequency |
---|---|---|---|
Komogortsev et al. [53] | 100 s | 16 % | 1000 Hz |
Holland et al. [40] | 60 s | 16.5 % | 1000 Hz |
Holland et al. [40] | 60 s | 25.6 % | 75 Hz |
Rigas et al. [33] | 60 s | 12.1 % | 1000 Hz |
Cantoni et al. [30] | 160 s | 22.4 % | 50 Hz |
Tripathi et al [38] | 60 s | 37 % | 1000 Hz |
Our result (gaze) | 20 s | 16.79 % | 30 Hz |
Our result (fusion) | 20 s | 6.82 % | 30 Hz |