Dataset includes 33,660 ocular samples from 1,122 subjects captured by 196 different mobile.
All images were resized to224×224. the complete dataset is classified into three age-groups: namely young (18 to 39), middle-aged (40 to 59), and older adults (60 to 79).
User recognition and gender classifier models are fine-tuned on randomly selected gender and age-group balanced subset from 780 participants. Subject-disjoint, gender and age-group-balanced subset selected from 342 subjects are used as the test set for authentication and 260 subjects (130 males and females) for gender classification.
Age classifier models are fine-tuned on balanced subset selected from 432 subjects and evaluated on subset from 132 subjects, across 6 age-groups, namely 18−29, 30−39, 40−49, 50−59, 60−69 and 70−79.
TABLE I: EER and FNMR at 0.01 and 0.1 FMR for user authentication using CNN models for Left (L), Right (R) ocular region,and their score-level fusion (L+R) for Young, Middle-Aged and Older adults evaluated on balanced version of UFPR ocular datasets.
CNN
Age-Group
EER(%)
FNMR(%) @ FMR
0.01
0.1
L
R
L+R
L
R
L+R
L
R
L+R
ResNet-50
Young
8.60
9.52
9.06
37.63
35.35
36.49
53.74
54.03
53.89
Middle-Aged
8.62
9.08
8.85
30.76
25.04
27.9
52.23
48.55
50.39
Older
11.01
11.00
11.01
15.47
19.34
17.405
30.67
30.67
30.67
MobileNet-V2
Young
7.75
7.32
7.54
33.21
26.11
29.66
51.61
48.28
49.95
Middle-Aged
9.08
8.68
8.88
29.18
28.14
28.66
51.17
51.31
51.24
Older
8.04
9.63
8.84
18.54
13.47
16.005
39.47
36.00
37.74
ShuffleNet-V2
Young
6.93
6.96
6.95
37.63
38.09
37.86
56.44
56.26
56.35
Middle-Aged
8.32
9.25
8.79
37.38
46.07
41.73
55.08
61.10
58.09
Older
9.72
8.18
8.95
30.53
30.80
30.67
44.93
53.47
49.20
EfficientNet-B0
Young
7.64
9.61
8.63
32.54
27.00
29.77
55.40
48.38
51.89
Middle-Aged
6.95
9.03
7.99
38.07
26.69
32.38
49.12
49.31
49.22
Older
9.34
12.08
10.71
20.80
23.33
22.065
39.73
41.74
40.74
TABLE II: Accuracy of CNN-based Gender Classification on Left Ocular Region among Young (18 to 39 years), Middle (40
to 59 years) and Older Adults (60 to 79 years).
CNN
Young
Middle-Aged
Older
Male[%]
Female[%]
Male[%]
Female[%]
Male[%]
Female[%]
ResNet-50
98.39
98.19
100
96.67
99.17
98.06
MobileNet-V2
99.97
99.9
100
99.7
100
100
ShuffleNet-V2-50
98.23
97.57
98.28
97.56
94.76
98.89
EfficientNet-B0
95.89
97.54
98.58
96.44
95.23
86.94
TABLE III: Accuracy of the CNN-based Gender Classification on Right Ocular Region among Young (18 to 39 years), Middle
(40 to 59 years) and Older Adults (60 to 79 years).
CNN
Young
Middle-Aged
Older
Male[%]
Female[%]
Male[%]
Female[%]
Male[%]
Female[%]
ResNet-50
97.98
99.61
97.07
99.78
92.86
98.61
MobileNet-V2
96.84
92.35
95.86
98.66
94.76
97.5
ShuffleNet-V2-50
98.51
98.91
98.79
99.33
97.38
98.61
EfficientNet-B0
97.18
95.14
95.15
97.33
94.52
90
TABLE IV: Exact and 1-off accuracies of Age-group Classification for Young Adults
Left Ocular
Right Ocular
CNN
Exact [%]
1-off [%]
Exact [%]
1-off [%]
ResNet-50
46.72
93
52.87
93.16
MobileNet-V2
54.14
91.78
59.76
94.9
ShuffleNet-V2-50
52.47
91.16
45.16
85.27
EfficientNet-B0
28.225
53.195
31.6
62.64
TABLE V: Exact and 1-off accuracies of Age-group Classification for Middle-Aged Adults.
Left Ocular
Right Ocular
CNN
Exact [%]
1-off [%]
Exact [%]
1-off [%]
ResNet-50
31.61
81.73
35.03
78.43
MobileNet-V2
27.95
86.35
31.86
72.69
ShuffleNet-V2-50
28.46
75.2
24.61
47.98
EfficientNet-B0
20.93
57.86
19.56
55.73
TABLE VI: Exact and 1-off accuracies of Age-group Classification for Older Adults.
