Deep Periocular Recognition Method via Multi-Angle Data Augmentation
, , , e | 19 apr 2021
INFORMAZIONI SU QUESTO ARTICOLO
Pubblicato online: 19 apr 2021
Pagine: 11 - 17
DOI: https://doi.org/10.21307/ijanmc-2021-002
Parole chiave
© 2021 Bo Liu et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Figure 1.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.
Figure 7.
Figure 8.
Figure 9.
Figure 10.
Figure 11.
Figure 12.
COMPARISON OF INCPETION V3 AND MOBILENET V2 METHODS
| Methods | Verification Accuracy | Test Accuracy | Model Size |
|---|---|---|---|
| IncpetionV3 | 98. 55% | 98. 5% | 93MB |
| MobileNetV2 | 98. 21% | 98. 4% | 24MB |
AMPLIFIED SAMPLES OF THE DATA SET AROUND THE EYE
| Eye Peripheral Data Set | Training Set | Verification Set | Test Set | Total sample | Sample Type |
|---|---|---|---|---|---|
| CASIA-Iris-Thousand | 42000 | 14000 | 14000 | 70000 | 1000 |
PARAMETER SETTINGS
| Parameter Types | Parameter Settings |
|---|---|
| Max number of steps | 20000 |
| Batch size | 24 |
| Learning rate | 0. 001 |
| Learning rate decay type | fixed |
| optimizer | RMSProp |
| Weight decay | 0. 00004 |
OVERALL ARCHITECTURE OF MOBILENET V2
| Input | Operator | t | c | n | s |
|---|---|---|---|---|---|
| 224×224×3 | Conv2d | - | 32 | 1 | 2 |
| 112×112×32 | Bottleneck | 1 | 16 | 1 | 1 |
| 112×112×16 | Bottleneck | 6 | 24 | 2 | 2 |
| 56×56×24 | Bottleneck | 6 | 32 | 3 | 2 |
| 28×28×32 | Bottleneck | 6 | 64 | 4 | 2 |
| 14×14×64 | Bottleneck | 6 | 96 | 3 | 1 |
| 14×14×96 | Bottleneck | 6 | 160 | 3 | 2 |
| 7×7×160 | Bottleneck | 6 | 320 | 1 | 1 |
| 7×7×320 | Conv2d 1×1 | - | 1280 | 1 | 1 |
| 7×7×1280 | Avg pool 7×7 | - | 1 | - | |
| 7×7×1280 | Conv2d 1×1 | - | 1000 | - |
IMPLEMENTATION OF THE MOBILENET V2 CORE BUILDING MODULE
| Input | Operator | Output |
|---|---|---|
| H×W×N | 1×1 conv2d, ReLU6 | H×W×t N |
| H×W×t N | 3×3 dwise s=s, ReLU6 | H/s ×W/s×t N |
| H s ×W s ×t N | linear 1×1 conv2d | H/s ×W/s ×t N |
NETWORK MODEL PARAMETER SETTING
| Parameter Types | Parameter Settings |
|---|---|
| Max number of steps | 100000 |
| Batch size | 32 |
| Learning rate | 0. 001 |
| Learning rate decay type | Fixed |
| optimizer | RMSProp |
| Weight decay | 0. 00004 |
OVERALL STRUCTURE OF THE INCEPTION V3 NETWORK MODEL
| Type | Size of Convolution Kernel/Step Size |
|---|---|
| convolution | 3×3/2 |
| convolution | 3×3/1 |
| convolution | 3×3/1 |
| pooling | 3×3/2 |
| convolution | 3×3/1 |
| convolution | 3×3/2 |
| convolution | 3×3/1 |
| Inception modules | 3 Inception Module |
| Inception modules | 3 Inception Module |
| Inception modules | 3 Inception Module |
| pooling | 8×8 |
| linear | logits |
| Softmax | Classification of output |
SAMPLE EYE PERIPHERAL DATASET
| Eye Peripheral Data Set | Original Training Set | Raw Verification Set | Raw Test Set | Total Original Sample | Original Sample Type |
|---|---|---|---|---|---|
| CASIA-Iris-Thousand | 6000 | 2000 | 2000 | 10000 | 1000 |