Table 6.
Comparison of finger shape recognition accuracy (unit: %) (*: ResNet-50 is used for SDU-DB whereas ResNet-101 is used for PolyU-DB).
| Method | EER | ||
|---|---|---|---|
| SDU-DB | PolyU-DB | ||
| Non-training-based method | Fourier descriptor (FD) [29] | 13.8753 | 22.855 |
| FD + PCA [28] | 13.574 | 22.8718 | |
| Training-based method | VGG Net-16 [56] | 13.2758 | 19.2294 |
| ResNet-50 | 7.98 | 10.914 | |
| ResNet-101 | 8.305 | 9.9553 | |
| Score-level fusion of ResNet-50 and ResNet-101 by weighted product rule | 7.5665 | 9.5631 | |
| Non-training and training-based method | Score-level fusion of FD + PCA [28] and ResNet * by weighted product rule | 7.8147 | 9.8094 |