Research Article
Fusing Two Kinds of Virtual Samples for Small Sample Face Recognition
Table 4
Classification error rates of different algorithms on the FERET database.
| The number of training samples | 1 | 2 | 3 |
| CRC-original | 0.5567 | 0.4160 | 0.5563 | CRC-mirror face | 0.5550 | 0.4050 | 0.5012 | CRC-symmetrical face | 0.5492 | 0.3950 | 0.4975 | CRC-the proposed scheme | 0.5417 | 0.3950 | 0.4625 |
| INNC | 0.5567 | 0.4170 | 0.4950 | INNC-mirror face | 0.5550 | 0.4240 | 0.4875 | INNC-symmetrical face | 0.5492 | 0.4100 | 0.4900 | INNC-the proposed scheme | 0.5408 | 0.4080 | 0.4713 |
| SFRFR | 0.5692 | 0.3840 | 0.4625 | SFRFR-mirror face only | 0.5550 | 0.3700 | 0.4150 | SFRFR-symmetrical face only | 0.5492 | 0.3630 | 0.4138 | SFRFR-the proposed scheme | 0.5492 | 0.3400 | 0.3987 |
| TPTSR | 0.5133 | 0.3700 | 0.4250 | TPTSR-mirror face only | 0.4942 | 0.3430 | 0.3713 | TPTSR-symmetrical face only | 0.4842 | 0.3390 | 0.3588 | TPTSR-the proposed scheme | 0.4692 | 0.3200 | 0.3400 |
|
|