| Popis: |
Fingerprint recognition systems have been shown to be vulnerable to spoof attacks with artificial fingers made from materials such as Play-Doh, gelatin and silicon. Anti-spoofing or liveness algorithms have been developed which determine at the time of capture whether a finger is live or spoof. Software algorithms rely on characteristics of the image captured by the underlying device and thus may be impacted by environmental conditions at the time of capture. In this study, we collected live fingerprints under various environment weather conditions (high/low/very low temperature, high humidity). Additionally we incorporated more spoofing materials of latex rubber, latex caulk, and latex paint. The algorithms were trained with a baseline dataset for Identix, Crossmatch, and Digital Persona devices with an average spoof/live equal error rate of 3.5%, 5.9% and 5.8%, respectively. Results showed an increase in error to 14.5%, 55.6% and 36.6%, respectively, when data of this type is not included in the training set. Similarly, we found the new spoof approaches developed defeat the liveness algorithm in almost all cases. When the algorithm is retrained to include new environmental and spoof images, the liveness algorithm is able to achieve an average error rate of 4.0%, 9.6%, and 11.4% for Identix, Crossmatch, and Digital Persona scanners, respectively. The impact of temperature, humidity, and novel spoof materials on anti-spoofing algorithm is significant and degrades performance. Performance can be restored when these factors are included in the training of the anti-spoofing model. |
