查看文章

Biometrics technology is very important in terms of security issues like the identification of personal identity. Many solutions have been offered regarding biometric technology such as eyes-iris recognition, face recognition and vein pattern recognition. Moreover, one of the today’s most important authentication methods is fingerprint recognition. Each fingerprint has different pattern of ridges, valleys, deltas and cores. Those pattern types indicate unique fingerprints such as arch, left loop, right loop, tent arch and whorl. The issue of fingerprint pattern recognition is a crucial prior step to speed up the matching process of fingerprint recognition systems. Therefore, an accurate pattern recognition method is always needed, especially for large fingerprint databases. Besides traditional methods, recently, CNN is mostly used for fingerprint pattern recognition and there are many studies in the literature which achieve high recognition rates. In this study, we propose an automated tecnique toward fingerprint classification using various pretrained CNNs Xception and NasNetLarge in order to increase recognition rates. We performed experiments using NIST Special database 4 and we achieved 97.3 98.5% recognition rates respectively, which are the best scores up to now, for four categories: arch, right loop, left loop and whorl. The models was also tested into 5 fingerprint classes which arch and tented arch were seperated as two different classes with the recognition rate of 91.5% and 90.2% respectively.