Publication: Register-transfer level implementation of enhanced k-nearest centroid neighbor algorithm for finger vein identification.
No Thumbnail Available
Date
2017-01
Authors
Lai, Kok Choong
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Finger vein is a relatively new comer to the biometric field. Every person has unique finger veins that can be used for verification and identification. These finger veins are not as easily stolen, tampered or spoofed than many other biometrics, and yet still very accessible. Many advances have been made on finger vein identification, however it still suffers from low accuracy and slow processing speed. Therefore, an enhanced k-nearest centroid neighbor classification algorithm is proposed to increase accuracy, and implemented in hardware to increase speed. The proposed algorithm is built upon k-nearest centroid neighbor algorithm with varying distance metrics when searching for nearest centroid neighbors. The best combination searches for 12 neighbors. The first neighbor was found using Chebychev distance, while the remaining neighbors were found using correlation distance. The result is 85.03% in identification accuracy on FV-USM database, 3.9% more than FkNCN algorithm. The proposed algorithm was implemented on application-specific integrated circuit using Verilog hardware description language. The hardware was synthesized on Silterra 0.18µm process technology, occupies 40 mm2 in die area and runs up to 99MHz and consumes 2.56W. A 4.25 times speed improvement was achieved on the hardware compared to the software implementation, taking up about 143 milliseconds to identify a single finger vein sample. When identifying the entire FV-USM database, 49.5 times speed improvement was achieved. With these accuracy and speed improvements, the proposed algorithm and hardware architecture can contribute to the advancement of finger vein biometrics.