Ensemble of multiple matchers for finger vein recognition
Loading...
Date
2017-06
Authors
Soh, Siang Loong
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Biometrics recognition system is important in identification and verification of an individual. Recently, the research on finger vein verification becomes popular due to the benefits such as hygiene and cannot be duplicated. Finger vein verification is also able to overcome community needs and health problems. Various feature extraction methods were proposed by researchers, such as repeated line tracking, Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Band-Limited Phase-Only Correlation (BLPOC. These methods are considered as hand-crafted feature extraction method. Learned feature extraction has not been used in finger vein verification yet. Hence, spatial pyramid pooling method is developed as learned feature extraction for finger vein verification. BLPOC is used as hand-crafted feature extraction which the scores obtained will be then fused together with the scores obtained from spatial pyramid pooling by using Support Vector Machine (SVM). The database used is FV-USM based on 123 individuals with 4 fingers each. In the result obtained, spatial pyramid pooling shows the highest EER, 4.368% , followed by BLPOC, 2.36% and the lowest is SVM, 0.1348%. As conclusion, fusion of learned feature and hand-crafted feature shows the best performance as compared to single feature matching.