Multispectral Palmprint Recognition Using The Quaternion Model

Biometrics is a technology that uses biometric features for recognition. Biometric features like fingerprints, palmprints, iris have been widely used. Among all the biometric features, palmprint is gaining more and more attention. Palmprint is not only reliable biometric feature but also easy to capture.Palmprint has been used in biometric recognition widely for many years. During these years, many methods or systems such as 3-D palmprint recognition system, touch-less palmprint recognition system and multispectral palmprint recognition system have been developed.Multispectral palmprint system can fully use palm infor mation under different illuminations, thus it can get better recognition accuracy. Previously, multispectral palmprint images were taken as multi-modal biometrics and fusion scheme was used for feature extraction or matching. However, some information or correlation will be lost during the matching score level or decision level fusion. In this study, we proposed a new multispectral palmprint feature level fusion method that represents the multispectral palmprint images by the proposed quaternion model and can fully exploit the multispectral information. Firstly, multispectral palmprint images captured under red, green, blue and near-infrared (NIR) illuminations were represented by the four components of the quaternion model. We modify the principal component analysis (PCA) and discrete wavelet transform (DWT) to quaternion principal component analysis (QPCA) and the quaternion discrete wavelet transform (QDWT). Then QPCA and QDWT were applied to extract palmprint features which include QPCA feature and QDWT feature. Finally, the Euclidean distances between tow QPCA features and two QDWT features were calculated and the fusion distance was used to measure the dissimilarity between two samples, then the nearest neighborhood classifier was employed for recognition decision. The experiment result showed that using the quaternion model can achieve higher recognition accuracy. Given 3000 test samples from 500 palms, the recognition accuracy could get up to 98.83%.