2D and 3D palmprint information, PCA and HMM for an improved person recognition performance

Biometric systems based on a single source of information suffer from limitations such as the lack of uniqueness, non-universality of the chosen biometric trait, noisy data and spoof attacks. Multimodal biometrics are relatively new systems that overcome those problems. These systems fuse information from multiple sources in order to achieve the better person recognition performance. In this paper, the 2D and 3D information of palmprint are integrated in order to construct an efficient multimodal biometric system based on fusion at matching score level and at feature extraction level. The observation vectors are created independently either from the original data of the two modalities (2D and 3D palmprint) or from their rotation invariant variance measures applied on textures. On each modality (or its corresponding invariant texture), we have applied the Principal Component Analysis (PCA) for reducing dimension of the feature vector. We have also used the multi-scale wavelet decomposition for each modality and the results of decomposition are combined and compressed using PCA for selecting the feature vectors. Subsequently, we have used the Hidden Markov Model (HMM) for modeling the feature vectors. Finally, Log-likelihood scores are used for palmprint evaluation. We note that the selected principal components of two modalities are fused at feature level and at matching score level. The proposed scheme is tested and evaluated using PolyU 2D and 3D palmprint database of 250 persons. Our experimental results show the effectiveness and reliability of the proposed system, which brings high identification accuracy rate.