A Multimodal Biometrics Authentication System

Amal Seralkhatem Osman Ali

Abstract: An automatic authentication system based solely on fingerprints or faces is often not able to meet the system performance requirements. Face recognition is fast but not extremely reliable, while fingerprint verification is reliable but inefficient in database retrieval. This work proposes a system, which integrates face and fingerprint modalities. The system overcomes the limitations of face recognition systems as well as fingerprint verification systems. The proposed system operates in the verification mode with an admissible response time. The proposed face modality incorporates the Gabor Wavelet features and the Local Binary Patterns Variance (LBPVar) features. Those two facial descriptors are complimentary in the sense that LBPvar captures small appearance details, while the Gabor features encodes facial shape over a broader range of scale. Both feature sets are high dimensional, so it is beneficial to use the Principal Component Analysis (PCA) to reduce the dimensionality prior to normalization and integration. The Kernel Discriminative Common Vector (KDCV) method is then applied to the combined feature vector to extract the discriminant nonlinear features for recognition. As for the fingerprint module, an algorithm based on extracting finger Minutia is adopted to build a feature vector for each sample fingerprint. The two modalities are fused at the score level using a simple rule. The proposed system performance is evaluated over CMU Multi-PIE face and CASIA-FingerprintV5 public databases. The performance of the proposed model in the verification mode surpasses the performance of a number of multimodal biometrics state-of-the-art systems with a maximum verification accuracy of 99.2%.

Keywords: LBPVar, Gabor Wavelet, PCA, KDCV