Palmprint Authentication Using Fusion of Wavelet Based Representations

Wavelets are widely used to extract the texture features for pattern recognition applications including biometric authentication. This can be attributed to the discriminating capability of wavelet features and the availability of fast algorithms for implementing discrete wavelet transform (DWT). In most of the wavelet based palmprint applications, distribution of energies in space-frequency domain are treated as features to classify the palmprints. Although wavelet energies are good discriminating features, they fail to characterize the palmprints sufficiently. In order to enhance the discriminating capability and the palmprint recognition accuracy, we extract the intramodal palmprint line and energy features from the same wavelet decomposition of palmprint. The score level (product rule and sum rule) fusion of these features improves the recognition accuracy significantly. We empirically found 39.38% relative improvement (RI) with an overall equal error rate (EER) of 1.37%, on PolyU online palmprint database (left hand palmprints). The computational burden on feature extraction is very less, and the EER is superior to other state of the art approaches. The comparison of the results with the state of the art wavelet and fusion based palmprint recognition approaches demonstrates the effectiveness of the proposed approach in classifying palmprints.