This paper proposes a wavelet-based palmprint verification approach which is efficient in terms of accuracy and speed. The prominent wavelet domain features such as subband energy distribution, histogram, and co-occurrence features fail to characterize the palmprints sufficiently due to coefficient perturbations caused by translational and/or rotational variations in palmprints. In this work, firstly, a novel approach, termed as adaptive tessellation of subbands, is proposed to effectively capture the spatially localized energy distribution based on the spread of principal lines. Secondly, a set of discriminating features, termed as high scale codes (HSCODEs), and a translation and rotation invariant matching technique are proposed. HSCODEs effectively characterize the palmprints by capturing the spatial patterns corresponding to the low frequency components. Energy features and selected HSCODEs are fused at score and decision levels. Particularly, score level fusion enhances the verification accuracy significantly. Effectiveness of the proposed approach is examined on PolyU-ONLINE-Palmprint-II (PolyU) database. The experimental results show an overall equal error rate (EER) of 0.22%, which is better than the existing wavelet-based palmprint recognition systems and comparable to the computationally complex state-of-the-art approaches. The speed of the approach is high as all the features are extracted from the same wavelet decomposition of palmprint.
Further, it is shown that the proposed feature extraction technique can be extended for speech signals as well and such features can be fused with palmprint features for accuracy enhancement.
