2D log-Gabor filters for competitive coding-based multi-spectral palmprint recognition

This paper presents a novel multi-spectral palmprint recognition approach based on multi-resolution 2D log-Gabor filtering aiming to enhance the recognition performances of the coding-based approaches using multi-spectral images. The proposed approach consists of the following three major steps: (i) the feature extraction step employs a 2D log-Gabor filter bank where the final feature map is composed using the bitwise competitive coding, (ii) the matching step uses the normalized bitwise Hamming distance to capture efficiently the similarities between feature maps, and (iii) in the decision step, the feature maps are fused to get a final score through a novel feature fusion technique allowing to eliminate the inherent redundancy of the features of neighboring spectral bands. The experiment carried out on the multi-spectral palmprint MS-PolyU database have shown that the proposed method outperforms to the state-of-the-art methods for the verification and identification modes.