Contactless Palmprint Identification Based on Patch Local Neighborhood Binary Pattern

• Main Authors: Doghmane, H., Amara Korba, M. C., Mentouri, Z., Boualleg, A.H, Bourouba, H., Sedraoui, M.
• Format: Conference Proceeding
• Language: English
• Subjects: Control systems; Feature extraction; Lighting; Palmprint recognition; Performance evaluation; Simulation
• ISSN: 2379-0067
• DOI: 10.1109/ICSC58660.2023.10449704

A new texture descriptor based on local neighborhood intensity variation is proposed in this paper, which enables contactless palmprint recognition. This proposed representation enables capturing the texture information of adjacent neighbours. It is therefore inspired by the principle that the neighbours of a given pixel contain a significant dataset of texture information, which can be exploited for an efficient texture representation in contactless palmprint recognition. The main benefit of using the mutual relationship between adjacent neighbors resides not only in considering the sign of the intensity difference between the central pixel and its neighbors, but also in taking into account the sign of the difference values between a given pixel and its adjacent neighbors, and between the central pixel and all neighboring pixels. As a result, the proposed model becomes much more robust when illumination is varied. Moreover, most local models, especially those of the LBP, are mainly focused on the sign information and therefore ignoring also its magnitude. It's important to emphasize that magnitude information has an auxiliary role, as it provides further information that complements the texture descriptor. It is therefore necessary to incorporate it into the proposed representation, in which the average absolute deviation of each pixel from its adjacent neighbors must be taken into account. This allows the development of a high-performance texture descriptor known as Patch Local Neighborhood Binary Pattern (P-LNBP). This descriptor generates sign and amplitude patterns based on the relative intensity difference between the pixel in question and the central pixel, where the corresponding adjacent neighbors are taken into account. Finally, these sign and amplitude patterns are concatenated into a single feature descriptor to generate an even more efficient feature descriptor. The proposed representation is applied on the CASIA database of contactless palmprints. Simulation results confirm that the resulting recognition rate becomes better than those provided by the most recent existing methods.