Finger Knuckle Image ROI Extraction using Watershed Transformation for Person Recognition

Personal Authentication using biometric traits has emerged as a powerful technology in various scenarios including criminal and civilian applications. A large number of studies related to Finger Knuckle Image recognition have been proposed recently. It is evident that an accurate segmentation of region of interest is very crucial stage to achieve superior recognition results. In this article, we have proposed a novel ROI segmentation algorithm for extracting a fixed size major knuckle ROI from finger dorsal image using Watershed Transformation Algorithm (WTA). At first, the finger dorsal image is aligned by constructing a local coordinate system to control the spatial variation occurred during acquisition. For this, we have used the magnitude of the finger image to obtain the center knuckle line of the phalangeal joint as one of the ROI coordinate. A Random Decision Forest classifier has been used to measure the performance of proposed FKI recognition algorithm. The algorithm has been evaluated over publicly available PolyU FKP database. The results have also been compared with state-of-art algorithm. It has been observed that proposed algorithm outperforms with EER drop close to 23 %. It clarifies that the proposed algorithm has been extracting FKP ROI more consistently and hence assist to improve the performance of conventional finger knuckle image biometric system.