Efficient finger segmentation robust to hand alignment in imaging with application to human verification

Abstract

Finger segmentation is the first challenging step in a Finger Texture (FT) recognition system. We propose an efficient finger segmentation method to address the problem of variation in the alignment of the hand. A scanning line is suggested to detect the hand position and determine the main characteristics of the fingers. Furthermore, an adaptive threshold and adaptive rotation step are exploited. The proposed segmentation scheme is then integrated into a powerful human verification scheme based on a finger Feature Level Fusion (FLF) method with the Probabilistic Neural Network (PNN). Three databases are employed for evaluation: IIT Delhi, PolyU3D2D and spectral 460 from the CASIA Multi-Spectral Palmprint database. The proposed method has efficiently isolated the fingers and resulted in best Equal Error Rate (EER) values for the three databases of 2.03%, 0.68% and 5%, respectively. Moreover, comparisons with related work are provided in this study.