Point Based Features for Contact-less Palmprint Images

In this paper, feature extraction and authentication scenarios for contact-based and contact-less palmprint images are investigated. The point-based feature extraction techniques like: Scale Invariant Feature Transform (SIFT), Harris corner detector, and Histogram of Gradient (HOG) in combination to Gabor filter are experimented for contact-based and contact-less palmprint authentication. In our experiments, we have used publicly available IITD database consisting of unconstrained contact-less palmprint images and compared its performance on these features with HongKong PolyU database acquired under constant illumination and constrained conditions. The presented work establishes that majority of the previous methods for palm print authentication, may work well with contact-based images, (with constrained environment) but fail to produce substantial results with unconstrained natural contact-less palmprint images. The proposed techniques give high recognition rate (97.5% and 93% GAR with Harris and SIFT respectively) for contact-less palmprint database. The experimental results in this paper have shown significant improvement for contact-based palmprint as well.