Projective invariant hand geometry

Hand geometry biometric authentication techniques recognize a person's identity by the mathematics of hand features. Traditional hand geometry techniques require peg-based and contact-based image acquisition devices. This causes some issues in user acceptability and system reliability. Furthermore, none of the techniques uses hand features that are preserved under projective transformations. Technically, the hand features used by traditional hand geometry techniques are not inherently associated with a hand. A peg-free and contact-free hand geometry system is more desirable. But it requires new projective invariant hand features to compensate the effect brought by the viewpoint changing.; In this thesis, a novel concept of non-contact non-intrusive projective invariant hand geometry is introduced and analyzed. A hand plane composed of the joined fingers and the palm can be approximately considered as a rigid 2D planar object. Thus, the images of the hand plane taken from different viewpoints are the perspective projections of each other. Accordingly, projective invariant hand features extracted from one image are possessed by all other images. In light of existing theories, we propose several original projective and permutation invariants (PPI). Both PPI and cross ratios obtained from 12 hand feature points are used to create the projective invariant hand features. The noise immunity and the discriminability of a hand descriptor constructed by different projective invariants are studied. It has been demonstrated that the proposed PPI hand features outperform cross ratios in terms of both noise immunity and the discriminability. Experimental results show that the system achieves an equal error rate (EER) of 2.11% by a five-dimensional projective invariant hand feature vector on a database of 52 hand images. The EER becomes 0.00% when the feature vector dimension increases to 18. This thesis suggests a high potential for success in using this technique in real world applications. It also opens a door to a new exploration of projective invariant biometrics.