Learning Local Correspondences for Static Signature Verification

摘要

This paper presents a new approach for off-line signature verification. Signature verification is performed by matching only well-selected regions of the signature images. More precisely, from the analysis of lower and upper contours of a signature image, region stability is estimated and the most stable regions are selected for verification, during the enrollment phase. In the verification phase, an unknown specimen is verified through the analysis of the selected regions, on the basis of a well-defined similarity measure. The experimental results, carried out on signatures from the GPDS database, demonstrate the potential of the proposed approach.