An Improved Graph-Matching Technique for Offline Signature Verification

P. Nemmikanti, N. Jain, and P. Kumar (India)


Offline signature, Graph Matching, Assignment Problem, Hungarian Method, and Pattern Recognition


An improved approach for offline signature verification of the bank checks is proposed. Lot of research has been carried out in the area of online signature verification and have reported accuracies of over 95%[1]. Although online signature verification is much more accurate, however, costlier and infeasible in most of the application areas. We present a graph-matching approach, which analyses the geometrical shape of the signature. Three sample signatures are taken, and the pixel distance matrices of the three possible combinations of the sample signatures are constructed. Hungarian Assignment method [6] is applied to pixel distance matrices to find the costs/distances of sample signatures. The maximum of the three costs is taken as the threshold. A cost matrix is calculated with one of the sample signature and the signature to be verified. The cost matrix of the test signature constructed with any one of the sample signatures is used to quantify the authenticity of the signature. Our method requires only three sample signatures as against a relatively large number of signatures required by other similar methods. The method is superior to various other offline methods, and its results are comparable to that of the online system.

Important Links:

Go Back