Offline signature verification and forgery detection using a 2-D geometric warping approach

We present a method of discriminating between authentic and forged signatures using 2-D geometric warping. After an initial coarse-alignment step, we use an automatic morphing correspondence algorithm to compute 2-D geometric warps that align the strokes of a questioned signature with those of known reference examples. We use distance maps to compute a difference metric, and then either accept the signature as genuine or reject it as a forgery depending on how different it is from the reference examples. Our method achieves equal error rate (EER) accuracies of about 94%–96% on our English dataset of blind forgeries and 87%–91% on casual forgeries (unprac-ticed imitations). Further evaluation of our method using the SigComp2011 competition dataset shows that our accuracies for skilled forgeries are comparable to those of several other recent methods. We are particularly encouraged by the performance of our method on the Chinese portion of the dataset, in which our EER accuracy (74%) is better than all but one of the systems that participated in the 2011 competition.