Event based offline signature modeling using grid source probabilistic coding

Abstract

A new offline handwritten signature modeling is introduced that confluences disciplines from grid feature extraction and information theory. The proposed scheme advances further a previously reported feature extraction technique which exploits pixel transitions along the signature trace over predetermined two pixel paths. In this new work the feature components, partitioned in groups, are considered as events of a grid based discrete space probabilistic source. Based on the 16-ary FCB2 feature, a set of 87 orthogonal event schemes, organized in tetrads, is identified. Next an entropy rule is drawn in order to declare the most appropriate tetrad scheme for representing a writer's signature. When skilled forgery is encountered verification results derived on both the GPDS300 dataset and a proprietary one, indicate enhanced EER rates compared to other approaches, including the previous reference of F CB2 as well. © 2013 Springer-Verlag.