English title

One of the basic problems in signature verification is variability and differences apparent on patterns of signature even for an individual. Signature segmentation to basic components, in addition to the access to the stable features, the hidden differences are revealed between genuine and forgery patterns. In this paper, signature patterns of two-dimensional are segmented by using dominant points. Segmentation decrease computation complexity also classification error of genuine and forgery signatures reduces. With extraction of strokes and one to one mapping their on the reference and test patterns based on algorithm of optimal subsequence bijection, parameter and functional features are defined. Applying the proposed approach to database extensive database with people in Farsi, Chinese, English and Turkish and contextual classification using linear, Parzen window and support vector machines, EER are 1.75% and 2.12% in group forger random and skilled, respectively.