Authentication Based on Modeling of the Signature Velocity

Nowadays, fast and accurate algorithms for signature verification are very attractive. In the area of dynamic signature verification, the features are classified into two groups: parametric and functional features. In parametric algorithms, although the speed of features extraction and classification process is faster than function based approaches but they are less accurate. The goal of this paper is modeling of the velocity signal that its pattern and properties are stable for a person. With using pole-zero models based on discrete cosine transform, a precise method is proposed for modeling and then features are extracted from strokes. These features are the deference of pole angles of strokes. Applying linear, parzen window and support vector machine classifiers, the proposed algorithm was tested on data set from Persian, Chinese, English and Turkish people and with common threshold, resulted equal error rates of 1.25% and 1.78% in the random and skilled forgeries, respectively.