ON-LINE SIGNATURE BASED RECOGNITION SYSTEM

The approach employed to protect on-line signature templates relies on a set of non-invertible transforms introduced. The considered protection scheme can be applied to any biometric template which can be expressed in terms of a set of different time/space-dependent discrete finite sequences specifically, a non-invertible transformation is applied to the set, thus generating a transformed template, which consists of discrete sequences. As described in Section IV, the resulting transformed templates can then be further processed, if the employed matcher is based on a sequence-based modeling approach (e.g., Hidden Markov Models), or directly stored in the system database, if the matcher works directly with sequence-based descriptions (e.g., Dynamic Time Warping).

          In this Section we detail the proposed protected on-line signature-based recognition system. Specifically, it is assumed that the employed acquisition device is able to capture the horizontal  and vertical  signature position trajectories, and the signal regarding the pressure applied by the pen, where  is a discrete time index, and  is the time duration of the signature in sampling units. A geometric normalization, consisting of position normalization followed by rotation alignment, is applied to the pen-position signals.  Other four discrete time sequences are then derived from and used as an additional set of functions, namely the path-tangent angle, the path velocity magnitude, the log curvature radius, and the total acceleration magnitude. The acquired on-line signatures are then represented by means of the set of sequences, where