Offline handwritten signature verification using back propagation artificial neural network matching technique

Handwriting is a skill that is highly personal to individuals and consists of graphical marks on the surface in relation to a particular language. Signatures of the same person can vary with time and state of mind. Several studies have come up with several methods on how to detect forgeries in signatures given to the security implication of signatures to daily business and personal transactions. This paper illustrates the proposed methodology for an offline handwritten signature identification and verification system which extracts certain dynamic features derived from velocity and acceleration of the pen together with other global parameters like total time taken and number of pen-ups in order to distinguish between forged signatures and genuine signatures signed under duress. Adaptive Window Positioning technique was employed for feature extraction, which focuses on not just the meaning of the handwritten signature but also on the individuality of the writer by dividing the handwritten signatures into 13 small windows of size nxn (13x13) such that it is large enough to contain ample information about the style of the author and small enough to ensure a good identification performance. Then, a signer specific codebook approach was used to generate a separate codebook of patterns for each individual signer such that the number of classes in each codebook varies as a function of the writing sample (signer), and a 3-layered Backward Propagation Artificial Neural Network (BPANN) method was used to produce a maximal matching and preserve the efficiency of the network. The proposed method was validated using a trained GPDS data set of 2400 original signatures of 100 different signers and comparing the results with those of two different known techniques of offline handwritten signature verification systems. The findings indicate that the proposed technique had the lowest ERR value of 7.23, indicating a more improved performance when compared against the two known techniques respectively thus proving to be a more efficient and superior method for offline handwritten signature identification and verification.