Neuro-Fuzzy Approaches to Signature Verification
Vamsi Krishna Madasu, Madasu Hanmandlu, Shweta Madasu · 2003
This paper presents neural networks based approach and fuzzy modeling based techniques for the verification of off-line handwritten signatures and detection of simple forgeries. The angle distribution within the signature box constitutes the features needed for the modeling of the signature. The angles made by signature pixels are computed with respect to a reference point, which is taken as the left hand corner of the box. This angle distribution is then clustered using the fuzzy C-means algorithm. By considering the clusters so obtained over the several samples as fuzzy sets, a Takagi-Sugeno model is constructed for the twin-purpose of verification and forgery detection. The same features are also fed to a back propagation neural network (BPNN). The results for both the approaches are demonstrated on sample signatures of four persons.