Signature Verification by Neural Networks with Selective Attention

Abstract

Automatic verification of handwritten signatures has numerous applications in checking the authenticity and validity of cheques and documents. Intra-class differences between genuine signatures and difficulty in collecting representative forgeries for comparison have been the main obstacles for its practical implementation. In this paper, a new standpoint of paying selective attention to the stable parts of genuine signatures is proposed to overcome the obstacles, and an experimental system based on it is given. To realize the selective attention, two strategies are addressed. One is to train the classifier with artificial forgeries generated by removing stable components from genuine signatures, so that the classifier can detect these stable components when verifying signatures. The other is to force the neural network classifier to pay special attention to local stable parts of signatures by weighting their corresponding node responses through a feedback mechanism. The experimental results demonstrate the potential of the proposed approach to compensate for the lack of representative forgeries for system training, and in improving the system's ability to identify skilled forgeries.