Multi-kernel approach to signature verification

Multi-kernel approach to signature verification

Signatures can be considered on-line, as multi-component discrete signals, or off-line, as gray-scale images. The problem of signature verification is considered within the framework of the kernel-based methodology for machine learning, more specifically, Support Vector Machine (SVM) approach. A kernel for a set of signatures can be defined in many ways and, at this moment, there is no method for finding the best kernel. We propose an approach of fusing several on-line and off-line kernels into a novel method embracing the whole training and verification process. Experiments with the public signature database SVC2004 have shown that our multi-kernel approach outperforms both single kernel and classifier fusion methods