On-Line Signature Verification Model

We have studied methods for feature selection and pattern classification based on linear transformations. In particular, our study shows how the K-L method for discrete function representation can be used to classify signatures. A motivation for the use of the K-L method in the presence of noise was also described. We developed a novel classifier which combines the advantages of the K-L transformation and hypothesis testing. Our experimental results demonstrated that the K-L method and our novel classifier lead to viable signature verification algorithms.