Authors: Bao LY VAN
Keywords: Online Signature
Abstract: This thesis contributes to the automatic identity verification using the online handwritten signature, which is often sampled by a digitizing tablet or a touch screen. The handwritten signature is a highly accepted biometric modality. The proposed algorithm is original, generic and independent of the experimental signature database. It can be installed with different acquisition devices without any adaptation.
The signature is modelled by a Hidden Markov Model. Firstly, we perform a personalized normalization of the signature features, which improves the quality of the Hidden Markov Model. In this stage, we experiment only the Likelihood information of the Hidden Markov Model, and show that the normalization of the signature features is crucial to the system performance.
Then, we exploit second information given by the Hidden Markov Model in order to verify the identity. It's the Segmentation of the signature, never used previously for this task. After, this information is fused with the Likelihood information to reinforce the verification system. The experiments show that the system performances are greatly improved compared to the exclusive use of the Likelihood information.
These experiments are performed on 4 signature databases, whose characteristics are very different, and then on the integrated database, which is simply a mixture of the 4 previous databases. The good system performances show the independence of the proposed algorithm with respect to the considered signature database, or to the signature acquisition device.