Quality measures for online handwritten signatures

Nesma Houmani (1) , Sonia Garcia-Salicetti (2, 3, 4)

1 SIGMA - Laboratoire Signaux, Modèles et Apprentissage Statistique
2 TSP - EPH - Département Electronique et Physique
3 SAMOVAR - Services répartis, Architectures, MOdélisation, Validation, Administration des Réseaux
4 CNRS - Centre National de la Recherche Scientifique

This chapter tackles the problem of quality of online signature samples. Several works in the literature point out signature complexity and signature stability as main quality criteria for this behavioral biometric modality. The drawback of these works is the measurement of such criteria separately. In this study, we propose to analyze such criteria with a different unifying view, in terms of entropy-based measures. We consider signature complexity as the intrinsic disorder of a signature instance and variability as the intra-class disorder of a set of genuine signatures. We introduce a novel statistical measure of complexity for signature samples and analyze it relatively to Personal Entropy that we proposed in former works. We study the power of both measures for an automatic writer categorization on several databases. We show that such categories retrieve separately on one hand degraded data and on the other hand good quality signatures. Finally, the degradation of signatures due to mobile acquisition conditions is quantified by our entropy-based measures

Mots clés: Signature complexity; Signature variability; Entropy; Hierarchical clustering; Hidden Markov Models; Writer categories; Online signature verification

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Nesma Houmani, Sonia Garcia-Salicetti. Quality measures for online handwritten signatures. Signal and Image Processing for Biometrics, 292, Springer, pp.255 - 283, 2014, Lecture Notes in Electrical Engineering, 978-3-642-54079-0. ⟨10.1007/978-3-642-54080-6_10⟩. ⟨hal-01263367⟩