Signature recognition using binary features and KNN

Abstract

This paper proposes the use of binary features in offline signature recognition systems. Indeed, offline signature recognition finds mainly its importance for the authentication of administrative and official documents in which a higher accuracy is needed. In the proposed approach, features are extracted by using two descriptors: binary statistical image features (BSIF) and local binary patterns (LBP). To assess the reliability of the method, experiments were carried out using two publicly available datasets, MCYT-75 and GPDS-100 databases. Using a k-nearest neighbour classifier, recognition performances reach values high as 97.3% and 96.1% for MCYT-75 and GPDS-100 databases respectively. In signature verification, the classification accuracy measured with equal error rate (EER) achieved 4.2% and 4.8% respectively on GPDS-100 and GPDS-160. In addition, the EER for the MCYT-75 database has attained 7.78%. All those accuracies outperformed various performance results reported in literature.