Using calligraphies features for off line writer identification

Abstract

This work proposes an off-line writer identification approach based on graphometrical and forensic features. We selected a set of features with independence of the text and some stability degree to natural changes in the writing. The system uses the LS-SVM classifier with RBF kernel, reaching up to 99.1% of success rate for an own database composed by 100 users with 10 samples per each one.