Hybrid features fusion for writer identification usingsingle handwritten words

Abstract

Handwriting as a part of behavioral biometrics has been proved to having the abilityto sufficiently differentiate any two individuals. Therefore, in this work, we proposea system for writer identification using single handwritten words. In this regard, wepropose, associated to Support Vector Machine (SVM) classifier, a hybrid features fusionthat combined features extracted from a new descriptor namely, Multiscale Local OrientedGradient (MLOG) and features generated from Convolutional Neural Network VGG-16.Two known approaches of writer identification were addressed: writer-dependent andwriter-independent. Experiments conducted on two standard datasets, showed satisfyingand very promising results