Writer Identification using Neural Network

Abstract

Computer Vision has been evolving everyday with advancement in the Deep Learning. Residual Neural Networks is one of such image classification techniques. This paper is an application of ResNet 50 for the purpose of writer identification using handwriting biometric – signature. Different signature verification competitions had used many approaches. Here SigComp2009 dataset is used and experimental results are discussed. ResNet 50 is able to achieved 92% accuracy for 780 signatures used randomly from ICDAR 2009 dataset of genuine signatures.