Studies On Gender Identification Based On Handwriting

Handwriting is influenced by postnatal learning.It contains the writer's congenital physical characteristics and can reflect the writer's writing habits and biometrics.The information extracted from handwriting can be used to judge the writer's gender,age and handedness.Among them,gender has great impact in the process of forming writing style.In various forensic and demographic investigations,it is useful to classify persons according to their genders.Confirming the gender of the writer can narrow the scope of the investigation and study,and improve the result of handwriting identification and verification.In addition,the combination of the gender and other biometrics has certain instructive for case analysis.We concern on studies of gender identification based on handwriting.Contour features,texture features and deep learning are used to achieve the goal of judging the gender of writers according to handwriting.Chain code and edge direction are designed to get contour information.We classify these feature with SVM and acquire the accuracy of 71.2% on IAM On-line database.LBP is researched and multi-scale LBP is used to obtain texture information.We build the multi-scale LBP and ensure appropriate K by experiments.Then,we classify these feature with KD tree and get the accuracy of 73.25% on IAM On-line database.In terms of deep neural network,we analyze the mature network structure.Based on dataset enlarging,we build a convolution neural network which contains seven convolution layers and other functional layers with caffe.Many skills are applied to improve the performance of the network.After setting parameters reasonably and fine-tuning,we acquire the accuracy of 76.17% on IAM On-line database,which is the highest one of gender identification based on handwriting on that database.