Research On Handwriting Identification Algorithms

As a kind of bioinformatics technology,handwriting identification is widely used in the fields of justice and finance.How to accurately identify the authenticity of handwriting and identify the identity of imitation handwriting has become an urgent need and a huge challenge in the judicial and financial industries.At present,there are still several difficulties that hinder the development of handwriting identification.The difficulty is how to use a small number of signature samples to realize the problem of signature handwriting identification,and the difference in handwriting written by the same writer at different times and at different places.How to effectively extracting the handwriting characteristics of the writer is another major difficulty,and the text will focus on these two issues.Aiming at how to use a small number of signature samples to realize signature handwriting identification,this paper proposes a semi-text dependent handwriting identification idea.The idea of text dependent handwriting recognition has achieved very good results in the field of handwriting identification.However,the text dependent discrimination idea relies on a large number of handwritings of the same content,that the existing handwriting is the same as the handwriting to be identified.However,it is very difficult to obtain a large number of signatures of the same person in practical applications,which requires extremely high labor and time costs.However,the handwriting of non-identical words is more abundant and easier to obtain,so the signature recognition using non-identical handwriting has significance.In the identification of signatures,it is desirable to be able to weaken the effect of existing signature handwritings,to make up for the daily writing strokes of non-identical words,and to explore the minimum number of signature handwritings required for signature authentication.Experiments have shown that daily handwriting with non-identical words can compensate for the reduced accuracy of the number of signature handwriting.In view of the large difference in handwriting written by the same writer at different times and different places,this paper uses the deep convolutional neural network to deepen the depth of the network,so that the extracted handwriting features have small differences and large differences between classes.In this paper,CliqueNet network is adopted.In order to solve the above problems at the same time,a method of combining semi-text correlation with CliqueNet network is proposed to train the model together.Experiments show that the handwriting features extracted by the model are more robust through the combination of semi-text correlation and CliqueNet network.The experimental results on GPDS,MCYT and the self-built Chinese data set prove the method of the paper in handwriting identification,where there are few samples of the same word.At the same time,the experiment proves that the cascaded SVM classifier makes the classification result more accurate and improves the accuracy of handwriting identification.