Signatures Verification Based On Texture Feature And Depth Feature

This days,more and more occasions use signatures as a means of identity verification,which has generated a large number of needs to identify the authenticity of signatures.Traditional signature verification requires signature authentication experts finding genuine signatures from a large number of samples manually,which takes a lot of time and costs.Computer signature verification is a technique that relies on a computer to extract features from a signature and uses machine learning models to automatically or assist experts in predicting the signature sample author.With the development of computer technology,computer signature verification technique has gradually attracted people's attention.This article introduced the research status of signature verification,and based on this,focused on the signature collection,preprocessing,feature extraction and feature matching technologies.This article mainly completed the following works:(1)Collect the signature image data required for the experiment,and use the scanner and 3D stereo microscope to obtain the 2D and 3D images of the signature,respectively.2D signature preprocessing included grayscale transformation,de-noising,and normalization of image-size.Signatures contained ink dot pollution,paper impurity pollution,and scanning mechanical pollution during the writing and scanning process.Pre-processing can effectively reduce pollution,avoiding affect the extraction of feature.3D signature pre-processing included surface correction,removal of outliers,and denoising.Paper is easy to bend,the paper surface contain impurities,and mechanical vibration will generate noise during scanning.Pre-processing can reduce the interference caused by experimental materials.(2)Extract signature features.Feature extraction is the key step of signature identification.2D features were extracted from the signature image's gray level cooccurrence matrix feature(GLCM)and local binary pattern feature(LBP),and the two features were combined to form a new texture feature,which would be used in signature verification.A method for extracting signature depth features was proposed.The signature depth matrix was extracted and the statistical features were calculated as the signature stroke features.Multiple strokes were merged as the signature features.(3)An overall framework based on texture feature and depth feature identification methods was designed,and experiments were performed on the GPDS data set and the local Chinese signature data set to verify the validity of texture features and depth features in signature verification.The experimental results showed that the GLCM and LBP features can effectively distinguish the authenticity of the signature when used alone on the locally forged Chinese signature data set.The fusion feature improved the best overall correct rate to 87.75%.The deep feature had an excellent discrimination effect on skilled forged signature data sets,reaching 97.378%,which was significantly higher than the traditional signature verification methods.