Offline Handwritten Signature Verification

Offline handwritten signature verification process is an authentication method using the specific rules of personal writing habits.It has a wide application prospect in the fields of justice,finance,business and government affairs as it is convenient,reliable and easily accepted by people.In recent years,many methods have been proposed in the field of offline handwritten signature verification.However,due to the high similarity between genuine signature and skilled forgery and the limited size of training database,offline signature verification is still a challenging task.This thesis proposes an offline signature verification method based on region metric fusion aiming to distinguish skilled forgeries from genuine signatures.Specifically,a convolutional neural network is adopted to extract multi-region feature from signature image.And a Siamese network is adopted to combine feature extraction with similarity measurement to realize end-to-end training in order to improve the performance.The main work of this thesis is as follows:1.Signature image preprocessing.In order to deal with the influence of background noise,slant,color of handwriting,position and size of signature,the image of signature is preprocessed.Preprocessing algorithms include background elimination,skew correction,stroke grayscale normalization and image moment normalization.2.Signature feature extraction based on convolutional neural network.Through comparative research of the classical convolutional neural networks,a network suitable for offline signature verification is designed for feature extraction combined with the DenseNet and the SE-Net.The network makes full use of the difference information between skilled forgery and genuine signature and enhances the expression ability of features by integrating deep and shallow features.3.Offline signature verification based on the Siamese network.This thesis constructs an improved Siamese network,and realizes end-to-end training by combining feature extraction and similarity measurement.Aiming at the problem that the main difference between skilled forgery and genuine signature lies in the details of stroke in signature,an offline signature verification method based on region metric fusion has been proposed.In this method,the signature image is divided into several signature regions to measure the similarities,and then the similarities are fused to improve the verification performance.In this thesis,the performance of proposed method is tested on the internationally opened database GPDS and CEDAR.And experimental results show the effectiveness of the proposed method.