Offline Signature Verification Based On Deep Learning

As the most convenient way of biometrics,offline handwriting identification plays an important role in daily life.As the most common application in the field of offline handwriting authentication,offline signature authentication has been widely used in the fields of finance,commerce,military and communication authentication.Therefore,it is of great significance and practical value to identify the authenticity of signature by studying the offline handwritten signature.In recent years,with the popularization of the concept of artificial intelligence technology,various industries have tried to combine traditional industries with artificial intelligence.Among them,deep learning technology is the most noticeable technology development direction in artificial intelligence technology.Therefore,this thesis hopes to combine the latest cutting-edge deep learning technologies to bring new technological breakthroughs to offline signature authentication applications.In addition,although domestic research on offline signatures has a history of decades,we have always faced the problem of lack of Chinese signature data in the field of handwritten signature data sets,which has seriously restricted the development of our country in the field of offline signature authentication.Therefore,this thesis has done the following work from the two aspects of Chinese signature dataset and algorithm model improvement:(1)A complete signature dataset collection scheme is proposed,and a signature collection system for automatic cutting and collection of scanned files is set up,which provides a basis for large-scale collection of signature data.Through this collection process and collection system,we produced a Chinese offline signature dataset containing nearly 10,000 handwritten signatures with 100 participants.The collection and collation of this dataset laid the foundation for the improvement of offline signature authentication algorithms.(2)A two-channel convolutional neural network model is proposed to solve the offline signature identification problem.The improvement of the model is from the perspective of model lightweight and loss function.In terms of model light-weighting,the model introduces the inception structure,dilated convolution,and global pooling layer,which can reduce the amount of model parameters and improve the accuracy of the model.In the aspect of loss function,this thesis introduces the consistency loss which is commonly used in the field of semi supervision to constrain the model and improve the feature extraction ability for effective information.The experimental results show that this method has obvious advantages over the baseline model in the self-made signature dataset and the open source signature dataset BHsig260-B.(3)Design and implement a signature verification platform based on off-line signature verification algorithm.In this platform,a signature extraction algorithm is implemented to extract the signature information in the scanned file.The offline signature authentication algorithm proposed in this thesis is applied and displayed through this platform.