Verification Of Off-line Handwritten Signature Based On The Improvedneural Network

With the development of artificial intelligence technology into scientific research and daily life in various fields,the related technical methods have brought new directions to the field of forensic science.Especially recently,active in speech recognition,image processing,and other aspects of Natural Language Processing deep learning neural network technology to the handwriting identification in quite a long time in the automatic identification of the bottleneck of the off-line signature handwriting brought to a breakthrough,improve the self identification based on handwriting signature handwriting identification.The development of handwriting recognition technology and system has a long history,especially in the field of finance,security,justice and so on.However,it is difficult to make effective use of handwriting identification in the practice of forensic authentication.The automatic recognition system of off-line handwriting development process stay in the conventional pattern recognition techniques,although the conventional image processing technology progress based on artificial neural network modeling,there are still major problems,two aspects of feature selection and feature value of the hand,especially for off-line handwriting recognition,the imitative rate is low;on the other hand,part of the work also leads to artificial presupposition and fully realize the automatic identification machine,there are subjective limitations and interference.Refer to the current application of mature handwriting recognition technology and artificial intelligence technology can be found using the depth of deep learning neural network algorithm or frame as classifier.Handwriting recognition is closely related to handwriting recognition,Thedepth learning algorithm which is active in the aspects of speech,image and natural language can provide a new way to solve the problem of feature selection and value problem.This paper attempts to explain the application of the improved neural network DL in off-line signature handwriting recognition from two aspects of theory and experiment.Combined with the traditional pattern recognition principle and practical effect,put forward the demand and requirement for the automatic recognition of handwriting features not only artificial handwriting identification when feature selection types are not limited to the traditional feature point automatic recognition of handwriting morphology.Accordingly,the characteristic value problem is also changed into an adaptive model.Therefore,the paper expounds the adaptability and the theoretical advantage of the deep learning neural network on the handwriting feature and the identification principle,and changes the quantitative and qualitative identification model.The part of experiment is mainly through the PaddlePaddle deep learning architecture Baidu released in 2016,the training data input Paddle,the depth of training the neural network model,namely the convolution neural network,reduce the false rate through the iterative training and test data were adopted to test the prediction,the realization of offline handwriting identification objective to identify,and try to make a signature imitation.From theory to experiment and verification to fully elucidate the automatic recognition of the depth of the neural network algorithm based on deep learning can be achieved offline signature,and to some extent of signature recognition;at the same time to solve the problem of feature and value in the traditional method,the optimization of machine automatic identification path,improve the degree of automatic identification,off-line handwriting automatic identification of true artificial intelligence to the direction of development.