Research On Handwritten Signature Recognition Technology Based On Convolutional Neural Network

As a characteristic of human biological behavior,handwritten signatures are stable and easy to obtain,and are widely used in finance,law,and other industries with high confidentiality.However,in the past,the authenticity of handwritten signatures mainly relied on manual and traditional physical methods,which existed the problems of low accuracy and long time.This paper applies deep learning knowledge to handwritten signature recognition tasks,the recognition rate is significantly improved,and the recognition time is shortened.Finally,the above research results are applied to the handwritten signature online recognition system.At first,because of the problem of low recognition rate of handwritten signatures,this paper proposes an improved Le Net-5 model.The model selects dual channels instead of traditional single channels,which makes the number of features is increased and the extracted feature information is more accurate.In the stage of exploring model,the model is continuously optimized by changing the parameter configuration.Finally,the experimental verification of the improved Le Net-5 model improves the recognition rate of handwritten signatures.Secondly,in view of the problem of small amount of signature recognition data and slow recognition speed,this paper proposes an improved Res Net-18 model.The model introduces a spatial transformation network to eliminate the effects of tilt and distortion during writing.Introduce transfer learning and combine unsupervised training to solve the problem of sparse data and improve model performance.It is verified through experiments that the improved Res Net-18 model improves the recognition rate while reducing the data requirements and shortening the time.Finally,combining the above research results and the actual application scenario of signature recognition,the design of this paper bases on a handwritten signature online recognition system of convolutional neural network.This systemcan present an interface on Windows.The writer moves the mouse to write a signature.By calling the improved Res Net-18 model,online recognition of handwritten signatures is realized.Through the test of the online recognition system for handwritten signatures by different writers,it can be seen from the test results that the online recognition system for handwritten signatures designed in this paper has high recognition rate,fast recognition speed and has certain application value.