Based On Hopfield Neural Network Off-line Handwritten Form Digital Recognition Analysis And Research

The handwritten digits identification is an important component of the optical character recognition and applied extensively in real life.There are two kinds of handwritten digits identification methods which are online and offline recognition.The algorithm of online handwritten digits identification is already very mature,and the recognition rate also has a high rate,so this method is used widely.Compared to this the rate of offline handwritten digits identification is very low which is detemined by its own characteristics.The offine handwritten digits lacks the dynamic information when writen,and have verious styles,so it is difficult to recognize the digits and has high rate of mistakenly identified.But,the offline handwritten digits identification has an irreplaceable position in some special occasions,such as the recognition of postal code in postal mail sorting, the identification of handwritten digits in bank cheque.Meanwhile in real life their is a great amount of digital information for offline handwritten digits identification to deal with,so we also need high requirement data processing speed.In short the identification of offline handwritten digital research has very important significance,this paper is to improve the offline algorithm.The technology of artificial neural network is becoming more and more mature throuch the research in recently dacades.It plays an important role in the field of various kinds of pattern recognition and signal processing etc,and achieves good effect.This topic is to recognize offline handwritten digits using artificial neural network.Firstly,in the first chapter this pape introduced and illustrated the research status and significance of handwritten digits,and presented some traditional identification methods of offline handwritten digits,then given the research work of this paper.Secondly,in the second chapter did some pretreatment for the offline handwritten digits image using the method such as gray change,binary and refined.In this part,not only presented the various algorithms of binary and refined,but aslo processed the handwritten digits image using the improved integral threshold value method.On the subject of extracting image characteristics vector,we first given some common feature extraction method,then focused on the wavelet transform method which used to extract eigenvector for the processde handwritten digits image.Finally,established the hopfield neural network model using MNIST handwritten digits database.This model set network connection right and retained the neural network model.Then,did the recognition tests by adopting a group of samples for offline handwritten digits.At the end of this paper,given the simulation experiments using Matlab Compared to the simulation results,we found that the method of wavelet transform combined with hopfield can achieve recognition rate than the traditional general BP neural network, and the rate is 83%.