Handwritten Text Recognition Based On Artificial Neural Network

With the extensive application of pattern recognition technology in the field of information science,the text recognition technology has been a critical issue,such as the handwritten document,license plates and verification code recognition,where the characters need to be identified.Because in many fields that need to automatically identify the information,the image recognition technology has a great theoretical significance and practical value.This paper studies for recognition of handwritten text with the method of Artificial Neural Network.This thesis discusses the significance of handwritten text recognition,and discusses the general process of recognizing characters,including graying,de-noising,binarization,normalization and character thinning.This thesis uses custom normalization algorithm for image pre-processing,transforming the 48×48 pixel image to normalized 16 × 16 pixel image,and then made the characteristic feature extraction pixel by pixel.After the preprocessing,the character features were converted to the input vector of the neural network.Twenty groups of handwriting samples were selected to train the BP neural network,and we take the other twenty sets of samples as a test sample input to the trained network to obtain the recognition result.The paper compared different activation functions in BP training algorithms,and finally used the BP algorithm with momentum factors and adaptive learning rate,and the experiment was conducted in the MATLAB platform,and the paper analyzed and summarized the experimental results.This study shows that the correct rate based on BP network handwritten text recognition technology is higher,and this algorithm has better anti-interference ability and deformation.In the future,it may be improved and applied in the relevant fields.