Machine Learning Tensor Flow Based Platform for Recognition of Hand Written Text

Abstract

Handwritten text recognition in a computer is an automatic mode that transcribe documents. Mainly two types of approaches are used in handwritten text recognition are artificial neural network (ANN) and hidden markov model. Here handwritten text recognition system is based on artificial neural networks. Feature of input images are improved by pre-processing method. After that for classifier, problem is simplified. For increasing the size of data set, data augmentation and also contrast normalization methods are used. To improve the feature of input image, classifiers are used a convolutional neural network. Information are propagated by recurrent neural network. Probability distribution of the character on every position of image are contained by matrix of the recurrent neural network outputs. The function of connectionist temporal classification is decoded the matrix into the final text. In the decoded text, to spelling error in final text post processing accounts are also present, from this accuracy of handwritten text is improved, after that it can be used in different fields like security system etc. Almost all the institutes and governments having large amount of handwritten papers are created every day. This insistent use computers to interpret handwritten texts, also create that is searchable and editable. Therefore, handwritten recognition became extremely research matter with the huge number of applications. It is resourceful to break up complex problems and reduce in extent of human action by change over the handwritten text documents into digital form.