Handwritten Text Recognition and Conversion Using Convolutional Neural Network (CNN) Based Deep Learning Model

Abstract

Recognition of handwritten documents has been an attractive area of research for the past few years. It is always intended to convert a piece of handwritten information into digital text for sharing or saving without typing the information manually. The proposed model takes a picture of a handwritten text as input and converts it into digital text. The Convolutional Neural Network (CNN) is used to study the features of similar objects from multiple image samples and to classify them. Since the text is sequential data, Long Short Term Memory (LSTM), an extension of Recurrent Neural Networks (RNN) with a longer memory is used. To deal with different placements of the text in the image, Connectionist Temporal Classification (CTC) loss is employed. The IAM Handwriting Database containing handwriting samples from over 600 writers and images of over 100,000 words is used for training. After training for multiple epochs, the model registered 94% accuracy and a loss of 0.147 on training data and 85% accuracy and a loss of 1.105 on validation data.