A Model for Building English Training Corpora from Scanned Documentation

Abstract Related to various fields like Academia, Government Organizations, Hospitals, and Mythology, where a lot of information is available in hard copies. These documents need to get digitized to text, to make the hard copies data more accessible and maintainable. To create a paperless environment Automatic handwritten text recognition system plays a major role which is possible through digitizing and processing handwritten copies. The proposed Automatic text recognition system works on image input to generate digital output. To develop Automatic handwritten text recognition system various domains like machine learning with computer vision and deep learning techniques are required to create abstract models for recognizing letters and words initially. The proposed model uses sequence to sequence architecture to generate the sequential output of digitized lines of text. The proposed model uses Conventional Neural Network model to perform feature extraction from the handwritten image. The extracted features are modeled with a sequence-to-sequence methodology and submit to ResNet-Transformer for encoding and decoding from the Input Image of visual features and the sequence of letters. The proposed model uses synthetic data augmentation for image preprocessing. The IAM dataset which contains a large amount of data will be used for the implementation of proposed model.