Handwritten Documents Text Recognition with Novel Pre-processing and Deep Learning

Data being the most valuable resource on earth today, there is a pressing need to transform all data digitally. The world has been historically and still in multiple sectors operating based on handwritten text. For example, Handwritten taxation data and calculations, claim forms, doctor’s prescription, legal documents, resumes, financial documents, accounts sheets, and many more. Handwritten data can be very easily misinterpreted even by human personnel. Hence, the most critical challenge that remains is the transformation of handwritten documents. Researches from different authors present an application wherein they take in word crop images generated from a document manually and extract the text from it. In our thesis, we present an end to end solution wherein a text image is uploaded, and handwritten text from the entire image is extracted as-is. The three main contributions of our thesis are 1. Improved text localizer trained on the various handwritten and printed dataset, 2. Novel classification model to segregate printed and handwritten words, 3. Effective image preprocessing techniques applied to handwritten word crops to make them eligible to be fed to the deep learning model for improved overall accuracy.