Identification of Printed and Handwritten Text Regions from FORM Document Using Morphological Transform Based Feature Set with CNN

Extraction of text from document images generally focuses on either printed text or handwritten text. In this world there are various types of documents like forms, money receipts, challans, cheque books where coexistence of handwritten text and printed text are very common. Thus it is very important to separate handwritten text from printed text with suitable text recognition techniques to formulate a digitized version of such documents. This paper describes a novel method to identify text regions from FORM documents. We have empirically studied different approaches to obtain bounding boxes for our own dataset. Gradient-based morphological approach gives better performance to draw the contour of text regions. Classification of extracted text images as printed or handwritten is done by Neural Network Models.