Filled-in Handwritten Data Extraction from Documents based on Geometrical Features

Abstract

Filling out forms is one of the oldest and widely used methods for collecting information in different fields from an applicant. Automating the scanning process for large volume of office data for example, bank cheque, aadhar card forms, driving license forms, new account opening form etc. not only escalates the office productivity but also reduces time consumption. A simple and systematic approach has been proposed based on geometrical features of the form, to extract handwritten data. The system is first fed with some geometrical and field details from an empty form, it then recognizes different fields within filled-in form using normalized correlation coefficient techniques and then estimates relative position of handwritten data. The handwritten data obtained is then processed with Hough Transform for removing unnecessary lines occurred due to bounding boxes, created to direct users for data entry area in blank form. The experiment was conducted over four different kind of forms with more than 120 filled forms, 30 instances of each type. An average accuracy of 93.82% has been recorded. Thus system extracts handwritten data efficiently.