Bangla character recognition for Android devices

Abstract

In this paper, we illustrate our attempt to create editable documents from images by retrieving the text. The process is widely known as Optical Character Recognition (OCR). We have tried to build an Android application for detecting Bengali characters. Previously, several attempts have been made in developing a Bengali OCR. However, there were a few limitations which drove us to work on this project. In order to recognize more characters and joint letters, we decided to work on reducing the error rate to preserve more texts. To serve our purpose, we found the Tesseract OCR engine and Leptonica Image Processing Library to be the best option. Tesseract is used in order to recognize the characters and Leptonica is used to build an Android application by extracting data from the text. We are using the Tesseract 3.03 version currently available to work on this project. Moreover, we demonstrate how we obtained better results by manipulating Tesseract along with Serak to create box files and trained data. In addition to that, we discuss how we dealt with joint letters, dangerous ambiguity and contrast issues in order to increase efficiency. Furthermore, we explain our analyzed data, our progress and the future scopes of improvement.