A benchmark dataset for printed Meitei/Meetei script character recognition

Yanglem Loijing Khomba Khuman; Salam Dickeeta Devi; Ch Ponykumar Singh; H Mamata Devi; N Ajith Singh

doi:10.1016/j.dib.2022.108585

A benchmark dataset for printed Meitei/Meetei script character recognition

Data Brief. 2022 Sep 15:45:108585. doi: 10.1016/j.dib.2022.108585. eCollection 2022 Dec.

Authors

Yanglem Loijing Khomba Khuman¹, Salam Dickeeta Devi¹, Ch Ponykumar Singh¹, H Mamata Devi¹, N Ajith Singh²

Affiliations

¹ Department of Computer Science, Manipur University, India.
² Department of Computer Science, South East Manipur College, India.

Abstract

The Manipuri language is the official language of the Indian state of Manipur. The language belongs to the Tibeto-Burman family of languages. A benchmark Meitei/Meetei script printed document images dataset is presented in this article. The dataset contains raw 824 pages of printed documents and binarized images, text files, and XML files for each raw image. It also includes 51,460 isolated character samples, composed of 27 consonants, 7 half-consonants, 8 vowels, and 10 numerical. This dataset could be used for optical character recognition (OCR) research and in the different research areas of natural language processing (NLP).

Keywords: Meitei/Meetei script; Natural language processing; Optical character recognition; Printed document images.