Performance Analysis of OCR Engines on the Bangla-CrossHair Dataset

Introducing a novel dataset for testing OCR Engines

Benchmarking OCR Engines for Bangla Character Recognition

We study the performance of the top OCR engines on a Bangla characters dataset called the Bangla CrossHair Dataset introduced by our team. The dataset is available on Kaggle. The paper has been submitted to ICCIT 2024 Conference.

Kaggle

The dataset is available on Kaggle. The code will be made public upon publication of the paper.

Python

Python libraries were used to perform the benchmarking.

Easy OCR

EasyOCR is a modern, deep learning-based OCR library developed by the Jaided AI team. It is built on top of deep neural networks and uses pre-trained models for detecting and recognizing text in various scripts and languages.

Tesseract OCR

Tesseract is one of the most widely-used traditional OCR engines. It was initially developed by HP and is now maintained by Google. Tesseract is a rule-based engine combined with machine learning techniques for recognizing text.

Motivation

A brief history

While there is a lot of work on OCR in Bangladesh, due to the lack of a proper understanding of the subject matter, no comparative study on OCR engines has been done at all until this research project which led to the realization of this research project. A novel dataset titled: Bangla-CrossHair was introduced by our team in order to test the OCR engines. This research project was done in collaboration with InteX Research Lab.

Where we are today

Currently, we are serving different roles individually. You can learn more about our team below.

Team

Hierarchy

The team was led by A.N.Chowdhury, who during the time of this work was collaborating as a Research Assistant with M.S.R.Kohinoor, the founder of InteX Research Lab. A.A.Sami contributed in every aspect of the process, while contribution to the dataset was done by S.Absar, S.M.P.Mamun, and F.R.Biswas.