| dc.description.abstract |
Accurate student identification is a critical aspect of public and recruitment examinations in Bangladesh, often involving handwritten identification numbers (e.g., roll or registration numbers). Current automated systems rely on filling circles corresponding to digits but are prone to errors from incorrect circle filling by students. To address this issue, this project explores the use of handwritten digit recognition (HDR) technology for cross-verification of student identification numbers. Traditional HDR datasets, such as MNIST, predominantly feature Western handwriting styles, leading to a gap in recognizing the diverse writing patterns of Bangladeshi students. Furthermore, such datasets focus on clean, isolated digits, while real-world answer scripts present challenges including low-resolution scans, noise, and clutter from additional markings.
This study aims to bridge these gaps by creating a comprehensive dataset of handwritten English digits from Bangladeshi students, and by evaluating machine learning algorithms for accurate recognition in this specific context. A prototype system will be developed to cross-check student identification numbers, improving the accuracy and reliability of student identification in examinations. The results of this project are expected to provide insights into machine learning-based handwritten digit recognition in challenging contexts, and facilitate future research by offering a new, contextually relevant dataset. |
en_US |