BPS: A Demographically Focused Dataset of Handwritten Bangla Primary Script for Early Writer Recognition

Bangla handwritten character recognition is challenging due to its large character set, compound forms, and style variations, and existing datasets rarely capture the handwriting of primary-level learners. To fill this gap, the BPS (Bangla Primary Script) dataset was created by collecting 24,420 grayscale images of 50 basic characters (11 vowels, 39 consonants) and 10 digits from 407 students across multiple districts in Bangladesh. The images were normalized to 32×32 pixels, organized into 60 balanced classes, and split into 60% training, 20% validation, and 20% test sets, providing a demographically representative resource for OCR and deep learning applications focused on early learners.