BanglaWriting: A multi-purpose offline Bangla handwriting dataset

Published: 19 October 2020| Version 1 | DOI: 10.17632/r43wkvdk4w.1
Dr. M. F. Mridha,


The BanglaWriting dataset contains single-page handwritings of 260 individuals of different personalities and ages. Each page includes bounding-boxes that bounds each word, along with the unicode representation of the writing. This dataset contains 21,234 words and 32,787 characters in total. Moreover, this dataset includes 5,470 unique words of Bangla vocabulary. Apart from the usual words, the dataset comprises 261 comprehensible overwriting and 450 incomprehensible overwriting. All of the bounding boxes and word labels are manually-generated. The dataset can be used for complex optical character/word recognition, writer identification, and handwritten word segmentation. Furthermore, this dataset is suitable for extracting age-based and gender-based variation of handwriting.


Steps to reproduce

The writings were conducted using regular stationery products. Writers were advised to write on a random topic. Only one page of writing was collected from each individual. The handwritings were further captured using scanners and smartphone cameras. Each captured image was cropped and annotated manually. The contains a converted dataset produced using the referred script.


Bangladesh University of Business and Technology


Image Processing, Optical Character Recognition, Handwriting Recognition, Handwriting, Segmentation