Bangla Sign Language Word Dataset Image based Sign Recognition

This data set was developed to facilitate the study of automatic identification of the Bangali Sign Language (BdSL) in response to the lack of publicly accessible materials for the Bangladeshi Deaf and hard-of-hearing community. It is a collection of 376 high-quality JPG pictures of 40 distinct BdSL words, including categories of actions, objects, context, feelings, status, and so forth. A still hand gesture of each word of Bangla was captured in each image and had been resized to 500 x 500 pixels to provide uniformity to the application of machine learning.

The information was gathered at the southeast university campus and residential areas of the authors at Dhaka, Bangladesh, from June 2025 to August 2025, using consumer-level smartphones (Redmi Note 7 Pro, Honor 200, Oppo A5, and POCO X3). Five participants repeated the signs according to demonstrations of references of reputable sign language teachers and Deaf actors. Several photos of each word were taken at different angles, different distances, and different lighting conditions to get the natural variations.

Each image is coded with a distinct identifier (e.g. W10101 where W is "word," the first digit is a word category, the next two digits are a word index, and the final two digits are an instance of the image). The data is hierarchically arranged within the semantic category into a folder structure that is easily reused to train, test, and benchmark recognition models. By offering one of the first organized datasets on word-level BdSL recognition, this dataset is designed to ease computer vision research, inclusive education, and assistive technology development.