Dataset of Hand written: Five English Words in Capital-Letter

Description

This dataset of 3,000 offline handwritten word images collected from 100 student participants of Nigerian nationality. Each participant hand-wrote five target English words – GAME, STAND, STOP, QUEST, and GIVE – in capital (block) letters, producing six repetitions of each word, for a total of 30 handwriting samples per subject (100 subjects × 5 words × 6 repetitions = 3,000 images). Handwritten sheets were digitised and segmented into individual word images, each saved as a JPEG file of constant width (287 pixels) and near-constant height (150–152 pixels). The images are organised in a three-level folder hierarchy by subject identifier (001–100), target word (word index 1–5), and repetition number (1–6), which supports both writer-independent and writer-dependent experimental protocols. The dataset provides a controlled-vocabulary resource for training and benchmarking capital-letter handwriting/word recognition and writer identification and verification systems, and, to the authors' knowledge, is among the few publicly documented handwriting datasets sourced specifically from a Nigerian student population, supporting research into the demographic and geographic generalisability of handwriting recognition models.

Steps to reproduce

This article describes a dataset of 3,000 offline handwritten word images collected from 100 student participants of Nigerian nationality. Each participant hand-wrote five target English words – GAME, STAND, STOP, QUEST, and GIVE – in capital (block) letters, producing six repetitions of each word, for a total of 30 handwriting samples per subject (100 subjects × 5 words × 6 repetitions = 3,000 images). Handwritten sheets were digitised and segmented into individual word images, each saved as a JPEG file of constant width (287 pixels) and near-constant height (150–152 pixels). The images are organised in a three-level folder hierarchy by subject identifier (001–100), target word (word index 1–5), and repetition number (1–6), which supports both writer-independent and writer-dependent experimental protocols. The dataset provides a controlled-vocabulary resource for training and benchmarking capital-letter handwriting/word recognition and writer identification and verification systems, and, to the authors' knowledge, is among the few publicly documented handwriting datasets sourced specifically from a Nigerian student population, supporting research into the demographic and geographic generalisability of handwriting recognition models.

Institutions

• University of Nigeria

  Enugu State, Nsukka
• Air Force Institute of Technology Kaduna

  Kaduna State, Kaduna
• Nile University of Nigeria

  FCT, Abuja
• University of South Africa

  Gauteng, Pretoria

Categories

Licence