pranita-mukhia/NHD-100

NHD-100: Nepali Handwriting Dataset

Description

NHD-100 is a Nepali offline handwriting dataset collected for research on writer identification and handwriting analysis. The dataset contains handwritten Nepali samples from 100 writers, with 6 handwritten pages collected from each writer, resulting in 600 handwritten pages in total.

The handwriting is written in the Devanagari script and contains a variety of Nepali words, characters, conjuncts, and numerals.

Dataset Structure

NHD-100/
|
|-- P-001/
|   |-- page_01
|   |-- page_02
|   |-- page_03
|   |-- page_04
|   |-- page_05
|   `-- page_06
|
|-- P-002/
|   |-- page_01
|   |-- ...
|   `-- page_06
|
|-- ...
|
|-- P-100/
|   |-- page_01
|   |-- ...
|   `-- page_06
|
|-- splits/
|   |-- train/
|   |-- validate/
|   `-- test/
|
`-- metadata.csv

Each writer is identified using a unique ID from P-001 to P-100.

What the Dataset Contains

• —100 writers
• —600 handwritten pages
• —6 pages per writer
• —Approximately 60–100 words per page
• —Nepali language written in Devanagari script
• —Handwritten samples covering different writing styles
• —All 36 Devanagari consonants
• —All 12 matras
• —More than 40 conjunct characters
• —All 10 Nepali numerals
• —10 text categories (C1–C10) with 20 writeups per category

The original data collection form contained seven pages. The first page was used for writer information, while the remaining six pages contained handwritten Nepali text. The first three handwriting pages used common passages, while the remaining three used individual writing prompts.

Demographic Information

The dataset was collected at St. Joseph's College, Darjeeling, India.

Data Splits

The dataset provides separate training, validation, and testing splits at the writer/page level for machine-learning experiments.

License

The NHD-100 dataset is released under the CC BY 4.0 license.