GHCR-A dataset for Grantha handwritten character recognition.

Area of Science:

• Computer Science
• Linguistics
• Digital Humanities

Background:

• The Grantha script, historically significant in South Indian languages, lacks sufficient digital datasets for research.
• Existing resources for Grantha character recognition are limited, hindering advancements in related AI applications.

Purpose of the Study:

• To introduce a comprehensive dataset of handwritten Grantha characters (numbers and vowels).
• To provide a benchmark resource for developing and evaluating machine learning models for Grantha character recognition.
• To facilitate research in Indian languages connected to the Grantha script.

Main Methods:

• Collected handwritten samples of 10 Grantha numbers and 34 vowels from diverse age groups.
• Digitized and preprocessed 5852 images, including segmentation, resizing, and grayscale conversion.
• Organized data into image and CSV formats with corresponding labels for machine learning.

Main Results:

• The final dataset contains 1330 samples for numbers and 4522 samples for vowels.
• The dataset comprises 44 distinct Grantha characters, with 133 handwritten samples per character.
• Data is available in both image and CSV formats, suitable for immediate use.

Conclusions:

• This dataset fills a critical gap for Grantha script research, particularly in numeral and vowel recognition.
• It serves as a foundational resource for machine learning initiatives focused on Grantha-influenced Indian languages.
• The availability of this dataset is expected to spur further innovation in script recognition and natural language processing.