HSC-IndicDB: A Hierarchical Multi-Task Dataset for Isolated Character and Numeral Recognition Across Six Indic Scripts

Published: 8 October 2026| Version 1 | DOI: 10.17632/dkmxgd8n79.1
Contributors:
,
,

Description

The dataset contains isolated handwritten and paleographical character samples across six Indic writing systems (Devanagari, Sharada, Nandinagari, Malayalam, Bengali, and Kannada), spanning Northern Nagari and Southern Dravidian lineages. The data was gathered from digitized historical manuscript folios, archival paper documents, and standardized handwritten transcription forms contributed by a diverse cohort of scribes and writers across different age groups. The document folios underwent high-resolution scanning, automated isolated glyph extraction, bounding-box cropping, and adaptive Sauvola binarization to mitigate substrate degradation, ink bleed, and background artifacts. After manually inspecting the corpus and discarding fragmented strokes, obscured contours, and illegible scribbles, the finalized benchmark repository comprises 36,152 digitized glyph images. The entire collection is organized into a clean directory hierarchy, sorted systematically into respective parent folders corresponding to each script lineage and script type, and subdivided into individual subfolders labeled by character and numeral class.

Files

Institutions

Categories

Computer Vision, Optical Character Recognition, Machine Learning, Deep Learning

Licence