AI recognizes writing on severely damaged South Indian palm leaves

A system tested on 1,250 historical folios reported high accuracy while identifying authentic manuscripts and separating writing from wood fibers.

Editorial illustration — not from the study.

EpigraphiX-AI combines image processing, character recognition and manuscript authentication tools for historical palm leaves written with incised styluses. The system is designed to distinguish shallow incisions from fiber patterns, decay and uneven surfaces, including in complex Grantha writing.

How well the system read

In tests on 1,250 historical palm-leaf folios, the system reported a Word Accuracy Rate of 97.4% and Character Accuracy of 98.6%. Its reported Character Error Rate was 1.4%, with precision of 98.6%, recall of 98.4% and specificity of 99.4%. The manuscript-authentication component reported 99.4% accuracy. The system was also described as rejecting non-manuscript images, including portraits, indoor and outdoor scenes, and digital interfaces.

Why damaged folios matter

South Indian palm-leaf manuscripts preserve texts in Sanskrit, Grantha and Old Malayalam on material that can be difficult for conventional optical character recognition systems to interpret. A tool that can identify authentic folios and recognize writing despite decay, wood fibers and uneven incisions could support the digitization and study of these historical records.

Evidence and open questions

The findings come from an experimental evaluation reported in a journal article using an archival corpus of 1,250 historical folios. The abstract does not give details about the corpus's source, language distribution, annotation process, train-test separation or the exact state-of-the-art systems used for comparison. It also does not report how well the system performs on manuscripts outside this corpus, or provide separate results for its translation and grammar components.

// Source

International Journal for Research in Applied Science and Engineering Technology · 2026 · DOI: 10.22214/ijraset.2026.84707

View paper (DOI)Open access versionPublished 2026-08-27

Authors: Adarsh S

Fine-particle levels around cement plants were above the WHO annual guideline

A meta-analysis found that fine-particle concentrations around cement plants had a weighted average of 20.76 micrograms per cubic metre, above the World Health Organization’s annual guideline of 5 micrograms per cubic metre. The researchers cautioned that the studies differed substantially, including in how long they sampled the air.

7 hours ago·1 views

AI & Computing

A drone-mounted camera recognized American Sign Language in early tests

Researchers developed a drone-based system that recognized American Sign Language gestures with 98.63% accuracy on a combined dataset. The system is designed to support sign-to-speech and speech-to-sign communication where fixed cameras or other communication infrastructure are unavailable.

1 day ago·5 views

Abstract editorial illustration accompanying the article: Why “Birds have hollow bones” can teach broad facts despite different interpretations

AI & Computing

Why “Birds have hollow bones” can teach broad facts despite different interpretations

People can interpret the same general statement in different ways, yet generic language may still pass on broad knowledge. Across three experiments, a model in which such statements have an uncertain threshold best explained participants’ varied interpretations of new generic claims.

4 days ago·28 views