Intellectual Property India Publishes Patent Application for Deterministic Topology-Preserving Edge-Ai System And Method For Offline Recognition, Reconstruction, And Semantic Decoding Of Highly Degraded Epigraphic Scripts Using Persistent Homology-Guided Quantized Attention-Based Glyph Tokenization

MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202641070289 A) filed by Rajith Kumar B K on June 05, 2026, for Deterministic Topology-Preserving Edge-Ai System And Method For Offline Recognition, Reconstruction, And Semantic Decoding Of Highly Degraded Epigraphic Scripts Using Persistent Homology-Guided Quantized Attention-Based Glyph Tokenization.

Inventors include Dr. H. S. Mohana; Rajith Kumar B K; and Shreedevi P.

The application for the patent was published on June 12, 2026, under issue no. 24/2026.

Abstract: An offline Edge-AI system and method for recognizing and reconstructing highly degraded epigraphic scripts are disclosed. The system comprises a multispectral capture sensor, an image acquisition module, a topology preservation engine, a persistent homology processor, a persistence signature generator, a spatial graph generator, a graph-based glyph tokenizer, a topology-aware attention module, a quantized transformer engine, an integer scaling engine, on-die SRAM memory, and a dual decoder module. Persistent homology analysis identifies stable topological structures within degraded inscription images and generates persistence signatures representing structural stability across filtration scales. A topology-preserved spatial graph representation of fragmented glyph structures is generated and transformed into structural glyph token sequences. The token sequences are processed using topology-aware quantized transformer inference employing low-bit integer arithmetic and SRAM-based computation. Persistence signatures are incorporated into attention weighting to improve reconstruction accuracy under severe degradation conditions. The system reconstructs missing glyph structures and simultaneously generates machine-readable Unicode text outputs. The invention operates entirely offline on resource-constrained edge hardware without requiring cloud connectivity and is particularly suitable for recognition, reconstruction, and preservation of degraded historical inscriptions and epigraphic scripts.

Disclaimer: Curated by HT Syndication.