MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641098012 A) filed by Chennai Institute Of Technology on August 13, 2026, for Detection Of Personally Identifiable Information In Government Documents.
Inventors include Dr. E. Kodhai; Mrs. Rajeswari C; Mrs. Sumathi S; Mrs. Vidhyalakshmi P; Mr. Rajeeth Prabhu S; and Mr. Jasirullah N.
The application for the patent was published on August 21, 2026, under issue no. 34/2026.
Abstract: The invention describes a robust, hybrid PII detection and anonymization framework for use in government document workflows. The six-stage sequential pipeline is composed of Regular Expression detection of structured identifiers such as Aadhaar, PAN, passport, voter ID, email and phone address stored in a global multi- geographic JSON schema, NER detection of unstructured PII such as person, geo and org names using NLP, checksum validation using the Verhoeff Algorithm for Aadhaar and the Luhn Algorithm for credit card numbers to eliminate false positives, contextual word analysis using proximity-based keyword analysis, POS tagging, and embedding-based similarity to disambiguate sensitive from non-sensitive entities, and anonymization through redaction, pseudonymization or generalization of validated PII. The system employs an OCR subsystem based on pytesseract which runs on adaptive pre-processing of grayscale conversion, adaptive thresholding, and deskewing on the input of scanned and photographed document images. For the task of detection of facial biometric PII in document images, the system utilizes a Haar cascade classifier. When classifying document types, the system incorporates a confidence scoring scheme for PII classification to reduce false positives across the non- sensitive text domain. The system was tested on 250 randomly sampled government documents (PDFs, text, and PDF-based images, etc.) and was able to identify structure and semi-structure PII semi-reliably. The modular system architecture is conducive to the addition of future transformer-based named entity recognition (NER) models, developing of an explainable AI system, and installing region-based OCR engines.
Disclaimer: Curated by HT Syndication.
