MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641073812 A) filed by Vellore Institute Of Technology on June 14, 2026, for Adaptive Semantic Forensic System For Generator-Agnostic Detection Of Ai-Generated Images.
Inventors include V Ezhil Arasi; and Pooja S.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: ADAPTIVE SEMANTIC FORENSIC SYSTEM FOR GENERATOR-AGNOSTIC DETECTION OF AI-GENERATED IMAGES The present invention relates to a computer-implemented adaptive semantic forensic system for generator-agnostic detection of AI-generated images. The system receives and preprocesses an input digital image and applies controlled perturbations using an adaptive perturbation controller regulated by semantic stability signals. Semantic embeddings are extracted from perturbed image instances using a vision-language representation model, and evolution of the embeddings is analysed through semantic drift trajectory modelling in a high-dimensional representation space. An Entropy-Weighted Embedding Stability Synthesis Index (ESSI) is computed to quantify semantic instability behaviour associated with AI-generated images. The system further performs multi-scale semantic stability analysis using global, region-level, and patch-level embeddings. A self-validation mechanism performs counter-perturbation-based verification for improving classification consistency and confidence. The system generates a forensic authentication output classifying the image as authentic, AI-generated, or inconclusive under zero-day generator conditions.
Disclaimer: Curated by HT Syndication.
