MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641099569 A) filed by Ms. M Thanmayee; Mr. P. Chandrashekar; Mrs Kunkulagunta Naga Maha Lakshmi; Ms. P. Kavitha; Mr. Srikanth. B; Mrs. Nelli Chandrakala; Mr. Syed Ashraf; Mr. V. Gopi Krishna; Mr. T. Pawanteja; and Mr. Shaik Ahmad on August 18, 2026, for System And Method For Smart Image Forgery Detection Using Combined Lightweight Artificial Intelligence Models.
Inventors include Ms. M Thanmayee; Mr. P. Chandrashekar; Mrs Kunkulagunta Naga Maha Lakshmi; Ms. P. Kavitha; Mr. Srikanth. B; Mrs. Nelli Chandrakala; Mr. Syed Ashraf; Mr. V. Gopi Krishna; Mr. T. Pawanteja; and Mr. Shaik Ahmad.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: Abstract The present invention discloses a system and method for smart image forgery detection using combined lightweight artificial intelligence models. The invention provides a unified end-to-end framework capable of detecting copy-move and splicing forgeries in digital still images, classifying images as Authentic, Copy-Move, or Spliced, and localizing tampered regions at the pixel level. The framework incorporates multiscale feature extraction using an EfficientNet-B3 backbone, internal self-correlation analysis for copy- move detection, a lightweight Feature Pyramid Network decoder for tamper localization, consistency-based contrastive training, balanced class sampling, and probability calibration for reliable forensic confidence scoring. Evaluated on VISION, DEFACTO, CASIA v2, and CoMoFoD datasets, the system achieves AUROC of 0.9830 and 0.9891 for copy-move and splicing detection respectively, with calibration reducing the false positive rate from 0.0416 to 0.0284.
Disclaimer: Curated by HT Syndication.
