MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641084287 A) filed by Peri Institute Of Technology on July 09, 2026, for Ai-Based Image Forgery Detection System.
Inventors include Sanjai M; Adithya S; Hemanth V; Sarathy K; Amirtha P; and Suryaswathi J.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: The present embodiment is directed to ah Artificial Intelligence (Al)-based intelligent image forgery detection system that can detect images that have been digitally manipulated by a combination of deep learning, digital forensic analysis and adaptive feature intelligence. Conventional verification methods are no longer effective due to the rapid development of image editing software and generative Al technologies, which have significantly broadened the realm of highly realistic fakes. However, with the rapid development of image editing software and generative Al technologies, traditional verification methods are no longer effective, as highly realistic fakes are now becoming more common. Current approaches are mainly based on metadata inspection or watermark verification, and manual forensic examination, which are all possible to circumvent or tamper with. The proposed invention provides an integrated multilevel image authentication framework so as to analyze the visible and invisible forensic traces without relying on metadata. It unifies Error Level Analysis (ELA), sensor noise pattern extraction, illumination consistency verification, texture anomaly detection, edge discontinuity analysis, compression artifact evaluation and adaptive pixel-distribution learning in a single deep learning architecture. The invention includes an Adaptive Multi-Forensic Fusion Engine which assigns confidence weights to various forensic features according to the quality of the image, the ratio of compression, the editing method and the acquisition device. The extracted forensic signatures are then fed into a hybrid Convolutional Neural Network (CNN) with Vision Transformer (ViT) attention networks for increasing localization and classification accuracy. The invention also creates a heat map of forgery confidence at a pixel level, allowing investigators to determine areas that may have been manipulated and their degree of confidence that the image has been tampered. The system continually refines its performance with incremental learning from new manipulation techniques, without the need for retraining the model. The proposed system achieves high detection accuracy for copy-move forgery, image splicing, object insertion, object removal, Al-generated image modification and deep image editing. The invention can be used in digital forensics, journalism, cybercrime investigation, insurance claim verification, law enforcement, social media authentication, judicial evidence validation, government security agencies, and others.
Disclaimer: Curated by HT Syndication.
