MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621079616 A) filed by Symbiosis International Deemed University on June 27, 2026, for System And Method For Explainable Multi-Domain Detection Of Ai-Generated Digital Media.
Inventors include Aadya Rastogi; Atherva Srivastava; Deshna Tendulkar; and Dr. Archana Chaudhari.
The application for the patent was published on August 14, 2026, under issue no. 33/2026.
Abstract: ABSTRACT SYSTEM AND METHOD FOR EXPLAINABLE MULTI-DOMAIN DETECTION OF AI-GENERATED DIGITAL MEDIA The present invention provides a system (100) and method for detecting artificially generated digital media through spatial-spectral feature fusion. The system (100) comprises a dual-branch architecture wherein a spatial feature extraction module (120) employing an EfficientNet-B2 backbone (122) extracts texture and semantic features while a frequency domain feature extraction module (130) employing Fast Fourier Transform derives spectral statistical features. A feature fusion module (150) concatenates and processes both feature vectors through a multilayer perceptron with layer normalization and dropout. A classification module (170) produces authenticity likelihood scores. An explainability module (180) employing Grad-CAM generates heatmap visualizations highlighting suspected manipulation regions. The system achieves approximately 95 percent accuracy and 0.97 ROC-AUC, generalizing across GAN and diffusion generators with robust performance against compression, blurring, and noise. The invention enables transparent forensic analysis for media verification applications. [
Disclaimer: Curated by HT Syndication.
