MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202621054338 A) filed by Symbiosis International Deemed University on April 28, 2026, for Deepfake Detection System Using Convolutional Neural Networks For Digital Forensics.
Inventors include Anjali Singh; Rashmi Kadu; Rohini Bharne; and Dr. Priya Dasarwar.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: ABSTRACT DEEPFAKE DETECTION SYSTEM USING CONVOLUTIONAL NEURAL NETWORKS FOR DIGITAL FORENSICS The present invention provides a deepfake detection system (100) for digital forensics applications comprising a data acquisition module (110), preprocessing unit (120), feature extraction module (130), classification engine (140), model repository (150), and user interface module (160). The system performs face detection, landmark identification, alignment normalization, and resolution standardization to 224x224 pixels. Convolutional neural networks extract hierarchical spatial features through Conv2D layers (132), MaxPooling layers (134), and dense layers (138). Binary classification with sigmoid activation generates authenticity predictions with confidence scores. The system achieves 96.28 percent detection accuracy through optimized CNN architecture trained with Adam optimizer and binary cross-entropy loss. Multiple deep learning architectures including Xception, DenseNet-121, and EfficientNetB0 enable comparative evaluation and transfer learning. A Flask-based web interface provides accessible media verification for non-expert users in forensic investigations. [
Disclaimer: Curated by HT Syndication.
