MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641078216 A) filed by Cmr Engineering College, Kandlakoyav, Medchal Road, Hyderabad, Medchal Malkajgiri, Telangana-, India. on June 24, 2026, for Attention-Driven Deep Learning Framework For Tampered Image Localization And Digital Forensics.
Inventors include Dr. T. Satya Narayana, Associate Professor, Electronics And Communication Engineering, Cmr Engineering College; Mr. K. Ramana Reddy, Associate Professor, Computer Science And Engineering, Cmr Engineering College, Kandlakoya, Hydeabad-; Mrs. Sushma Priyadarshini, Assistant Professor, Computer; Mr S Sai Bhushanam, Assistant Professor, Computer Science And; Mr. Chilaka Venkateswarlu, Assistant Professor, Computer; Mr. Y. Sampath Kumar, Assistant Professor, Computer Science And Engineering Aiml, Cmr Engineering College, Kandlakoya, Hydeabad-.; and Mrs. M. Sabitha, Assistant Professor, Computer Science And Engineering Aiml, Cmr Engineering College, Kandlakoya, Hydeabad-..
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The present invention discloses an Attention-Driven Deep Learning Framework for Tampered Image Localization and Digital Forensics for detecting image manipulations and accurately localizing forged regions. The framework is designed to address challenges associated with digital image tampering, misinformation, and content authenticity in modern multimedia environments. Digital images collected from cameras, social media platforms, surveillance systems, and forensic databases are processed through preprocessing stages including resizing, normalization, noise reduction, and data augmentation. The system integrates advanced deep learning architectures with attention mechanisms such as spatial attention and channel attention to enhance feature extraction and focus on suspicious regions within images. Deep learning models including Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) are employed to analyze image characteristics and identify manipulation artifacts. The framework generates pixel-level localization maps, forgery masks, and heatmaps to accurately highlight tampered regions and improve interpretability. Furthermore, the system classifies images as authentic or manipulated and identifies forgery types including copy-move, splicing, object removal, and synthetic image generation. The proposed framework provides visual forensic reports and confidence scores to assist investigators in digital evidence analysis. The invention achieves high localization accuracy, robustness across multiple datasets, and computational efficiency. It is applicable in law enforcement, cybersecurity, media verification, surveillance systems, judicial investigations, and digital content authentication, thereby enhancing trust and reliability in digital media ecosystems.
Disclaimer: Curated by HT Syndication.
