Image Forgery Detection Using Deep Learning-Based Stable Keypoint Feature Extractor and Multiscale Caswide Residual Networks
Year: 2024
Type: article
Abstract: Today, when the advancement of manipulation techniques has great speed, the ability to detect the image forgery becomes extremely important for maintaining the probity and reliability of the digital content. In this work, we propose a Deep Learning (DL) framework composed of a Stable Keypoint Feature Extractor to effectively select robust features, and Multiscale CASWide Residual Networks for accurate and dependable classification. Additionally, the Stable Keypoint Feature Extractor designed to reliably detect forgery prone regions, and the multiscale architecture to learn intricate global and local patterns for better forgery type detection encompassing copy-move, splicing and AI generated manipulations. The system is found to be resilient to noisy and compressed speech, compression artifacts, and various geometric transformations and in doing so, outperforms current methods in terms of robustness and scalability. Its efficiency is validated by extensive experimental results on benchmark datasets to achieve improved accuracy with reduced false positives. Proposed methods achieve 91.45 % accuracy in feature selection and 92.46% accuracy in classification. These results demonstrate the promise of using this adaptable and computationally efficient framework to deploy digital forensics, content authentication, and real time forgery detection in security critical applications.
Author: X. Tubax
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7218001647
