Image Forgery Detection Using Transfer Learning

Image Forgery Detection Using Transfer Learning

Name :- Abbas Asif Tisekar                                                   Name :- Mahesh Ratnakar Yewate

Mail-id:-abbastisekar27@gmail.com                                     Mail-id:-maheshyewate2020@gmail.com

Department of Computer Engineering                                  Department of Computer Engineering

Name :- Prajakta Prakash Hande                                            Name :- Dnyanashree Kishor Palve

Mail-id:-prajaktahande713@gmail.com                                 Mail id:- dnyanashreepalve2533@gmail.com

Department of Computer Engineering                                   Department of Computer Engineering

Guide Name: Prof. Apeksha Pande

SIDDHANT COLLEGE OF ENGINEERING SUDUMBARE, TAL- MAVAL DIST-PUNE – 412109.

----------------------------------------------------------------------***--------------------------------------------------------

Abstract: The increasing prevalence of image manipulation technologies has raised significant concerns about the authenticity of digital media, making image forgery detection an essential task in various fields such as digital forensics, media verification, and legal investigations. Despite the advances in image manipulation techniques, detecting forgeries remains a challenging problem due to the sophisticated nature of modern editing tools. In this research, we address this challenge by combining traditional image analysis methods with state-of-the-art deep learning techniques to improve the accuracy and robustness of forgery detection systems. Specifically, we propose a hybrid approach that integrates the Error Level Analysis (ELA) method with transfer learning.

 The ELA method, a well-known image forgery detection technique, helps identify inconsistencies in error levels across different regions of an image, which is a characteristic of tampered images. However, traditional ELA methods may struggle to detect advanced forgeries, particularly those made with modern editing software. To overcome these limitations, we employ transfer learning by fine-tuning pre-trained convolutional neural networks (CNNs) on the CoMoFoD and CASIA datasets. These datasets are widely used in the domain of image forgery detection, containing both genuine and tampered images. By leveraging the power of transfer learning, we can enhance the detection capabilities of the model, allowing it to learn from large-scale, real-world datasets and detect subtle forgery traces that might be overlooked by classical methods.

Our experimental results show that the hybrid approach significantly outperforms traditional ELA techniques, with notable improvements in detection accuracy and robustness. The model achieved higher precision, recall, and F1-score when compared to baseline methods, demonstrating its effectiveness in identifying forged images across various manipulation types, including copy-move, image splicing, and more. These results underscore the potential of combining classical techniques like ELA with modern deep learning methods to create a more powerful and reliable forgery detection system.

Key Words: Image Forgery Detection, Image Tampering, Convolutional Neural Networks (CNN), Copy-Move Forgery Detection (CMFD), Splicing Forgery Detection, ELA