International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 03 | Mar 2025
p-ISSN: 2395-0072
www.irjet.net
Unmasking the Deception: An Image Manipulation Detection using ResNet and UNet Architecture Onasvee Banarse1, Harsh Shah2, Kaustubh Kabra3, Akash Mete4, Dr. K. S. Wagh5 1,2,3,4 Student, Dept. of Computer Engineering, All India Shri Shivaji Memorial Society’s Institute of Information
Technology Pune, India
5Professor, Dept. of Computer Engineering, All India Shri Shivaji Memorial Society’s Institute of Information
Technology Pune, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Images from digital cameras have been used in
image manipulation. It has become difficult to distinguish between real photographs and modified images due to the accessibility of image editing software and the expansion of manipulated images online. It is possible to employ image alteration to propagate false information, trick people, or damage someone’s reputation. It has become more crucial than ever to create effective methods for detecting and flagging image tampering, as the need to defend against the dissemination of false information, protect people’s reputations, and guarantee the integrity and authenticity of the photos used as evidence in various judicial procedures has increased. The rising problem of manipulated photos can be solved by building an image manipulation detection application, which can help to create a more secure and reliable online environment.
an increasing variety of applications because of the extraordinary rise that digital photography has seen in recent decades. There are several pieces of software available now that can be used to alter images to make them appear as they did in the original. Pictures are utilized as authenticated proof of any crime; thus, it will be problematic if they stop being genuine. These kinds of forgeries are difficult to detect now. It might be difficult to tell whether a digital image is authentic or has been altered. It might be difficult to spot signs of tampering in digital images. In this research paper, a novel approach is proposed for detecting image manipulation using a combination of ResNet101 and Unet neural networks. The need for such techniques has become paramount due to the rise in manipulated images in digital media. A literature survey of more than 20 research papers on image manipulation detection was conducted, which revealed that deep learning techniques, specifically CNNs, have shown promising results in identifying manipulated images. The proposed methodology involves using ELA for feature extraction, followed by Unet for pixel-wise segmentation of the manipulated area. The approach was tested on the CASIA dataset and achieved overall accuracy of 93.30%. With the help of ELA preprocessing technique, the binary predicted mask, predicted mask, and numerous statistical graphs on evaluation metrics are generated. Hence, the proposed approach can be used as a reliable tool for identifying image manipulation, thereby mitigating the spread of manipulated images in digital media.
Fig -1: Examples of tampered images from different datasets. The names of dataset are mentioned at the bottom.
Key Words: Transfer Learning, Image Segmentation Model, ResNet-100, Unet, Image Manipulation Detection.
Digital image manipulation is now simpler than ever thanks to the expanding range of tools and software available for it. Image modification has the potential to be exploited for deceit, propaganda, and other harmful ends, even while it can have genuine uses in art, entertainment, and advertising. As a result, there is a rising need for tools and methods that can identify and stop picture alteration, especially in industries like journalism, forensics, and security where image authenticity is crucial.
1.INTRODUCTION Artificial intelligence (AI), and more specifically image processing, has made impressive strides in recent years. AI models can now complete a variety of complicated image processing tasks, like picture recognition, segmentation, and restoration, with remarkable accuracy and speed thanks to the development of deep learning techniques. These developments have also raised the possibility of
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 595