AI Powered Fake Image Detection System using ResNet and U-Net

Abstract

Nowadays, it is very easy to edit images and create fake photos using advanced tools and deepfake technology. This creates problems like spreading false information and security risks. In this project, we developed an AI-based system to detect fake images, mainly for use on social media. The system uses ResNet to understand important features of an image and U-Net to find exactly which part of the image is edited. It not only detects whether an image is fake but also shows the modified area clearly. The model is designed to run on a Raspberry Pi, making it lightweight and suitable for real-time use. This system can be useful in areas like social media monitoring, document verification, and cybercrime investigation

Introduction

This text describes a deep learning-based system for detecting and localizing manipulated (fake) images, motivated by the growing problem of image forgery and deepfakes on social media. Since traditional detection methods are no longer effective against modern image editing techniques, the proposed solution uses AI to improve accuracy and reliability.

The system combines ResNet for image classification (to determine whether an image is real or fake) and U-Net for segmentation (to identify and highlight the exact tampered regions). It is also designed to run on a Raspberry Pi, enabling real-time, low-power deployment for applications like social media verification, surveillance, and security systems.

The literature review shows that image forgery detection has evolved from traditional feature-based methods to advanced deep learning approaches such as CNNs, U-Net variants, hybrid models, and attention-based architectures. While deep learning models generally achieve high accuracy, common challenges include high computational cost, poor generalization across datasets, sensitivity to compression, and difficulty detecting complex or subtle manipulations. Lightweight and edge-device solutions exist but often sacrifice robustness.

The proposed system addresses these gaps by combining efficient feature extraction (ResNet) with precise tamper localization (U-Net) in a lightweight framework suitable for edge deployment.

The methodology outlines a pipeline consisting of image input, preprocessing, feature extraction, classification, localization, and output visualization. The system ultimately provides both a binary decision (real/fake) and a visual map of manipulated regions, making it useful for practical real-time forensic applications.

Conclusion

In our project, we developed a system to detect fake images using AI technology. The system uses ResNet to understand image features and U-Net to find edited parts. It can not only tell whether an image is fake but also highlight the manipulated area. Compared to older methods, this approach gives better accuracy and results. The model is designed to be lightweight and efficient. It can run on a Raspberry Pi, making it suitable for real-time use. This makes it useful for social media monitoring and security purposes. It can also help in detecting cybercrimes and verifying digital images. Overall, the system is simple, effective, and practical for real-world applications. In the future, it can be improved to detect more complex deepfake images.

References

In our project, we developed a system to detect fake images using AI technology. The system uses ResNet to understand image features and U-Net to find edited parts. It can not only tell whether an image is fake but also highlight the manipulated area. Compared to older methods, this approach gives better accuracy and results. The model is designed to be lightweight and efficient. It can run on a Raspberry Pi, making it suitable for real-time use. This makes it useful for social media monitoring and security purposes. It can also help in detecting cybercrimes and verifying digital images. Overall, the system is simple, effective, and practical for real-world applications. In the future, it can be improved to detect more complex deepfake images.

Copyright

Copyright © 2026 Sandhya N, Shalini S, Tejashwini S, Thrisha P, Dr. Anita P. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.