“Robust Detection of Image Fraud Employing Convolutional Neural Networks (CNNs): The ‘Imageguard’ Application”

A surge in digital manipulation necessitates robust methods to combat image tampering, particularly the prevalent technique of image splicing. This paper addresses this challenge by proposing a real-time image splicing detection system. The system leverages a pre-trained convolutional neural network (CNN) model, specifically ResNet50, adapted for the task. The CASIA V2 dataset, containing both authentic and spliced images, serves as the foundation for model training and evaluation. To enhance performance, data augmentation techniques and meticulous hyperparameter tuning are employed. Additionally, exploration of alternative model architectures is conducted. The trained model interacts with a user-friendly Android mobile application through an API endpoint hosted on a web server. This enables real-time image splicing detection directly on the mobile device. Initial testing yielded an accuracy of 78 %, subsequently improved to 83 % through further analysis and model modifications. Continuous training is expected to further enhance accuracy. The Android application seamlessly displays the splicing detection results in real-time. In essence, this paper demonstrates a real-time image splicing detection system utilizing a modified ResNet50 architecture and an Android application. This approach offers significant potential across multiple domains, such as aiding forensic investigations by analyzing digital evidence, verifying the authenticity of visual content in journalism, and combating the spread of manipulated imagery on social media platforms to ensure the integrity of digital media.