Image Forgery Detection based on Fusion of Lightweight Deep Learning Models

Abstract: The popularity of capturing images has increased in recent years, as images contain a wealth of information that is essential to our daily lives. Although various tools are available to improve image quality, they are often used to falsify images, leading to the spread of misinformation. This has resulted in a significant increase in image forgeries, which is now a major concern. To address this, a decision fusion method is proposed in this project, which uses lightweight deep learning-based models for detecting image forgery. The proposed approach involves two phases that utilize pretrained and fine-tuned models, including SqueezeNet, MobileNetV2, and ShuffleNet, to extract features from images and detect image forgery. In the first phase, lightweight models are used to extract features from images without regularization, while in the second phase, fine-tuned models with fusion and regularization are employed to detect image forgery.

Keywords: Image Forgery, Deep Learning, Lightweight models, Convolutional Neural Networks (CNN)