Abstract

With the rapid advancement of digital imaging technologies and editing tools, image forgery has become a significant threat to information integrity, posing challenges in various domains such as forensics, journalism, and authentication systems. Traditional forgery detection techniques often struggle to keep up with sophisticated manipulation methods like copy-move, splicing, and deepfake alterations. To address these challenges, this research proposes a novel image forgery detection framework based on the fusion of lightweight deep learning models. Unlike conventional deep learning approaches that rely on computationally expensive architectures, our method integrates multiple lightweight neural networks to enhance detection accuracy while maintaining efficiency. The fusion mechanism effectively extracts both spatial and frequency domain features, enabling the model to identify forgery patterns with higher precision.The proposed framework consists of a multi-branch feature extraction module, where each branch employs a different lightweight deep learning model to capture diverse feature representations. These extracted features are then fused using an attention-based mechanism to emphasize critical regions affected by forgery operations. The model undergoes rigorous evaluation on publicly available benchmark datasets, including CASIA, CoMoFoD, and DEFACTO, demonstrating superior performance over state-of-the-art methods. Our experiments show that the proposed approach achieves significant improvements in accuracy, precision, recall, and F1-score while maintaining a low computational footprint, making it suitable for real-time applications and resourceconstrained environments. Furthermore, we conduct ablation studies to analyze the contribution of each component within the model, providing insights into the effectiveness of different fusion strategies. This research advances the field of image forensics by offering a scalable, robust, and computationally efficient solution for detecting forged images across diverse scenarios.