Research on Image Forgery Detection Based on Fusion Models

The rapid proliferation of digital image forgery has exacerbated a crisis of trust in society. Consequently, image forgery detection has garnered increasing attention in computer vision field. Despite the existence of numerous algorithms for image forgery detection, the utilization of lightweight models remains significantly insufficient. Many researchers encounter challenges due to limited resources for training and deploying models. To address this issue, this paper proposed an approach composed of different lightweight models. Specifically, it comprises two key components: 1) a fusion model module that acquires diverse feature maps from various lightweight networks, enabling the fused network to attend to a broader range of information; 2) treating image forgery detection as a binary classification problem, with the fused feature map fed into a Random Forest for classification. Furthermore, we also conducted some ablation experiments to validate the superior performance of our method.