Fake Image Detection Based on Attention-based Feature Aggregation
Weinan Zhang, Sanshuai Cui, Jingxi Xue · 2024
In recent years, AI images (referred to as deepfake images) have become increasingly difficult to distinguish from real images.Existing detection techniques perform poorly when processing images of unknown origin.To address this problem, this paper proposes a new network that introduces Multi-type Stepwise Pooling and attention to aggregate deep and shallow features and efficiently process complex information, thereby enhancing the generalization ability and detection performance of the network.Experiments show that the model performs well on multiple datasets, and we increase the average Acc to 86.3% and AP to 92.4%.Our research provides new ideas for image authenticity detection and technical support for preventing the abuse of deep fake technology.