关键词: 深度学习, 神经网络, 伪造人脸检测, 图像增强, 特征融合

Abstract: With the rapid evolution of generative models such as Generative Adversarial Networks (GAN), the quality of forged facial images has continued to improve, posing severe challenges to social media, identity authentication and public opinion security. Forged image detection has become a research hotspot in the current field of information network security. The existing methods mainly focus on directions such as spatial domain texture analysis, frequency domain artifact extraction or temporal consistency modeling, and introduce multiple attention mechanisms and feature fusion strategies to improve performance. However, these methods usually have problems such as weak generalization ability and rigid feature fusion mechanisms, making it difficult to adapt to the constantly evolving forgery techniques. This paper proposes a forged face detection model based on a dual-branch structure in response to the above problems. It extracts multi-dimensional features in the spatial domain and the frequency domain respectively, and introduces a trainable dynamic feature fusion module to achieve adaptive weighted fusion between feature domains and enhance feature complementarity. Meanwhile, an image enhancement strategy based on random channel masks is designed to effectively improve the robustness of the model in various forgery scenarios. The experimental results show that the method proposed in this paper has achieved excellent performance on multiple benchmark datasets and demonstrated strong generalization ability in cross-dataset tests, providing an efficient and scalable solution for forged image detection.

Key words: deep learning, neural network, forged face detection, image enhancement, feature fusion