【摘要】 深度伪造技术是一种基于深度学习的图像和音视频合成技术。它使用深度神经网络生成高度逼真的虚构内容，尤其在人脸篡改方面，已对日常生活产生了不良影响。随着深度学习的发展，传统的卷积神经网络难以准确识别当前的人脸篡改行为。为应对该挑战，提出了一种多模态人脸篡改识别方法，创新性地结合了频域处理、错误水平分析以及语义信息进行检测。经过在FaceForensics++数据集上进行实验测试，结果表明，该方法的准确率高达83.10%，是一种有效的人脸篡改检测方法。更多还原

【Abstract】 Deepfake technology is a deep learning based image and audio video synthesis technique. It uses deep neural networks to generate highly realistic fictional content, especially in the area of facial tampering, which has had a negative impact on daily life. With the development of deep learning, traditional convolutional neural networks are difficult to accurately recognize current facial tampering behavior. To address this challenge, a multimodal facial tamper recognition method has been proposed, which innovatively combines frequency domain processing, error level analysis, and semantic information for detection. After experimental testing on the FaceForensics++dataset, the results show that the accuracy of this method is as high as 83.10%, making it an effective face tamper detection method.更多还原