“MAD”多注意力网络伪脸显廓系统

“MAD”fake face detector based on multi-attentional network

更为高级的人工智能换脸技术——局部人脸替换已经被研发、实现,这种技术能够保持原人脸被识别的同时仅仅替换面部某些部位,依然保证替换者的特征被识别。利用局部人脸替换技术进行身份伪造,会严重威胁社会安全和稳定。多注意力网络伪脸显廓系统采用高分辨率网络与多注意力网络相结合的方式,可以检测出目标人脸是否为虚假人脸以及其换脸区域,保证了虚假人脸的检测准确性、局部换脸检测的实用性,解决了局部换脸的虚假人脸伪造身份问题。方案在已有研究的基础上推陈出新,成本相对较低,一定程度上可用于解决人工智能换脸造成的法律问题,具有很好的可行性、实用性。

A more advanced artificial intelligence face replacement technology,partial face replacement,has been developed and implemented.This technology can keep the original face being recognized while replacing only certain parts of the face,so that the characteristics of the replacement are also recognized.The use of partial face replacement technology for identity forgery can seriously threaten social safety and stability.Our Multi-Attention Detector uses a combination of high-resolution network and multi-attention network for false faces with local artificial intelligence face swapping,which can detect whether the target face is a false face and the face-swapping part of the false face,ensuring the accuracy of false face detection,the practicality of local face-swapping detection,and solving the problem of falsification identity with partial face swapping.Our work pushes forward on the basis of existing research,with relatively low cost,and can be used to a certain extent to solve the legal problems caused by AI face replacement,with good feasibility and practicality.