中圖分類號：TP183 文獻標識碼：A DOI： 10.19358/j.issn.2096-5133.2020.05.003
引用格式：易旭，白天.基于Star-Gan的人臉互換算法［J］.信息技術與網絡安全，2020,39（5）：12-16.

Abstract： The effect of face swap algorithm based on deep learning will be affected by the brightness of the background face environment,facial expression,posture and other factors.There are inherent disadvantages in the current face swap models.The discriminator structure based on PatchGan can enhance the local consistency of human face through full convolution network.The generator takes the feature output of the Unet encoder as the input,and considers the multilayer information details.Among them,StarGan model is adopted as the overall model architecture,and case normalization layer is introduced to ensure the image independence.Finally,it is validated on FaceForensics++ face exchange data set.The results show that the optimized model has better generation effect and details.

Key words : deep learning;face swap;generative adversarial networks