A Method for Face Expression Forgery Detection Based on Reconstruction Error

The development of AIGC has facilitated a creation of forgery face expression, once these generated fake images are abused by criminals, it will bring great harm to our society and country. The current face forgery detection usually stays at the perspective of spatial domain, lacks consideration of information sharing and does not have the ability to trace the differences between original face and generated face. To further promote the development of facial forgery detection, this paper provides a method for face expression forgery detection based on reconstruction error. Reconstruction error measures an error between a real face image and its reconstruction image and can serve as a bridge to distinguish real images and generated images. Furthermore, this method uses a residual neural network as its backbone, which introduced autoencoder module to obtain hidden space information during the reconstruction process. Also, we build a human face dataset with 30000 real images and 30000 face expression edited images. In the comparative experiment, our method achieved an accuracy of 90.13%. Overall, this method implements: efficiently and accurately identify forgery face expression.