The invention discloses a face video
forgery detection method, which comprises the following steps of: 1, acquiring a
data set, adopting a
diffusion model as a core generation framework, extracting a stable and high-discrimination identity embedding vector from the input
data set, and injecting the stable and high-discrimination identity embedding vector into a
generation process of the
diffusion model in a conditional constraint manner so as to generate a forgery
data set; and 2, completing face counterfeiting detection by using the counterfeiting data set generated in the step 1 based on a fine-grained distribution migration strategy. The method provided by the invention is used for solving the technical problems of insufficient quality of forged data, lack of
time sequence consistency, lack of identity
feature modeling, poor cross-identity
forging adaptability, insufficient sensitivity of a detection model to fine-grained artifacts and poor robustness.