Deepfake Detection Method Based on Cross-Domain Fusion

Algorithm 1

The key algorithm of Cross-domain model.

/ Frequency domain feature extraction process /
Input: preprocessed image .
Output: frequency domain feature .
(1)	Use formula (3) to calculate , , .
(2)	, , is mapped into an edge matrix by formula (4).
(3)	Q = operate on ReLu layer, batch normalization layer, and maximum pooling layer.
(4)	= the tile of Q.
/Obtain high-level semantic spatial domain feature/
Input: preprocessed image .
Output: high-level semantic spatial domain feature .
(5)	For i ≤ 4 do
(6)	P = operate on convolution layer, ReLu layer, batch normalization layer, and maximum pooling layer.
(7)	End for
(8)	= the tile of P.
/Fuse features /
Input: frequency domain feature and spatial domain feature .
Output: the image facial feature vector .
(9)	Obtain image feature vector by formula (6).
(13)	Return