Manipulated Faces Detection with Adaptive Filter

Figure 2

Framework of the proposed SAF. The input is a face to be detected. Spatial and frequency information is extracted by CNN, then they are fused by SPPF. The output is the result (real or fake).