Research Article
Manipulated Faces Detection with Adaptive Filter
Algorithm 1
Spatial Pyramid Pooling Fusion.
(i) | Input: spatial feature IS (M M Ch1); | (ii) | frequency feature IF (N N Ch2); | (iii) | SPP levels L; | (iv) | pooling type T | (v) | Output: fusion feature FF | (1) | cnt = 0; | (2) | S = []; | (3) | F = []; | (4) | while cnt < L do | (5) | Cnt + = 1; | (6) | Kernel_S = (M/cnt, M/cnt); | (7) | Kernel_F= (N/cnt, N/cnt); | (8) | S = [S, Pooling(IS, T, Kernel_S)]; | (9) | F = [F, Pooling(IF, T, Kernel_F)]; | (10) | end | (11) | Attention_S = S; | (12) | Attention_F = F; | (13) | Fusion1 = Attention_S; | (14) | Fusion2 = Attention_F; | (15) | FF = [Fusion1, Fusion2] |
|