Forgery Detection in Digital Images by Multi-Scale Noise Estimation

. 2021 Jul 17;7(7):119. doi: 10.3390/jimaging7070119

Algorithm 1 Pseudo-code for the proposed method

Input: image I of shape

N_{x} \times N_{y}

with C color channels.
Parameters:

W = 256

macroblock side,

S = 0.5

stride, num_scales = 3 number of scales.

1:
$M_{x} = ⌊ N_{x} / (W \times S) ⌋ - 1 .$ ▹ horizontal number of macroblocks
2:
$M_{y} = ⌊ N_{y} / (W \times S) ⌋ - 1 .$ ▹ vertical number of macroblocks
3:
$macroblocks_list \leftarrow list of all W \times W macroblocks with S stride .$
4:
for each scale s do
5:
for each channel c do
6:
$I_{s}^{c} \leftarrow get image in scale s and channel c .$
7:
$f_{I_{s}^{c}} \leftarrow noise curve estimation for I_{s}^{c} using$ [29] as described in Appendix A.
8:
$H^{c} \leftarrow zeros (M_{x} \times M_{y}) .$
9:
for each macroblock in macroblocks_list do
10:
$M_{s}^{c} \leftarrow get macroblock in scale s and channel c .$
11:
$f_{M_{s}^{c}} \leftarrow noise curve estimation for M_{s}^{c} using$ [29] as described in Appendix A.
12:
$H^{c} [M_{s}^{c}] \leftarrow percentage of bins of f_{M_{s}^{c}} below f_{I_{s}^{c}} .$
13:
end for
14:
end for
15:
$H_{s} \leftarrow geometric mean of the heatmaps H^{c} .$
16:
end for
17:
$H_{aux} \leftarrow sum and normalization of heatmaps H_{s} .$
18:
$H \leftarrow compute for each pixel the average of H_{aux} for each macroblock containing it .$
19:
returnH.