Digital Image Forgery Detection Using JPEG Features and Local Noise Discrepancies

Algorithm 1

Algorithm description.

Begin

Load image 𝐼(𝑚,𝑛);

Generate BAG feature 𝐵_𝑥𝑦;𝑥=1,2,...,⌊𝑚/8⌋,𝑦=1,2,...,⌊𝑛/8⌋ for 𝐼(𝑚,𝑛);

Divide 𝐼 into 𝐾 segments 𝑆_𝑘,𝑘∈{1,2,...,𝐾} by SLIC superpixels;

  For each 𝑆_𝑘:

      Extract noise feature 𝐹_𝑘;

Assign label 𝐿_𝑘(𝑚,𝑛)∈{0,1} for 𝑆_𝑘 by graph cut;

Generate noise feature 𝐴_𝑥𝑦=(1/64)∑^8𝑦_{𝑗=8𝑦−7}∑^8𝑥_{𝑖=8𝑥−7}𝐿(𝑖,𝑗)

Calculate image quality score 𝑄;

Calculate coefficient 𝛼=𝑓(𝑄);

Set 𝐶_𝑥𝑦=  0;

Calculate 𝐶_𝑥𝑦=𝛼⋅𝐵_𝑥𝑦+(1−𝛼)⋅𝐴_𝑥𝑦;

Morphological operation-closing and opening (𝐶•𝑀_𝑐)∘𝑀_𝑜

End