Detection of Image Seam Carving Using a Novel Pattern

Algorithm 2

The algorithm of feature selection.

(1)	Input: Grayscale image (I)
(2)	Output: Histogramfinal(LNMOP)
(3)	Initial: n > r, m ⟵ 5 : 10
(4)	Begin
(5)	sub {c1, c2,…, cn} ⟵ block(I, n)
(6)	//The image I is divided into n subblocks with equal size and no overlap.
(7)	for each block c ∈ sub, do
(8)	histogram {HOG1, HOG2, …, HOGn} ⟵ extract (c, HOG)//extract HOG features
(9)	for r ⟵ 1 : m, do
(10)	histogram {LNMOP|^11, LNMOP|^21, …, LNMOP|^r1, LNMOP|^12, LNMOP|^22, …, LNMOP|^r2,
(11)	…, …, LNMOP|^1n, LNMOP|^2n, …, LNMOP|^rn} ⟵ extract (c, LNMOP)
(12)	//extract LNMOP features
(13)	end
(14)	end
(15)	for each block c ∈ sub, do
(16)	l ⟵ rank(HOG, histogram|HOG) //Determination the rank of HOG features
(17)	histogram {LNMOP|1, LNMOP|2, …, LNMOP|n} ⟵ LNMOP|^l
(18)	//Screening LMOP features matching HOG levels
(19)	end
(20)	histogramfinal(LNMOP) ⟵ concatenate (histogram {LNMOP|1, LNMOP|2, …, LNMOP|n})
(21)	//Concatenation into the final features