Adaptive Steganalysis Based on Selection Region and Combined Convolutional Neural Networks

Algorithm 1

Finding the most effective region for steganalysis.

Input: input parameters: an input image , the width and the height of the region.

Output: output result: the MER , save as PGM format.

() Initialize the probability map (matrix) with random weights ;

() //Select a pixel from the input image and calculate the probability.

() for  , Rows do

()   for  , Column do

()       Calculate change costs using Eq. (1);

()       Compute using Eq. (4);

()       Set the probability using Eq. (2);

()       Store in ;

() Initialize the MER with 0;

() //Select the upper left corner coordinates of the area with size of .

(11) for  , Rows-  do

(12)   for  , Column-  do

(13)       Calculate the sum of the probability in an matrix () with top-left corner of ;

(14)       Statistics out all of the sum and its corresponding ;

(15) Select the maximal value of those sums;

(16) Cut the area in an input image according to and ;

(17) Save this area as a PGM format image;