Unknown issue with image refinement process - image

Question

As introduced in this work (code), the following code sould refine an image (i.e. a gray scale saliency map).
Refinement Function:
function sal = Refinement(y, dim)
th_2 = graythresh(y);
if dim == 1
sal = y;
sal(y<th_2) = 10*(y(y < th_2))/th_2 - 10;
sal(y>=th_2) = 10*(y(y >= th_2) - th_2)/(1-th_2);
sal = 1 ./ (1 + exp(-sal)) + y;
sal = normalization(sal, 0);
elseif dim == 2
[r, c] = size(y);
y_col = reshape(y,[1 r*c]);
sal_col = y_col;
sal_col(y_col<th_2) = 10*(y_col(y_col < th_2))/th_2 - 10;
sal_col(y_col>=th_2) = 10*(y_col(y_col >= th_2) - th_2)/(1-th_2);
sal_col = 1 ./ (1 + exp(-sal_col)) + y_col;
sal = reshape(sal_col, [r c]);
end
end
normalization function:
function matrix = normalization(mat, flag)
% INPUT :
% flag: 1 denotes that the mat is a 3-d matrix;
% 0 denotes that the mat is a matrix;
%
if flag ~= 0
dim = size(mat,3);
matrix = mat;
for i = 1:dim
matrix(:,:,i) = ( mat(:,:,i) - min(min(mat(:,:,i)))) / ( max(max(mat(:,:,i))) - min(min( mat(:,:,i))) + eps);
end
else
matrix = ( mat - min(min(mat)))/( max(max(mat)) - min(min(mat)) + eps);
end
However, after applying the function values of the image matrix will be changed, the result remains the same as the image before the refinement.
Is there a conceptual error with that or the implementation failed?
P.S. The input image (saliency map) for refinery is something like below. in the refined saliency map, the foreground (satrfish in this image) should stand out(becomes homogeneously white as possible) and the background noise should be removed (becomes homogenously black as possible):