How can I make *this* program faster? image ownership & tampering authentication scheme

My program is written after the paper "A watermarking-based image ownership and tampering authentication scheme" by Chin-Chen Chang, Yih-Shih Hu and Tzu-Chuen Lu, Pattern Recognition Letter 27 (2006) 439-446.

It works so far, but it is extremely slow.

Can anyone give me some tips how to make it faster?

I've seen the vectorization tip of loops, but I don't know how to use it in my case. If this is a possibility can you show me an example of how to do it here?

I also used different Hash functions by Jan Simon (DataHash http://www.mathworks.com/matlabcentral/fileexchange/31272-datahash and this one http://www.mathworks.com/matlabcentral/fileexchange/7919-md5-in-matlab) but it didn't make much time difference.

Edit: you have to call it by rgb2gray(imread('your_picture.type')) Thank you!

function [PixelSignificant, PixelValues] = schemel(I)

%get picture info
A = size(I);
height = A(2);
length = A(1);

%preallocation -- more needed?
vector1 = ones(1,length);
vector2 = 1:length;
P2 = zeros(length, 3);
PixelSignificant = cell(length-2, height-2); 
PixelValues = zeros(length, height);
exOrNew = zeros(1,4);
n = 0; 

%Get all Pixel values
for i = 1:height
  P2 = impixel(I, vector1, vector2);
  for j = 1:length
    PixelValues(j,i) = P2(j,1);
  end
  vector1 = vector1+1;
end

%create the new pixel values, with md5
%without edge
%IDK = 56, Users secret key SK =189 vll später über scheme eingeben

    for j = 2:length-1
      for i = 2:height-1
    %feature extraction + hash
        ConVal = strcat(num2str(PixelValues(j+1,i-1)), num2str(PixelValues(j,i-1)), num2str(PixelValues(j-1,i-1)), num2str(PixelValues(j-1,i)), num2str(PixelValues(j-1,i+1)), num2str(PixelValues(j,i+1)), num2str(PixelValues(j+1,i+1)), num2str(PixelValues(j+1,i)), num2str(n*(length-2)), '56', '189');
        Y1 = mMD5(ConVal); 
        X1 = dec2bin(hex2dec(Y1(1:32)));

    %length correcting: 132'bit' --why isnt't it always 128 bits long?
    %somehow I had to concatenate each '0' by itself, otherwise they got lost
        if size(X1,2) == 124 
          X2 = horzcat('0', X1, '0000');
          X3 = horzcat('0', X2);
          valuesBitwiseLonger = fliplr(horzcat('00', X3));

        elseif size(X1,2) == 125 
          X2 = horzcat('0', X1, '0000');
          X3 = horzcat('0', X2);
          valuesBitwiseLonger = fliplr(horzcat('0', X3));

        elseif size(X1,2) == 126
          X2 = horzcat('0',X1,'0000');
          valuesBitwiseLonger = fliplr(horzcat('0', X2));

        elseif size(X1,2) == 127
          valuesBitwiseLonger = fliplr(horzcat('0',X1,'0000'));

        elseif size(X1,2) == 128
          valuesBitwiseLonger = fliplr(horzcat(X1,'0000'));
        end

    %sigma and r 
      sigma = (PixelValues(j+1,i-1) - PixelValues(j,i-1))^2 + (PixelValues(j,i-1) - PixelValues(j-1,i-1))^2 + (PixelValues(j-1,i-1) - PixelValues(j-1,i))^2 +    (PixelValues(j-1,i) - PixelValues(j-1,i+1))^2 + (PixelValues(j-1,i+1) - PixelValues(j,i+1))^2 +    (PixelValues(j,i+1) - PixelValues(j+1,i+1))^2 +(PixelValues(j+1,i+1) - PixelValues(j+1,i))^2 + (PixelValues(j+1,i) - PixelValues(j+1,i-1))^2;

      if (16 <= sigma) && (sigma <= 255)
        r = 4;

      elseif (8 <= sigma) && (sigma < 16)
        r = 3;

      else %0 <= sigma < 8
        r = 2;
      end

    %folding  start    
    %first two blocks folded
        for u = 1 : r
          ExOr1 = valuesBitwiseLonger(u);
          ExOr2 = str2num(valuesBitwiseLonger(u + r));
          exOrNew(u) = bitxor(str2num(ExOr1), ExOr2);
        end

    %Rest folded -- very 'slow' here -> 'too' many loops
        for l = 2*r : 132 - r + 1
          NewExOr = '';
          for p = 1 : r
            NewExOr = horzcat(NewExOr, num2str(exOrNew(p)));
          end

          for u = 1 : r
            ExOr1 = valuesBitwiseLonger(l + u - 1);
            ExOr2 = str2num(NewExOr(u));
            exOrNew(u) = bitxor(str2num(ExOr1), ExOr2);
          end
        end
        PixelSignificant(j,i) = {NewExOr};

    %folding ende
      end 
    %check wie schnell + pixelindex zähler    
   n = n+1
    end 
  end

Answer 1

[Walter Roberson]

You should be passing a second argument to dec2bin() so that you get out a consistent length.

Be warned that the results you get out of the dec2bin() are going to be mostly nonsense. You are doing hex2dec() on a 32 nibble hex value, which is going to give you a result between 0 and 340282366920938463463374607431768211456 but only 53 of the (up to) 129 bits at a time are ever going to be meaningful. MATLAB does not have any 128 bit integer data type.

Answer 2

[Anton]

Thank you so much, together with your help on this question I was able to speed up my program a lot!