I have written a logic for co-occurance matrix but it is taking a lot of time, is there any other to built the logic?

4.Calculate four co-occurrence matrices P, take the distance 1, the angles are 0, 45, 90, 135 respectively

%M&N ARE ROWS &COLOUMNS OF THE IMAGE
%-------------------------------------------------------------------------- 

P = zeros(16,16,4);   %Generate 4 16 * 16 0 matrix

for m = 1:16 

      for n = 1:16 

          for i = 1:M 

              for j = 1:N 

                  if 
  j<N&Gray(i,j)==m-1&Gray(i,j+1)==n-1  %0 ° m for the row n for the column
                      P(m,n,1) = P(m,n,1)+1; 
                      P(n,m,1) = P(m,n,1); 
                  end 
                  if i>1&j<N&Gray(i,j)==m-1&Gray(i-1,j+1)==n-1  %45° 
                      P(m,n,2) = P(m,n,2)+1; 
                      P(n,m,2) = P(m,n,2); 
                  end 
                  if i<M&Gray(i,j)==m-1&Gray(i+1,j)==n-1  %90° 
                      P(m,n,3) = P(m,n,3)+1; 
                      P(n,m,3) = P(m,n,3); 
                  end 
                  if i<M&j<N&Gray(i,j)==m-1&Gray(i+1,j+1)==n-1  %135° 
                      P(m,n,4) = P(m,n,4)+1; 
                      P(n,m,4) = P(m,n,4); 
                  end 
              end 
          end 
          if m==n 
              P(m,n,:) = P(m,n,:)*2; 
          end 
      end 
  end

Answer 1

[Walter Roberson]

You could vectorize. Or you could use graycomatrix()

[teja jayavarapu]

Vectorize??

[Walter Roberson]

https://www.mathworks.com/help/matlab/math/matrix-indexing.html#bq7egb6-1

For example, you have

                  if i<M&j<N&Gray(i,j)==m-1&Gray(i+1,j+1)==n-1  %135° 
                      P(m,n,4) = P(m,n,4)+1; 
                      P(n,m,4) = P(m,n,4); 
                  end

inside loops on i and j. You could replace those loops with several pieces including:

i = 1 : M - 1;   %certain to be < M so do not need to test i<M
j = 1 : N - 1;   %don't run off the end of the array
mask = Gray(i, j) == m-1 & Gray(i+1, j+1) == n-1;
P(m, n, 4) = P(m, n, 4) + nnz(mask);
P(n, m, 4) = P(m, n, 4);

[teja jayavarapu]

previously i attached the greylevel co-occurance matrix code,using that code was not giving the same resuts using greycomatrix in matlab ,can you please identify if there is any error in the code. And In case if the code was right can you please provide idea about texture parameters(capacity,moment of inertia,relevance)

[Walter Roberson]

Instead of looping over m and n and testing Gray(i,j)==m-1&Gray(i,j+1)==n-1 (for example) you should just form all the diference pairs and increment the corresponding locations. For example,

V = @(M) M(:);
accumarray(1+[V(Gray(:,1:end-1)), V(Gray(:,2:end))], 1)

[teja jayavarapu]

Sorry I didn't get you,can you explain more precisely.

[Walter Roberson]

Under the conditions that all values in Gray are integers in the range 0 to 15, then your code

for m = 1:16

      for n = 1:16 

          for i = 1:M 

              for j = 1:N 
  j<N&Gray(i,j)==m-1&Gray(i,j+1)==n-1  %0 ° m for the row n for the column
                      P(m,n,1) = P(m,n,1)+1; 
                      P(n,m,1) = P(m,n,1); 
             end
           end
         end
       end

can be replaced by

V = @(M) M(:);
P0 = accumarray(1+[V(Gray(:,1:end-1)), V(Gray(:,2:end))], 1);

This will make P0 a 16 x 16 matrix containing the applicable counts.

Your other three conditions can be created with similar lines. You can then put all four of them together into your P matrix.

P = cat(3, P0, P45, P90, P135);

Answer 2

[Roger Stafford]

In the code you have written the tests are false the great majority of the cases, and that suggests strongly that the method can be made more efficient.

Since you mention “image” in connection with rows and columns of ‘Grey’, I will assume all the values in ‘Grey’ are non-negative integers. I will add 1 to the values of ‘Grey’ and replace any that then exceed 17 by 17 itself. This latter will make P have just one extra row and one extra column of false values which will be discarded at the end.

P = zeros(17,17,4);
G = reshape(min(Grey(:)+1,17),size(Grey));
for i = 1:M
 for j = 1:N-1
  m = G(i,j); 
  n = G(i,j+1);
  P(m,n,1) = P(m,n,1)+1;
 end
end
for i = 2:M
 for j = 1:N-1
  m = G(i,j); 
  n = G(i-1,j+1);
  P(m,n,2) = P(m,n,2)+1;
 end
end
for i = 1:M-1
 for j = 1:N
  m = G(i,j); 
  n = G(i+1,j);
  P(m,n,3) = P(m,n,3)+1;
 end
end
for i = 1:M-1
 for j = 1:N-1
  m = G(i,j); 
  n = G(i+1,j+1);
  P(m,n,4) = P(m,n,4)+1;
 end
end
P = P(1:16,1:16,:); % Trim off one row and one column
P = P + permute(P,[2,1,3]); % Replaces the "P(n,m,-)=P(m,n,-)" above

[teja jayavarapu]

I am writing the code co-occurance matrix, so we have to check with the neighbouring pixels,I think you are not checking with the neighbouring pixels.

[Image Analyst]

Again, just use the built-in graycomatrix() and graycoprops().

[Roger Stafford]

@Teja: When I get m and n from m = G(i,j) and n = G(i,j+1) for example, and then add 1 to P(m,n,1), this is updating the co-occurance matrix for neighboring pixels.

Why don't you just try out the code and see if it agrees with your own results, (and hopefully is faster.)

[teja jayavarapu]

P = P + permute(A,[2,1,3]); % Replaces the "P(n,m,-)=P(m,n,-)" above

what is A here I think it is P.And why are you adding with P

I think P=permute(P,[2 1 3])

[Roger Stafford]

@Teja: Yes, that is an error. I have corrected it now. Thank you.

[teja jayavarapu]

Can you give me idea about guassian internal normalisation

Answer 3

[teja jayavarapu]

Ok,thank you for the support,can you please help me with the Gussian internal normalisation for texture parameters of (capacity,entropy,relevance,inverse difference)