parfor loop??? making it faster how?

i have wrote something like this to search for a color reference in live feed from webcam:

opengl hardware 

while (nFrame<z)

      b = step(vidobj);
      n=0;
      yb=size(b,1); %rows
      zb=size(b,2); %columns
      %u=zeros(zb,2); %------------------------------
      v=0;
      for g=1:1:yb
          for h=1:1:zb
              if (b(g,h,1)>=rot_min) && (b(g,h,1)<=rot_max)
              n=n+1;
              v(n,1)=g;
              v(n,2)=h;
              end
          end
      end
  v;
%end of rot search

so now i want to exploit the core i5 to use all for cores and make the code even more faster...

Answer 1

[Mandar]

i know the syntax. here are the probs that i am facing:

here are 2 problems:

a) i can't use the counter 'n' and parfor doesn't allow me to use the index from the earlier loop :O :( please tell me how should i use the parfor to search position of the red reference value!!! :(

b) Also i used the preallocation of v matrix with zeros and at the end i want the final v matrix without zero i.e. with the values other than 0 only. v=(v(v~=0)) works and gives the v matrix in a column matrix. that is not what i want. :(

Please Help!

[Iain]

1. Put the parfor in the outer loop.

2. You can't use that n in a parfor loop. You may, instead, create "v" to be 2 by yb by zb, and initialised (or set) to NaN when you don't have a valid answer. You can then eliminate the NaNs by

 v = reshape(v(~isnan(v),2,[]);

Your problem, however has a faster/simpler solution.

 indices = find(b(:,:,1)>= rot_min & b(:,:,1)<=rot_max);

That gets you linear indices to find your qualifying values of b. ind2sub can then be used to return, into a couple of variables (or columns), the rows and column values of the qualifying elements.

Answer 2

[Nitin]

you need first to use:

matlabpool open, then use parfor

You will find some nice examples here