how to preallocate the variable in matlab code?

while implementing kernal based grapg cut method, i am getting error at line: area(1+1)=999999999; suggestion for correcting the error is line 147: the variable 'area' appears to change size on every loop iteration(within a script). consider preallocation for speed. so please suggest me how to preallocate the variable area,in the matlab code?

i am attaching the code below

clear all; close all; 

%%%%%%%%%%%%%%%%%%%%%%%Main inputs and parameters%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%Note: The RBF-kernel parameters are given in function kernel RBF.m%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%Example with a color image%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%path = 'Images\Color_image.jpg';
%im = im2double(imread(path)); 
%alpha=1; %The weight of the smoothness constraint
%k =8; %The number of regions

%%%%%%%Example with a SAR image corrupted with a multiplicative noise%%%%%%
%%%%%%%%%%%%%%%%Uncomment the following to run the example)%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

path = 'Images\Sar_image.tif';
im = im2double(imread(path));
alpha=0.6;
k =4;

%%%%%%%%%%%%%%%%%%%%%%%%%%Example with a brain image%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%Uncomment the following to run the example)%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%path = 'Images\Brain_image.tif';
%im = im2double(imread(path));
%alpha=0.1;
%k =4;

sz = size(im);
Hc=ones(sz(1:2));
Vc=Hc;
i_ground = 0; % rank of the bakground for plotting, 0: the darkest; 
%k-1 the brightest; 99: nowhere

diff=10000;
an_energy=999999999;
iter=0;
iter_v=0;
energy_global_min=99999999;

distance = 'sqEuclidean'; % Feature space distance

% Initialization: cluster the data into k regions
tic,
disp('Start kmeans');
data = ToVector(im);
[idx c] = kmeans(data, k, 'distance', distance,'EmptyAction','drop','maxiter',100);
c=c(isfinite(c(:,1)),:);
k=size(c,1);
k_max=k;
kmean_time=toc;

Dc = zeros([sz(1:2) k],'single');   
c;

tic
while iter < 5
    iter=iter+1;
    clear Dc
    clear K
    c;
    for ci=1:k
        K=kernel_RBF(im,c(ci,:));
        Dc(:,:,ci)=1-K;
    end   
    clear Sc
    clear K
    %%The smoothness term
    Sc = alpha*(ones(k) - eye(k)); 
    gch = GraphCut('open', Dc, Sc, Vc, Hc);
    [gch L] = GraphCut('swap',gch);
    [gch se de] = GraphCut('energy', gch);
    nv_energy=se+de;
    gch = GraphCut('close', gch);

      if (nv_energy<=energy_global_min)
          diff=abs(energy_global_min-nv_energy)/energy_global_min;
          energy_global_min=nv_energy;
          L_global_min=L;
          k_max=k;

          nv_energy;
          iter_v=0;
          % Calculate region Pl of label l
          if size(im, 3)==3 % Color image
          for l=0:k-1
              Pl=find(L==l);
              card=length(Pl);
              K1=kernel_RBF(im(Pl),c(l+1,1));K2=kernel_RBF(im(Pl),c(l+1,2));K3=kernel_RBF(im(Pl),c(l+1,3));
              smKI(1)=sum(im(Pl).*K1); smKI(2)=sum(im(Pl+prod(sz(1:2))).*K2); smKI(3)=sum(im(Pl+2*prod(sz(1:2))).*K3);
              smK1=sum(K1);smK2=sum(K2);smK3=sum(K3);
              if (card~=0)
                  c(l+1,1)=smKI(1)/smK1;c(l+1,2)=smKI(2)/smK2;c(l+1,3)=smKI(3)/smK3;
              else
                  c(l+1,1)=999999999;c(l+1,2)=999999999;c(l+1,3)=999999999;
              end
          end
          end

          if size(im, 1)==1 % Gray-level image
          for l=0:k-1
              Pl=find(L==l);
              card=length(Pl);
              K=kernel_RBF(im(Pl),c(l+1,1));
              smKI=sum(im(Pl).*K);
              smK=sum(K);
              if (card~=0)
                  c(l+1,1)=smKI/smK;
              else
                  c(l+1,1)=999999999;
              end
          end
          end

          c=c(c(:,1)~=999999999,:);
          c_global_min=c;
          k_global=length(c(:,1));
          k=k_global;

      else
          iter_v=iter_v+1;
          %---------------------------------
          %       Begin updating labels
          %---------------------------------
          % Calculate region Pl of label l
          if size(im, 3)==3 % Color image 
          for l=0:k-1           
              Pl=find(L==l);
              card=length(Pl);
              K1=kernel_RBF(im(Pl),c(l+1,1));K2=kernel_RBF(im(Pl),c(l+1,2));K3=kernel_RBF(im(Pl),c(l+1,3));
              smKI(1)=sum(im(Pl).*K1); smKI(2)=sum(im(Pl+prod(sz(1:2))).*K2); smKI(3)=sum(im(Pl+2*prod(sz(1:2))).*K3);
              smK1=sum(K1);smK2=sum(K2);smK3=sum(K3);
              % Calculate contour Cl of region Pl
              if (card~=0)
                  c(l+1,1)=smKI(1)/smK1;c(l+1,2)=smKI(2)/smK2;c(l+1,3)=smKI(3)/smK3;
              else
                  c(l+1,1)=999999999;c(l+1,2)=999999999;c(l+1,3)=999999999;
                  area(l+1)=999999999;
              end
          end
          end

          if size(im, 3)== 1 % Gray-level image 
          for l=0:k-1           
              Pl=find(L==l);
              card=length(Pl);
              K=kernel_RBF(im(Pl),c(l+1,1));
              smKI=sum(im(Pl).*K);
              smK=sum(K);
              % Calculate contour Cl of region Pl
              if (card~=0)
                  c(l+1,1)=smKI/smK;
              else
                  c(l+1,1)=999999999;
                  area(l+1)=999999999;
              end
          end
          end

          c=c(c(:,1)~=999999999,:);
          k=length(c(:,1));
      end
  end

L=L_global_min;
energy_global_min;
c=c_global_min;

size(c,1)
iter;

%Show the results

if size(im, 3)==3 % Color image 
img=zeros(sz(1),sz(2),3);
j=1;
imagesc(im); axis off; hold on; 

for i=0:k_max-1
    LL=(L_global_min==i);
    is_zero=sum(sum(LL));
    if is_zero
         img(:,:,1)=img(:,:,1)+LL*c(j,1);
         img(:,:,2)=img(:,:,2)+LL*c(j,2);
         img(:,:,3)=img(:,:,3)+LL*c(j,3);
         j=j+1;
    end
    if i~=i_ground
        color=[rand rand rand];
        contour(LL,[1 1],'LineWidth',2.5,'Color',color); hold on;
    end
end
figure(2);
imagesc(img); axis off;
end

if size(im, 3)==1 % Gray-level image 
img=zeros(sz(1),sz(2));
j=1;
imagesc(im); axis off; hold on; colormap gray; 

for i=0:k_max-1
    LL=(L_global_min==i);
    is_zero=sum(sum(LL));
    if is_zero
         img(:,:,1)=img(:,:,1)+LL*c(j,1);
         j=j+1;
    end
    if i~=i_ground
        color=[rand rand rand];
        contour(LL,[1 1],'LineWidth',2.5,'Color',color); hold on;
    end
end
figure(2);
imagesc(img); axis off;
end

Answer 1

[Adam]

That's way too much unformatted code to look through everything, but

area = zeros(k,1);

will pre-allocated the variable 'area' (or alternatively zeros(1,k) if you specifically want a row vector).

Pre-allocation is a speed improvement suggestion though. You say you are getting an error, which is a different matter if it is actually an error and making a speed improvement won't fix an error.

[Prerna Surbhi]

thanks for your advise, successfully solved the preallocation problem, but getting error at another line

[Adam]

This is completely independent of the question you are asking at the top. But since Guillaume is already responding to this problem below I won't double up on giving the same advice to that too!

Answer 2

[Guillaume]

Since your loop runs from 0 to k and you're using the loop index + 1 to assign to area, its maximum size is going to be k columns, thus:

area = zeros(1, k);

However, if you get an error, preallocation in itself is not going to make it go away. Please give the exact text of the error message.

[Prerna Surbhi]

thanks for your advise, successfully solved the preallocation problem, but getting error at another line

[Guillaume]

Well, the error message is clear, you don't have a function ToVector. Either it does not exist or the folder where that function exist is not on matlab's path.

If the former, create that function, if the latter, add that function's folder to matlab's path.

Answer 3

[Bjorn Gustavsson]

It is often simple, sometimes tedious other times impossible to preallocate the exact size of your variables. In this case I cant easily see. What you do is before the loop (if possible):

area = zeros(1,max_of_number_l);

If possible another trick is to run for-loops from the largest index downwards:

for i1 = max_i:-1:1,
  area(i1) = sum(randn(1,10));
end

A more exhaustive description of allocation issues in matlab you can find here: Preallocation performance

HTH

[Prerna Surbhi]

thanks for your advise, successfully solved the preallocation problem, but getting error at another line