Image to patches for local features extraction

Hi All,

I have an image and I need to extract local features,so I divided it to 32 * 32 patches and size of each patch =16 *16 pixels and extract that three color histograms features locally from each patch, now I need to :

1-Store these features in the descriptor x like in figure:

2-Store these feature discroptors in the X like in figure:

3-Store these X's in the global matrix called G like in figure:

I tried this code :

patches_counter=1;
for g=1:32
  for u=1:32
    if nom_ones>area_of_patch_by_two
      patch_h=h_seg_image_cell{g,u} ;
      patch_s=s_seg_image_cell{g,u} ;
      patch_v=v_seg_image_cell{g,u} ;
      
      % Describe each patch by 3 color histograms features      
      [patch_h_ftr1,xh]=imhist(patch_h);
      [patch_s_ftr2,xs]=imhist(patch_s);
      [patch_v_ftr3,xv]=imhist(patch_v);
    end
    
    % Create features vector
    x=[patch_h_ftr1;patch_s_ftr2;patch_v_ftr3];
    X(:,patches_counter) = x(patches_counter);
    patches_counter=patches_counter+1;
  end
end

But I got this ERROR

In an assignment  A(I) = B, the number of elements
in B and I must be the same.

Error in my_im2blocks1 (line 72)
        X_image_vector(patches_counter) =
        x_featurs_vector_patch;

Hint : Don't care about this condition

if nom_ones>area_of_patch_by_two

It works well.

Please make any correction and thanks in advance.

Answer 1

[Image Analyst]

Capital X is undefined in the code you gave and it may not like you to poof it into existence like that. Also, before that line do

whos X
whos x
whos patches_counter

and make sure that the number of rows in X matches the length of patches_counter.

[hamed abdulaziz]

Thank you for your fast response, I changed Capital X is to X_image_vector and I defined it at the beginning as :

X_image_vector = zeros(768,1024);

And elso changed x to

x_featurs_vector_patch

Please notice that the number of rows in X_image_vector matches the length of patches_counter where 32 * 32 =1024.

Thank you and I am waiting your answer

[hamed abdulaziz]

Image Analyst:I am waiting your answer

[Image Analyst]

As far as I can tell from your code, patches_counter = 1 while the number of rows of X_image_vector is 768. And x_featurs_vector_patch doesn't even show up in your code, which means I can't see it or debug it, so it appears the bet is for you to use the debugger and figure out what sizes don't match up. Can you do that?

[hamed abdulaziz]

At the beginning I also changed x to

x_featurs_vector_patch to become :

X_image_vector (patches_counter) = x_featurs_vector_patch;

and the initial value of patches_counter = 1 then it will be increased in this line :

patches_counter=patches_counter+1;

please could you correct it now,thanks.

[Image Analyst]

You never followed my suggestion of doing "whos" and you didn't attach your m-file. It's difficult for me to try to figure this out without being able to run it, because you have variables that are just suddenly there (like h_seg_image_cell{g,u}, etc.), and variables change names, etc. So I think the best bet is for you to step through this in the debugger. Don't you agree?

[hamed abdulaziz]

O.K I agree I will attach my m-file now,to take look on it please wait....

Answer 2

[hamed abdulaziz]

I attached my_tried_code.m ,Otsu code for segmentation,and image for test; please could you run it and correct it with my thanks. Hint: I am tried to implement the steps in the thesis (page 44 to 49 in pdf-file)on this website for local features :

thesis

[hamed abdulaziz]

Image Analyst:I am waiting your answer

[hamed abdulaziz]

could I have an answer?thanks