how can i find the displacement in pixels for the 1st, 3rd and 5th squares in the picture

the bridge in the picture was loaded 4 times and a picture was taken after each load, im trying to find the displacement in pixels for the 1st, 3rd and 5th squares in the picture. the 2nd and 4th squares are fixed. i have the displacement readings from the dial gauges but i also need the pixel displacement. im not familiar with image processing techniques, any suggestion would be appreciated.

Answer 1

[Thorsten]

% number here found by visual inspection
I = im2double(imread('1.JPG'));

index_center = size(I,2)/2+100:size(I,2)/2+250;
offset = [-1910 0 1854]; % left, center, right
Noffsets = numel(offset);
Nimages = 5;

% preallocate result
dx = zeros(Nimages, Noffsets);
dy = zeros(Nimages, Noffsets);

for o = 1:3
  for im = 1:Nimages
    I = im2double(imread([int2str(im) '.JPG']));

      % crop central part of image with the black square
      Ic = I((1610:1750),index_center + offset(o));
      subplot(1,3,2), imshow(Ic)
      title({['Square from image ' int2str(im) ], 'to be matched to S1.'})
      drawnow

      if im == 1 % just the black square
        S = Ic;
        s = 40;
        S = S(s:end-s, s:end-s);
        szS = size(S);
        % find square of 
        subplot(1,3,1), imshow(S)
        title('S1: square from image 1')
        drawnow
      end

      szIc = size(Ic);
      sz = szIc - szS;

      R = nan(sz(1), sz(2));

      % find best match; simple brute force method
      for i = 1:sz(1)
        for j = 1:sz(2)
          R(i,j) = sum(sum(abs(S - Ic(i:i+szS(1)-1, j:j+szS(2)-1))));
        end;
      end

      [dy(im,o), dx(im,o)] = ind2sub(size(R),find(R == min(R(:))));
    end

end

% relative displacement: subtract dy of image 1
dy_rel = bsxfun(@minus, dy(2:end,:), dy(1,:));

subplot(1,3,3)

plot(2:5, dy_rel, 'o-')

axis([2 5 0 12])
xlabel('Image #')
ylabel('Offset in pixels')
box off

legend({'left', 'center', 'right'})
legend boxoff

[ZERO]

thanks for the code, i tried to run it there is an error in "R(i,j) = sum2(abs(S - Ic(i:i+szS(1)-1, j:j+szS(2)-1)));" (line 42).

[Thorsten]

Sorry, I used sum2, one of my helper functions. I fixed it in the code above. Please try again.

[ZERO]

it runs perfectly , thank you.

where did you get 100 and 250 from. what do they represent? index_center = size(I,2)/2+100:size(I,2)/2+250;

also how did you get the offset values? offset = [-1910 0 1854]; % left, center, right

[Thorsten]

Just by trial and error and visual inspection.

Answer 2

[KSSV]

You have to read the image first using imread, specify the location where you want pixels.

Pseuodcode:

RGB = imread('myimage');   % read the color values of the image 
% specify the row and column position where you want to extract pixels 
c = [12 146 410];           
r = [104 156 129];
pixels = impixel(RGB,c,r) ;  % get pixels at the given locations

[ZERO]

im not sure what you mean by "specify the row and column position where you want to extract pixels". may be my question wasn't clear enough. im trying to find the number of pixels that a square moved down after the load. the location of the square in the image with no load is considered zero, after the load is applied, how many pixels did the square move down?

[KSSV]

You will know that, only after getting the pixles at your square region, and finding the difference in each case. Isn't it?

[ZERO]

ok, let me ask you this, how would i specify the location that i want in pixels, how did you get the values for r, c?

[Image Analyst]

I told you how to get the centroid of the square rings in my answer. Maybe if I have time later I can do it for you, if you can't.

[ZERO]

thank you for trying to answer, i wouldn't have asked this question if i knew what i need to do.

Answer 3

[Image Analyst]

This is pretty easy.

1. First I'd mask out the vertical lanes where your reference squares are, just to make sure there's no clutter anywhere that could be confused with the squares.
2. Then threshold. This could get the white squares and the white sticks so
3. label and call regionprops and look at the bounding box width and height and throw out (with ismember) any blobs where the width to height ratio is not close to 1.
4. Then call regionprops again with the "good" squares getting the centroids. You're basically done now.

See my Image Segmentation Tutorial for a demo of this: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862