Detect all the rectangles in image

Hey all,

in the following image, all the rectangles are a little bit deformed due to power leakage effect. I want to detect all the rectangles and obtain the positions of the rectangles.

Is there a way to detect all rectangles?

Thank you

Answer 1

[Image Analyst]

It's trivial. Just use regionprops(). See my Image Segmentation Tutorial

[simon xu]

Thank Image Analyst very much!

After some trivial, I found using regionprops() is straightforward and helpful to detect the centroids of the rectangles. If I also want to obtain the corners (or boundaries) of each rectangle, is there a easy way to detect?

[simon xu]

I found the property 'BoundingBox' of regionprops() is helpful to detect the boundaries of rectangles.

Thanks a lot !

[Image Analyst]

regionprops() gives you [xLeft, yTop, width, height]. From that you can get row1 (y1), row2 (y2), col1 (x1), and col2 (x2), as well as the centroid (xLeft + width/2, yTop + height/2).

You can also get a lot of other things from regionprops().

[simon xu]

Thanks.

I'm wondering whether there is a useful method to remove the 'tails' (caused by power leakage effect) of the rectangles, illustrated in the following image

A naive way may be graythresh(), but it may be not satisfying, the following image is a binary image after thresholding

As a result, the rectangles which we detected may be not perfect, as shown below

So, if there is a more powerful method (more powerful than graythresh()) to remove these 'tails' of the rectangles, the detected shape would be more satisfying.

Is there a more powerful method?

Thank Image Analyst very much again!

[KSSV]

Thanks is accepting the answer.

[Image Analyst]

You can try imopen().

[simon xu]

The following image is the original image

imopen() is helpful to smooth the boundaries and get rid of the 'tails' of these rectangles, as shown below. The following image is the image after imopen()

There are no notable 'tails', and then I use imadjust() to adjust the contrast, and then I use graythresh() and im2bw() to binarize the image and obtain the following binary image.

The top-right rectangle is missing after graythresh(). So, is there a better way of thresholding and binarizing to aviod missing any rectangle?

Thanks a lot!

% here is my code
MainImg = imread('originalImg.jpeg');
se = strel('rectangle',[4,5]);
openImg = imopen(MainImg,se);
enhenImg=imadjust(openImg);
BinImg = im2bw(enhenImg,graythresh(enhenImg));

[Image Analyst]

How about you just compute the centroid, even with the tails, then take a vertical and horizontal line through the centroid and use find() to find the top and bottom lines and left and right columns of the rectangle?

[simon xu]

How to binarize the grayscale image in the attached grayscale.mat file?

My goal is binarizing the grayscale image to get the rectangles and get rid of noise.

Thank you very much!

Answer 2

[KSSV]

I = imread('myImage.jpeg') ;
[y,x] = find(I>50) ;

imshow(I)



hold on
idx = kmeans([x y],4) ;
for i = 1:4
    plot(x(idx==i),y(idx==i),'.')
    % GEt L and B of rectangles 
    x0 = min(x(idx==i)) ; x1 = max(x(idx==i)) ;
    y0 = min(y(idx==i)) ; y1 = max(y(idx==i)) ;
    %
    L = x1-x0 ;
    B = y1-y0 ;
    coor = [x0 y0 ; x0 y1 ; x1 y1 ; x1 y0] ;
    patch(coor(:,1),coor(:,2),rand(1,3))
    
end

[simon xu]

Thank KSSV very much!

But if the number of rectangles are arbitrary (maybe unknown), how to detect?

[islam elsayed]

can you please explain in brief

[islam elsayed]

@ KSSV