How detect, crop and straighten a particular shape

Hi all, I have to detect, crop and straighten this particular rectangular shape with rounded corner from the image, in order to classify this "film" used in dosimetry application.

These are different films belonging to different class:

I have already tried with common methods like is shown in these posts without any useful result:

https://it.mathworks.com/matlabcentral/answers/134898-detecting-rectangle-in-real-images

https://it.mathworks.com/matlabcentral/answers/35243-detecting-rectangle-shape-in-an-image

Any suggestion?

Answer 1

[Image Analyst]

See attached m-file that produces the figure below:

[riccardo isi]

Thank you again!! I have just modified two lines: when I have to rotate the image I choose the bilinear method because the image seems to keep a better quality, and I have added an if/else to choose the angle because some of these films sometimes are oriented in the opposite way.

Answer 2

[Image Analyst]

Look at the outer band of the image, like a rectangular "ring" within 40 pixels (or whatever) of the image. Take the mode of those pixels in each color band to get the most common color. Create masks for each color channel and then combine them to form an overall mask. Then invert that to find blobs that are not the background color. Then fill holes with imfill() and get the convex hull. That will get you the rounded rectangle. Then call bwboundaries to get the coordinates and identify the flat parts to get the angle, then call imrotate() to rotate/fix it.

[riccardo isi]

Thank you so much for your advices. I tried to follow, but I notice two problems:

1) I do not understand how get the angle for rotate the image.

2) There is some manner to remove the two supports that hold the film? I thought about some morphological operator like the erosion/dilation but it didn't work so good...

below there is the code I used:

img = imread('Test_0577.tif');
red = img(:,:,1); % Red channel
green = img(:,:,2); % Green channel
blue = img(:,:,3); % Blue channel
tollerance = 7;

figure; imshow(img), title('RGB image');

%%Red mask
r_mode = mode(red);
r_max1 = max(r_mode(:,1:40));
r_min1 = min(r_mode(:,1:40));
r_max2 = max(r_mode(:,1565:1605));
r_min2 = min(r_mode(:,1565:1605));

r_max = max(r_max1, r_max2)+tollerance;
r_min = min(r_min1, r_min2)-tollerance;

r_mask = zeros(size(red));
for k = 1 : size(r_mask,1)
    for i = 1 : size(r_mask,2)
        if (red(k,i)>r_min && red(k,i)<r_max)
            r_mask(k,i) = 1;
        end
    end
end

%%Green mask
g_mode = mode(green);
g_max1 = max(g_mode(:,1:40));
g_min1 = min(g_mode(:,1:40));
g_max2 = max(g_mode(:,1565:1605));
g_min2 = min(g_mode(:,1565:1605));

g_max = max(g_max1, g_max2)+tollerance;
g_min = min(g_min1, g_min2)-tollerance;

g_mask = zeros(size(green));
for k = 1 : size(g_mask,1)
    for i = 1 : size(g_mask,2)
        if (green(k,i)>g_min && green(k,i)<g_max)
            g_mask(k,i) = 1;
        end
    end
end

%%Blue mask
b_mode = mode(blue);
b_max1 = max(g_mode(:,1:40));
b_min1 = min(g_mode(:,1:40));
b_max2 = max(g_mode(:,1565:1605));
b_min2 = min(g_mode(:,1565:1605));

b_max = max(b_max1, b_max2)+tollerance;
b_min = min(b_min1, b_min2)-tollerance;

b_mask = zeros(size(blue));
for k = 1 : size(b_mask,1)
    for i = 1 : size(b_mask,2)
        if (blue(k,i)>b_min && blue(k,i)<b_max)
            b_mask(k,i) = 1;
        end
    end
end

figure; imshow(r_mask), title('red mask');
figure; imshow(g_mask), title('green mask');
figure; imshow(b_mask), title('blue mask');

%%Overall mask
mask = zeros(size(red));
for k = 1 : size(mask,1)
    for i = 1 : size(mask,2)
        if (r_mask(k,i) == 1 || b_mask(k,i) == 1 || g_mask(k,i) == 1)
            mask(k,i) = 1;
        end
    end
end

mask = imfill(mask, 'holes');
mask = ~mask;

figure; imshow(mask), title('overall mask');

%%Find boundaries
[B,L] = bwboundaries(mask,'noholes');

figure; imshow(img), title('image with boundaries');
hold on;
for i = 1:length(B)
    boundary = B{i};
    plot(boundary(:,2), boundary(:,1), 'r', 'LineWidth', 3)
end