Creating a bounding box that is not a perfect rectangle

Currently, my code allows me to select a ROI in an image and then it creates a bounding box. This bounding box, however, is a rectangle while I would like the cropped image to remain as the ROI shape and not a perfect rectangle.

clear varaiables, close all
if ~exist("image_analysis.mlx") 
    % 1. Load the file and scale it 
    before = "blue_close";  white = "white_close"; after = "group1_close";
    A = imresize(imread(before,'jpg'),1.00);
    B = imresize(imread(white,'jpg'),1.00);
    C = imresize(imread(after,'jpg'),1.00);
    
    %     % 2. manually specify the ROI using the original image
    imshow(A)
    h = drawpolygon('FaceAlpha',0);
    h.Color = 'yellow';
    % find a bounding box of ROI to further cut the image smaller
    % bbox = [xmin ymin xmax ymax]
    hmin = min(h.Position); hmax = max(h.Position);
    bbox = [hmin hmax-hmin]; 
    %     3. Crop ROI for both image a and image b using the bounding box
    Cb = imcrop(B,bbox);  
    Ca = imcrop(A,bbox);  
    Cc = imcrop(C,bbox);
    
    clear A B C
    save sqa.mat  
else
    load sqa.mat
end

Thanks!

[Matt J]

Images have to be rectangular.

[ISABELLE GUNDRUM]

Is there a way for the RGB values in a non-rectangular region of an image to be analyzed?

[Matt J]

Yes, certainly. Exactly how though depends on the analysis to be done.

[Walter Roberson]

For example if you were to create a mask from the ROI, and you were to double() that so that the values become numeric 0 and numeric 1 instead of logical 0 (false) and logical 1 (true), then you could pass that numeric array to regionprops as the "label" array, and pass in a grayscale image as the next parameter. You can then ask about the properties and it will only calculate the properties for the marked locations.

That said... if you were calculating a property such as Solidity which has to do with the area occupied relative the area of the bounding box, then the bounding box is still going to be calculated as a rectangle.

Answer 1

[DGM]

As Matt says, it depends what you intend to do with the image region. If all you intend to do is some sort of color analysis, then:

% a test image
inpict = imread('peppers.png');

% create ROI object
% i'm explicitly setting the point list since i'm doing this in the forum
imshow(inpict)
ROI = drawpolygon(gca,'position',[295 184;256 207;281 217;298 222;304 193]);

% create mask from the ROI, expand as necessary
mask = repmat(createMask(ROI),[1 1 size(inpict,3)]);

% get samples from selected locations
% this is a Mx3 array of RGB values
sampledpixels = reshape(inpict(mask),[],3)

sampledpixels = 962×3

255 183 0 255 186 0 255 185 0 255 186 0 255 185 0 255 181 0 255 184 0 255 182 0 255 183 0 255 181 0

Of course, this won't be of much help if you're trying to do anything spatial.

Answer 2

[Image Analyst]

See attached manual-drawing ROI definition examples.