How to extract images from groundTruth object?

Hello all,

is there a way to extract (i.e. saving as separate images) all the individual objects/regions which were defined as ROI label with the Image Labeler app. and stored within a groundTruth object? The aim is to apply some batch-modifications (e.g. contrast or rotation) to these images to achieve a better result of the cascade object detector.

Thank you for your help!

[Joel Borja]

I would like to find an answer to this too.

Answer 1

[awezmm]

Rummy18 and Joel,

You can import the labeled images and mask at different class labels. Each element in a labeled image matrix is a number representing which class a pixel in the original image belongs to.

If you want to apply rotations or contrast for some type of learning, check out data augmentation:

https://www.mathworks.com/help/deeplearning/ref/imagedataaugmenter.html

[Theresa Pflüger]

awezmm, thanks for your input but I do not quite understand what you mean.

I am also trying to find a way to extract the labeled ROI (bounding boxes) into single images which I can save separately.

So, what do you mean by "importing labeled images and mask at different class labels"? Can you please explain the way to go. Is there a code I need to write and run for the gTruth (my labeled ROI) to generate separate images?

Thanks in advance for any help!

[awezmm]

Can you upload one of your label images here

[Theresa Pflüger]

Sure!

Below you can see one example of my labeled images from the ImageLabeler App.

Just to be clear again, I would like to save the four stomata as four separate images.

[awezmm]

Is this the actual labeled image? Can you please attach the labeled image after you export after labeling.

Answer 2

[Rummy18]

Hi Theresa,

you can export your labels using the "export labels" function from the Image Labeler App. If you choose to export the labels directly to your Matlab workspace, select "table" as the export format. Lets call this table "plant". The first column of this table indicates the path to your images used in the Image Labeler App and the second column (plant.Stomata) (=name of the table.name of your labels) shows the coordinates of your labeled rectangles (x,y,x',y'). x' and y' are the number of pixels you have to add to x and y, respectively, to describe the corners of your labeld rectangles.

You can now apply some loops and export the labels:

% set your directory to the folder, where you want to save the individual images 
cd 'path/folder';
% to find the number of pictures in your table "plant" use e.g.:
number_pictures = height(plant);
% find the number of labeled rectangles:
for c = 1:number_pictures
    number_rectangles = (numel(plant.Stomata{c,1})/4); 
    % export the rectangles as individual images
    for d = 1:number_rectangles
    name = plant.imageFilename{c}; % path to the image used in the Image Labeler session
    picture = imread(plant.imageFilename{c}; % imports the image into Matlab
    picture_crop = imcrop(picture,[plant.Stomata{c}(d,1),plant.Stomata{c}(d,2),plant.Stomata{c}(d,3),plant.Stomata{c}(d,4)]); % crops the individual label using the coordinates 
    imwrite(picture_cop,[name,num2str(d),'.jpg'],'Quality',100); % saves the rectangles as .jpg image to your directory
    clear name
    clear picture
    clear picture_crop
    clear number_rectangles
    end
end

Hope this helps.

Rummy18

[Jose Agustin Barrachina]

This is ALMOST what I need, but I have polygones and not rectangles... Any idea there?