How to subtract background from a leaf image .

I have a leaf data set.so i want to remove background from each leaf in that dataset.how i remove this

Answer 1

[Sulaymon Eshkabilov]

You shall use ROI mask: https://www.mathworks.com/help/images/specify-roi-as-binary-mask.html

Here are a couple of threads on this issue:

https://www.mathworks.com/matlabcentral/answers/501391-extract-leaf-from-background?s_tid=answers_rc1-1_p1_MLT

https://www.mathworks.com/matlabcentral/answers/237106-extracting-leaf-from-background?s_tid=answers_rc1-2_p2_MLT

[Alia Rehman]

I didn't want to use roi because I want to make automated system

Answer 2

[Image Analyst]

That may or may not be appropriate. Show an example of your image with a leaf in it, and your background image so I can evaluate it. Also say WHY you think you need to do this.

[Alia Rehman]

This my leaf rust image. I want to subtract background (all part of the image except leaf) and detect foreground (leaf). I have done it by lazzy snapping method but that was restricted to one leaf. After that I have used freehand roi to select roi of an image ,convert into binary and used that as a mask. Give me better result for all image but my project will not be automated by doing that. So need suggestions what can I do for background subtraction??? Thank you

[Image Analyst]

If you want to extract just the images and not the background and text, I'd get a mask that is just for the pure white, and then invert it and take blobs larger than 10% of the image only. Then look over all pictures, saving them to disk. Something like (untested)

[r,g,b] = imsplit(rgbImage);
[rows, columns, numberOfColorChannels] = size(rgbImage);
whiteMask = (r == 255) & (g == 255) & (b == 255);
% Extract only blobs bigger than 10% of the image.
minAcceptableArea = rows * columns * 0.10; % 10% of the area.
pictureMasks = bwareaopen(~whiteMask, minAcceptableArea);
% Give each blob an ID number so we can extract those pixels one blob at a time.
[labeledImage, numBlobs] = bwlabel(pictureMasks);
% Loop over all blobs, cropping and saving to disk.
for k = 1 : numBlobs
    % Get a mask of just this one picture.
    thisPictureMask = ismember(labeledImage, k);
    % Find its bounding box.
    props = regionprops(thisPictureMask, 'BoundingBox');
    % Crop that sub image out of the full image.
    croppedImage = imcrop(rgbImage, props.BoundingBox);
    figure;
    imshow(croppedImage);
    % Create filename
    baseFileName = sprintf('Image #%3.3d', k);
    % outputFolder is defined at the start of the program.
    fullFileName = fullfile(outputFolder, baseFileName);
    % Write the image to disk
    imwrite(croppedImage, fullFileName)
    fprintf('Saved "%s".\n', fullFileName);
end

[Alia Rehman]

what is the output folder?? Hoe can i define this

[Alia Rehman]

This is the output that comes from my below code. But its output is not good, What changes i do in my code.

[Image Analyst]

The connected region is green, and could possibly be part of the leaf, as far as the color segmentation knows. Since you have expert knowlege I suggest you use drawfreehand to mask out any part that is connected by mistake. Demo attached.

[Alia Rehman]

i have done it by freehand tool but i want to automate my system so this method is not good

[Alia Rehman]

Sir there is another method on the basis of which i subtract background of images like that. I have a leaf dataset and i have subtract background but not by freehand tool.

[Alia Rehman]

Here its another leaf image but here also the same result. There is no greenery in th ebackground attached with the leaf. Why??? Sir plz help

[Image Analyst]

You're going to have to adjust your thresholds. Maybe the background is slightly greenish but it's also bright so lower your upper V threshold to exclude the white regions.

[Alia Rehman]

How can I directly extract rust part from leaf. making all image black except rust part of leaf???

[Image Analyst]

@Alia Rehman, again, it's the Color Thresholder. Once you start using it I think you'll understand.