How to detect contents of a beaker in an image

I have an image of a beaker in which I want to automatically detect the contents of the beaker. I cannot use the intensity for this because sometimes the intensity of the beaker content is the same as the one in the walls. Also the the length and height of the beaker content is not always the same. Is there a method that I can use to create a mask for just the beaker content without having to do it manually?

[DGM]

If the object properties vary, then it's important to know what properties don't vary. With this description and only one image, that information cannot be known. If the intensity of the fluid and the shape of the container are variable, to what degree may they vary? Is the boundary line between the two always present? If not, and if the fluid and container may have the same intensity, can the thickness of the bottom of the container be inferred from the visible geometry of the container? Is the background variable? Is it possible that the container is ever empty?

[Walter Roberson]

It would help to have additional images such as when the intensity were the same

[DGM]

I just realized this was from last year. I don't mind making projects out of answering old questions, but it's disappointing when you know you can never get clarification.

[MM]

Thank you both for showing interest in this problem.

I have attached an example where the intensity inside the contained and at the wall is the same and the height of the contained content is lower. The boundary line is always there so it might be possible to use that but it's not always well-defined. Any tips on how to solve this problem are still welcome.

[Walter Roberson]

Is it possible that the original images might be a lossless file format, not JPEG? You are running into notable artifacts because of JPEG.

filename = 'https://www.mathworks.com/matlabcentral/answers/uploaded_files/1287945/beaker_2.jpg';
img = rgb2gray(imread(filename));  %jpeg are almost always RGB
imshow(img)

img2 = imclearborder(imcomplement(img));
imshow(img2)

BW = img2 > 10;
BW = bwareafilt(BW, 1);
img3 = img2 .* cast(BW, 'like', img2);
imshow(img3)

img4 = img3;
img4(img4 >= 224) = 0;
imshow(img4)

[Walter Roberson]

filename = 'https://www.mathworks.com/matlabcentral/answers/uploaded_files/1212173/image.jpeg';
img = rgb2gray(imread(filename));  %jpeg are almost always RGB
imshow(img)

img2 = imclearborder(imcomplement(img));
imshow(img2)

BW = img2 > 10;
BW = bwareafilt(BW, 1);
img3 = img2 .* cast(BW, 'like', img2);
imshow(img3)

img4 = img3;
img4(img4 >= 224) = 0;
imshow(img4)

[Image Analyst]

Attach the original image, not a screenshot with tick marks, tick labels, and other stuff. We want just the bare image. Not a figure or .fig file, just the original .tiff, .jpg, or .PNG file.

[MM]

Attached you can find three ".mat files" containing the data matrix needed to make the bare figure. Each file has a different situation in which either the volume is different or the content of the beaker has different intensities. I will also try out DGMs suggestion which looks promising.

Answer 1

[DGM]

Here's an initial attempt. How well this works really depends on how vague the gap between the blobs can be. As Walter mentions, if there are images other than screenshots, that would probably help too.

% read the images
A = imread('bk1.png');
B = imread('bk2.png');

imshow(generatemask(A))

imshow(generatemask(B))

function mk = generatemask(inpict)
    % try to make things less sensitive to value
    inpict = imadjust(inpict);
    
    % binarize (idk how well this will work across all images)
    mk = imbinarize(inpict,'adaptive','sensitivity',0.7);
    
    % clean up the mask
    mk = imopen(mk,ones(5)); % try to break any small bridges
    mk = imfill(mk,'holes'); % fill any holes
    mk = bwareaopen(mk,10E3); % get rid of small blobs
    mk = bwpropfilt(mk,'extent',[0.85 1]); % keep blobs which are roughly grid-aligned rectangles
    mk = bwareafilt(mk,1); % keep the largest remaining blob
end

As to how to apply watershed segmentation to this problem, I'm not the one to ask, but if it's a particularly good use-case, I'd like to see an example applied to this specific problem.

Answer 2

[Mandar]

Looking at the image, the watershed-based segmentation approach could be a good starting point to segment out the content in the bicker. Refer the following MATLAB documentation link to know more about watershed segmentation.

Link: https://in.mathworks.com/help/images/marker-controlled-watershed-segmentation.html