How to apply colours of my choosing to labels in a binary image based on their individual areas

Hi, I'm trying to create a program that can pick out screws and bolts in an image but the screws and bolts will be a different colour each. I've segmented the image successfuly and used the 'bwlabel' function to apply labels to the images. I know the 'label2rgb' function can apply colours to all labels but I can only make it so all labels are a random colour. I'm a little stumped and any help would be greatly appreciated!

[Image Analyst]

Post your image and code and I'll do it for you. If you can see holes in the nuts, then you can use Euler number to pick out nuts vs. bolts. You might also be able to use area. Then make up your colormap and pass it in to label2rgb().

[Callum Moore]

Thank you so much, I've just attached the image and file, I've managed to fill the holes in the nuts its the differentiating between the items thats been the issue. Thanks again for the help!

[Image Analyst]

Can you get more uniform illumination, because that's causing a problem. Or is that part of your project - to do a background correction. Attached is what I have so far, but it's not segmenting right yet. I'm thinking of trying CLAHE with adapthisteq to flatten the background since imclose() didn't do such a great job. I'll also have to get two masks: one for brighter than the local background and one for darker since you have parts of the items that are both brighter and darker than the background.

It'a after midnight here. I'll look at it more tomorrow.

[Callum Moore]

The goal of the project is to segment the image, and then after segmentation differentiate the objects in the image (screws and washers) and apply a coloured layer to each type of object. Thank you for the progress you've made

[Image Analyst]

Is this homework? You didn't tag is as such. If it is, you need to add the homework tag. I'm sure your instructor doesn't want me to do the project for you then turn in my code as your own. So if it is, i'll probably only be able to provide ideas for algorithms, not actual code. So I'll have to leave it up to you if it's your school assignment.

Another thing I might try is to make a first guess at the background some how and then using scatteredInterpolant() to estimate what the background is in the non-background parts (the screws and washers). Then subtract the two images to find only the screws and washers. See attached demo for scatteredInterpolant.

Don't you have control over the image capture conditions? I mean, could you put the items on a black background, like black velvet? Or on a colored background like pure red or green?

[Callum Moore]

Sorry this was my first time asking a question on here and didn't realise I had to do that. Differentiating between the screws and washers is the last part of the section of the assignment I'm doing. After segmenting it as I did I'm meant to differentiate them by appropriate features and label them different colours accordingly. That's where my idea to colour them based on their area came from.

[Image Analyst]

It looks like you can ask regionprops for the major and minor axes lengths and then compute aspect ratios. Then you can create a color map for N colors, where you have N blobs, where the color is either red or blue depending on if the aspect ratio is larger than some value or not. Would you get in trouble if I gave you that code?

[Callum Moore]

I'm not sure but for now I'll attempt to implement what you've suggested and I'll let you know what I get. Thanks for all your help on this it's been really throwing me through a loop

[Callum Moore]

I've managed to get the aspect ratio successfuly thanks to your help, this may be a stupid question but how exactly do I interact with the blobs individually so I can apply the relevant colour map that I've made to them?

Answer 1

[Image Analyst]

Add this to your segmentaiton code:

% Get the aspect ratio of each blob.
props = regionprops(mask, grayImage, 'MajorAxisLength', 'MinorAxisLength', 'Area');
aMajor = [props.MajorAxisLength]
aMinor = [props.MinorAxisLength]
allAreas = sort([props.Area])
aspectRatios = aMajor ./ aMinor
numBlobs = length(props)
cmap = zeros(numBlobs+1, 3);
for k = 1 : numBlobs
	if aspectRatios(k) > 2 % Whatever value you want.
		cmap(k+1, :) = [1, 0, 0]; % Red.
	else
		cmap(k+1, :) = [0, 1, 0]; % Green.
	end
end
cmap
h3 = subplot(2, 2, 3);
labeledImage = bwlabel(mask);
imshow(labeledImage, []);
colormap(h3, cmap);

Note: my segmentation is not good because it didn't find the lower right washer nice and the screw and the washer are touching and I didn't try to separate them. But the colormap code works well.

[Callum Moore]

I see what I was doing wrong now, the last line where you called the colourmap is why mine wasnt applying correctly that makes me feel a little stupid I didn't realise that haha! Thanks again for your help you were really helpful with this!

[Image Analyst]

Yeah, it's a bit tricky. Starting a few releases ago the colormap applies to all axes on the figure and if you want a special colormap for just one particular axes, you either have to pass it in to imshow() with the colormap option, or pass the handle of the axes in to the colormap() function.

[imaggeprocessor01]

@Image Analyst Hello, I am facing a similar problem like the one in this post, could you please explain the above code you have provided in a bit more detail please? like what each line is doing?

[Image Analyst]

@Mathews

% Make measurements of each blob.
props = regionprops(mask, grayImage, 'MajorAxisLength', 'MinorAxisLength', 'Area');
% Get the major axis into a vector for convenience.
aMajor = [props.MajorAxisLength]
% Get the minor axis into a vector for convenience.
aMinor = [props.MinorAxisLength]
% Get the areas into vectors for convenience.
allAreas = sort([props.Area])
% Compute aspect ratios.
aspectRatios = aMajor ./ aMinor
% Get the number of blobs (regions) into a new variable for convenience.
numBlobs = length(props)
% Initialize a colormap that we will modify.
cmap = zeros(numBlobs+1, 3);
% For each blob number assign the color to be used for it.
% This depends on that blob's aspect ratio.
for k = 1 : numBlobs
    if aspectRatios(k) > 2 % Whatever value you want.
        cmap(k+1, :) = [1, 0, 0]; % Red.
    else
        cmap(k+1, :) = [0, 1, 0]; % Green.
    end
end
% Print out the colormap to the command window just for curiosity.
cmap
% Create a new axis in the lower left.
h3 = subplot(2, 2, 3);
% Turn the binary image into a labeled image where each blob now has
% a gray level that is equal to the blob's ID number (label number).
labeledImage = bwlabel(mask);
% Display the labeled image.  It will be in gray scale at this point.
imshow(labeledImage, []);
% Apply the colormap to turn each gray level (which is the same as the blob
% ID number) into a RGB color that depends on the blob's aspect ratio.
colormap(h3, cmap);

[imaggeprocessor01]

@Image Analyst Thank you! this is really useful. Do you know how we can compare and get the dice score of the output image of this and a given "ground truth" image (basically a grayscale image correspond to images in the input dataset with only 2 labelling values. The washers are labelled with 1, small screws value 2 and background value 0).

[Image Analyst]

Just split them up into two classes

testWashers = ismember(labeledImage, 1);
diceWashers = dice(testWashers, groundTruthWashers);
testScrews = ismember(labeledImage, 2);
diceScrews = dice(testScrews, groundTruthScrews);