How to detect other types of colors?

Hello! I am developing a project that needs to identify silver wires through colors.

I found this code but it shows a "Threshold" for each color:

%%Initialization
redThresh   = 0.24; % Threshold for red detection
greenThresh = 0.05; % Threshold for green detection
blueThresh  = 0.15; % Threshold for blue detection

vidDevice = webcam;
hblob = vision.BlobAnalysis('AreaOutputPort', false,'CentroidOutputPort', true, ...
           'BoundingBoxOutputPort', true','MinimumBlobArea', 600,'MaximumBlobArea', 3000, ...
           'MaximumCount', 10);                              % Set blob analysis handling
hshapeinsBox = vision.ShapeInserter('BorderColorSource', 'Input port','Fill', true, ...
           'FillColorSource', 'Input port','Opacity', 0.4);  % Set box handling
htextinsRed  = vision.TextInserter('Text', 'Red   : %2d','Location',  [5 2], ...
                                   'Color', [1 0 0],'Font', 'Courier New','FontSize', 14);% Set text for number of blobs
htextinsGreen= vision.TextInserter('Text', 'Green : %2d','Location',  [5 18], ...
                                   'Color', [0 1 0],'Font', 'Courier New','FontSize', 14);% Set text for number of blobs
htextinsBlue = vision.TextInserter('Text', 'Blue  : %2d','Location',  [5 34], ...
                                   'Color', [0 0 1],'Font', 'Courier New','FontSize', 14);% Set text for number of blobs
htextinsCent = vision.TextInserter('Text', '+      X:%4d, Y:%4d','LocationSource', 'Input port', 'Color', [1 1 0], 'Font', 'Courier New','FontSize', 14);% set text for centroid
hVideoIn = vision.VideoPlayer('Name', 'Final Video', ... % Output video player
                                        'Position', [100 100 1300 750]);% Output video player
nFrame = 0; % Frame number initialization
%%Processing Loop
while(1)
  rgbFrame = snapshot(vidDevice); % Acquire single frame
  rgbFrame = flipdim(rgbFrame,2); % obtain the mirror image for displaying
  diffFrameRed = imsubtract(rgbFrame(:,:,1), rgb2gray(rgbFrame)); % Get red component of the image
  diffFrameRed = medfilt2(diffFrameRed, [3 3]); % Filter out the noise by using median filter
  binFrameRed = im2bw(diffFrameRed, redThresh); % Convert the image into binary image with the red objects as white
  diffFrameGreen = imsubtract(rgbFrame(:,:,2), rgb2gray(rgbFrame)); % Get green component of the image
  diffFrameGreen = medfilt2(diffFrameGreen, [3 3]); % Filter out the noise by using median filter
  binFrameGreen = im2bw(diffFrameGreen, greenThresh); % Convert the image into binary image with the green objects as white
  diffFrameBlue = imsubtract(rgbFrame(:,:,3), rgb2gray(rgbFrame)); % Get blue component of the image
  diffFrameBlue = medfilt2(diffFrameBlue, [3 3]); % Filter out the noise by using median filter
  binFrameBlue = im2bw(diffFrameBlue, blueThresh); % Convert the image into binary image with the blue objects as white
  [centroidRed, bboxRed] = step(hblob, binFrameRed); % Get the centroids and bounding boxes of the red blobs
  centroidRed = uint16(centroidRed); % Convert the centroids into Integer for further steps 
  [centroidGreen, bboxGreen] = step(hblob, binFrameGreen); % Get the centroids and bounding boxes of the green blobs
  centroidGreen = uint16(centroidGreen); % Convert the centroids into Integer for further steps 
  [centroidBlue, bboxBlue] = step(hblob, binFrameBlue); % Get the centroids and bounding boxes of the blue blobs
  centroidBlue = uint16(centroidBlue); % Convert the centroids into Integer for further steps 
  rgbFrame(1:50,1:90,:) = 0; % put a black region on the output stream
  vidIn = step(hshapeinsBox, rgbFrame, bboxRed, single([1 0 0])); % Instert the red box
  vidIn = step(hshapeinsBox, vidIn, bboxGreen, single([0 1 0])); % Instert the green box
  vidIn = step(hshapeinsBox, vidIn, bboxBlue, single([0 0 1])); % Instert the blue box
  for object = 1:1:length(bboxRed(:,1)) % Write the corresponding centroids for red
    centXRed = centroidRed(object,1); centYRed = centroidRed(object,2);
    vidIn = step(htextinsCent, vidIn, [centXRed centYRed], [centXRed-6 centYRed-9]); 
  end
  for object = 1:1:length(bboxGreen(:,1)) % Write the corresponding centroids for green
    centXGreen = centroidGreen(object,1); centYGreen = centroidGreen(object,2);
    vidIn = step(htextinsCent, vidIn, [centXGreen centYGreen], [centXGreen-6 centYGreen-9]); 
  end
  for object = 1:1:length(bboxBlue(:,1)) % Write the corresponding centroids for blue
    centXBlue = centroidBlue(object,1); centYBlue = centroidBlue(object,2);
    vidIn = step(htextinsCent, vidIn, [centXBlue centYBlue], [centXBlue-6 centYBlue-9]); 
  end
  vidIn = step(htextinsRed, vidIn, uint8(length(bboxRed(:,1)))); % Count the number of red blobs
  vidIn = step(htextinsGreen, vidIn, uint8(length(bboxGreen(:,1)))); % Count the number of green blobs
  vidIn = step(htextinsBlue, vidIn, uint8(length(bboxBlue(:,1)))); % Count the number of blue blobs
  step(hVideoIn, vidIn); % Output video stream
  nFrame = nFrame+1;
end
%%Clearing Memory
release(hVideoIn); % Release all memory and buffer used
release(vidDevice);
clear all;
clc;

I want to detect other kind of colors like: yellow, lime green, orange, purple, hot pink. But I do not know the "Threshold" for these colors according to the code.

Please, who can help me because I am a layman, I thank you very much!

[Raphael Ferreira]

follows the code file:

Answer 1

[Image Analyst]

You didn't show your image so it's hard to advise you. Basically if you have a megapixel image, you could potentially have anywhere from one up to a million colors. You might want to try rgb2ind() to use one algorithm that quantifies colors into a specified number of simple colors. Or you could use KNN to classify colors into the closest color from a training set of colors. See attached.

[Raphael Ferreira]

I can not show the actual image because my advisor does not allow me to show in public. Can I send you an email? my email: [e-mail]

[Image Analyst]

To get the gray/silver regions, convert to hsv color space with rgb2hsv, then select/threshold regions with saturation less than about 0.3 or so.

[Image Analyst]

I don't know what the detail (colored shapes) indicates. All the wires look the same to me.

[Raphael Ferreira]

Each one will indicate the channel of the graph. Example: ch1, ch2, chn ...

[Raphael Ferreira]

I already have the code of geometric shapes, lack detect colors.

[Image Analyst]

What does that mean? What code is that exactly?

The wires do not appear to have unique, distinct colors.

[Raphael Ferreira]

There are 4 geometric shapes for four colors. It can be any color less green (can be a green in lighter shade), blue and red. These objects will be on the electrodes, soon after their detection, each one would receive a name.

Understood?

[Raphael Ferreira]

Distinct colors of green, blue and red. Can be: Yellow green, orange yellow, Magenta and purple red.

[Raphael Ferreira]

Forgive me if I am not being clear with my doubt.

[Image Analyst]

Are you saying that in the "real" photo, that the wires would all be wrapped/tagged with a colored sticker oriented normal to the optic axis?

[Raphael Ferreira]

Yes! In this case, they are colored geometric shapes printed on the 3D printer, electrodepositions. What I need is to detect colors as mentioned before: yellow green, orange yellow, magenta and purple red.

I found this code that I posted at the beginning of the question, besides I found other similar ones that talk about this "Threshold", but I do not know how to do it for other colors.

[Raphael Ferreira]

this code does this:

I want to do this with other colors.

[Image Analyst]

Basically you have to find the distance in colorspace of your image to each of the reference colors.

distanceImage1 = sqrt((redChannel - redRef1).^2 + (greenChannel - greenRef1).^2 + (blueChannel - blueRef1).^2);
distanceImage2 = sqrt((redChannel - redRef2).^2 + (greenChannel - greenRef2).^2 + (blueChannel - blueRef2).^2);

and so on for any additional reference colors.

Then whichever reference color has the smallest distance, that's what color that pixel belongs to.

[Raphael Ferreira]

Ok, I will test.

[Image Analyst]

I'd take any of the color channels and run stdfilt() on it to identify uniform areas. Those will have a standard deviation of zero. Once you have those regions identified, call regionprops for each color channel and ask for 'MeanIntensity' to get the RGB values of each uniform region.

[Raphael Ferreira]

This way I get the "Threshold" of each color?