Best method for identifying the color

Hello. I have read a book in image processing. I have worked it out with matlab too. I want to find out the location (co-ordinates) of a particular color in the image. How should I do this? Which is the best method for color identification?

Thanks in advance.

[Walter Roberson]

How is the color denoted? If it is by name, then there is no definition for when that colorname stops and a different (or unnamed) color begins. Is (250, 11, 3) "red" or not? When does "olive green" end and "forest green" begin?

[i Venky]

It's very simple. All I need to do is to find red, yellow and blue colored boxes.

[Walter Roberson]

Is (250, 11, 3) "red" or not? What are the exact rules you want to use for determining whether something is "red"? "yellow"? "blue" ?

[Muhammad Muhammad]

How to identify yellow colour in images by Matlab software.

[Walter Roberson]

You start by defining exactly what yellow means to you. For example is sunlight yellow, or is it yellow-green (as science tells us), or is it white?

[Image Analyst]

Try the Color Thresholder app on the Apps tab of the tool ribbon.

Or check out some demos in my File Exchange.

Answer 1

[Image Analyst]

Color image analysis is not so easy, but it's one of my specialties as you can see from my logo and my File Exchange:

http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

Check out the color segmentation demos I have there. There are three different ways of detecting colors. You can also specify how close you want the colors to be or how different you'll allow them to be from each other (like how much spread there is in the detected color gamut).

[chadi fahmi]

Please answer this question : http://www.mathworks.com/matlabcentral/answers/257830-given-an-image-with-a-colored-object-in-it-i-need-a-code-to-display-the-name-of-the-colored-object-i with a code please

Answer 2

[Jonathan Sullivan]

Think of the color of a pixel being a vector in a 3 dimensional space. What you want to do is find the angle of the color vector of the pixel to the color vector of the ideal by using the dot product.

For example:

blue = [0 0 255];
pixel = [10 10 200];
ang_thres = 25; % degrees. You should change this to suit your needs
ang = acosd(dot(blue/norm(blue),pixel/norm(pixel)));
isBlue = ang <= ang_thres; % Apply angle threshold

You might also want to apply a magnitude threshold (i.e. is the pixel dark enough). This would filter out any really faint colors (i.e. [0 0 1]);

For example:

blue = [0 0 255];
pixel = [10 10 200];
ang_thres = 25; % degrees. You should change this to suit your needs
ang = acosd(dot(blue/norm(blue),pixel/norm(pixel)));
mag_thres = 64; % You should change this to suit your needs
mag = norm(pixel);
isBlue = ang <= ang_thres & mag >= mag_thres; % Apply both thresholds

Hope this helps!

[Alejandro Hernandez6]

thanks to all for this post. Anyway what means 'norm(blue)','norm(pixel)'? normalized ? My task is (given an image representing a floor plan) remove from it all green, red and blue pixels, considered noise, and later I will have to recognize all black segments to produce an floor plan.

[Image Analyst]

You're best off just attaching your image and telling us what you need to do, extract, or measure in a brand new question. Like what, exactly, is considered "noise."

You can also search the Image Analysis literature for papers on floor plans by clicking here.

Answer 3

[Walter Roberson]

Unfortunately, the hue-angle solution from Jonathan, and the hsv solution from Chandra, both will likely classify pink as being red, since pink is a saturated form of red. This is why it is important to define exactly what "red" and "yellow" and "blue" mean to you.

[Jonathan Sullivan]

With all do respect, Walter, I don't believe that the solution I gave will include pinks. Here, let's look at an example of all colors.

% Set user parameters
mag_thres = 0.5; % Set magnitude threshold
ang_thres = 10; % Set angular threshold
r = cat(4,1,0,0); % Define "red"
% Create all colors
[R G B] = meshgrid(linspace(0,1,100),linspace(0,1,100),linspace(0,1,100));
RGB = cat(4,R,G,B);
% Threshold
mag = sqrt(sum(RGB.^2,4));
ang = acosd(sum(bsxfun(@times,r,bsxfun(@rdivide,RGB,mag)),4));
f = ang < ang_thres & mag > mag_thres;
% Display Results
figure
image(cat(3,R(f),G(f),B(f)))

As you can tell, no "pinks" are included.

[Image Analyst]

I ran it and it looks like it gives a red Chirp or sawtooth function. Not sure I understand.

[DGM]

The example selects colors in a conical region around red.

% Set user parameters
mag_thres = 0.5; % Set magnitude threshold
ang_thres = 30; % Set angular threshold
r = cat(4,1,0,0); % Define "red"

% Create all colors
vv = single(linspace(0,1,200));
[R G B] = meshgrid(vv);
RGB = cat(4,R,G,B);

% Threshold
mag = sqrt(sum(RGB.^2,4));
ang = acosd(sum(bsxfun(@times,r,bsxfun(@rdivide,RGB,mag)),4));
f = ang < ang_thres & mag > mag_thres;

% Display Results
% this is a Nx1x3 stripe of colors within the selected volume
image(cat(3,R(f),G(f),B(f)))

% plot the slected volume
isosurface(R,G,B,f,0)
grid on
axis equal
ll = [0 1];
xlim(ll)
ylim(ll)
zlim(ll)
xlabel('R'); 
ylabel('G')
zlabel('B')
view(-73,35)

... but it doesn't seem generalized. The thresholds aren't relative to the specified color, so choosing any color other than red doesn't work.