I determined the shape of the object in image, how i can determine the object's color?

My project is about color and shape sorter, I used just two objects a circle and rectangle, Also three colors, Red, Green and blue. I determined what the shape is, But my problem now is how i can determine what the object color is? For example like this image.

    clc
  close all
  clear
  %%Reading Image And Save It In Array.
  Image = 'Gertie-Ball-Image-Green-transparent.png';
  RGBImage = imread(Image);
  RGBImage = imresize(RGBImage,[500,500]);
  subplot(2,2,1)
  imshow(RGBImage)
  title('Input Image')
  %%Image Conversions To GrayScale.
  GrayImage = rgb2gray(RGBImage);
  subplot(2,2,2)
  imshow(GrayImage)
  title('GrayScale Image')
  %%Binarize The Image.
  BinaryImage = imbinarize(GrayImage);
  subplot(2,2,3)
  imshow(BinaryImage)
  title('Initial (Noisy) Binary Image')
  %%Remove Small Objects.
  BinaryImage = bwareaopen(BinaryImage, 50);
  % BinaryImage = imcomplement(BinaryImage); % Complement image
  subplot(2,2,4)
  imshow(BinaryImage)
  title('Cleaned Binary Image')
  %%Measure Properties Of BinaryImage Regions.
  [LabeledImage, NumberOfObjects] = bwlabel(BinaryImage); % Label Connected Components In 2-D Binary Image.
  Stats = regionprops('Table', LabeledImage, 'FilledArea','Perimeter');
  %%Get The Outermost Boundaries Of The Objects.
  FilledImage = imfill(BinaryImage, 'holes'); % Fill Image Regions And Holes.
  Boundaries = bwboundaries(FilledImage); % Trace Region Boundaries In Binary Image.
  %%Collect Some Of The Measurements Into Individual Arrays.
  Perimeter = [Stats.Perimeter];
  FilledArea = [Stats.FilledArea];
  %%Calculate Circularies.
  Circularities = Perimeter.^2./(4*pi*FilledArea);
  %%Determaine The Shape
  if Circularities(NumberOfObjects) < 1.10
       disp Circle
      elseif Circularities(NumberOfObjects) < 2.0
       disp Square
      else
       disp Something
  end
  %%Determaine The Color.

[Adam]

If you have a mask representing your shape then isn't this simple, or am I missing something?

You'd have to take an average colour from applying the mask I guess, or if it is literally just red, green or blue and is supposed to be only one of those then you could just take the colour of any pixel known to be in your shape and calculate its distance from red, green and blue.

How you want to define colour distance is a whole other question and may or may not be relevant, depending how far away from the primary colours your shape may be.

[odai kiwan]

Well, At my code i didn't use a mask for determining the shape, I calculate the circularizes and compare it with a threshold.

if a took a pixel and it's distance in this case the result will not be a 100% correct i guess. i don't know, Could you explain more please.

[Adam]

It all depends on how wide is the scope of objects and colours you will be presented with and what variation of colour they are allowed, but since you seemed to say only 3 colours are allowed - Red, Green and Blue, and that the shape must be one of these then a sample of pixels from inside the shape should do. Obviously just 1 is not very robust, but if your shape were uniformly very close to green or blue or red then it would still give the right answer. Since you appear to have white patches in your shape though a sampling of points with a mean (and exclusion of white) would work fine to give you the data to determine if it is red, green or blue.

Answer 1

[Benjamin Kraus]

Steve Eddins posted a series of blog posts that may help with this question:

• What color is green?
• Two-Dimensional Histograms
• a* and b*
• Freehand segmentation in the a*-b* plane

[Benjamin Kraus]

Based on the blog post by Steve Eddins, I suggest working in the L*a*b* color space:

The "a" in L*a*b* gives you a separation between red and green, while the "b" gives you separation between blue and yellow.

I would suggest reading the blog post for more context. The script below may not be perfect, you should definitely test it to make sure it works accurately, but it is a first shot that may work.

% Convert to L*a*b*
lab = applycform(RGBImage, makecform('srgb2lab'));
lab = lab2double(lab);

% Apply mask
a = lab(:,:,2);
a(~FilledImage) = NaN;
b = lab(:,:,3);
b(~FilledImage) = NaN;

% Average the a and b values
aavg = mean(a(:),'omitnan');
bavg = mean(b(:),'omitnan');

% Based on the blog post:
% blue = b < 0
% green = b > 0 & a < 0
% red = b > 0 & a > 0
if bavg < 0
    disp Blue
elseif aavg < 0
    disp Green
else
    disp Red
end

[Benjamin Kraus]

Another tip you may want to consider:

When you are calling imread, you are ignoring the third output, which provides transparency values:

[RGBImage,~,mask] = imread(Image);

The mask output gives you the transparency values from your PNG. I would recommend using this as a first step in determining your shape, as you can probably immediately throw away any values that are completely transparent (have a value of 0), and only keep values that are not transparent (have a value of 255).

Image = 'Gertie-Ball-Image-Green-transparent.png';
[RGBImage,~,mask] = imread(Image);
RGBImage = imresize(RGBImage,[500,500]);
mask = imresize(mask,[500,500]);
subplot(2,2,1)
im = imshow(RGBImage);
im.AlphaData = mask;
title('Input Image')

Answer 2

[murk hassan memon]

But what about when we have object in biconcave or disc shape? what value should be use in order to represent the biconcave and disc shape? i am working on microscopic blood images and normal blood cells are biconcave or disc in shape. kindly help me out.