Image Segmentation not working properly

I have written a code to identify the green and black color seeds and then calculate the ratio of the area occupied by green seeds to that of black seeds. I used the color thresholder app to identify the two colors and generated 2 functions- one to segment the green color and another to segment the black. I saved the 2 segmented images in Binary format. Then I found each of the area using bwarea() and hence found out their ratio.

However, the result so obtained is not accurate. For 50:50 ratio (black:green), I get 42:68 and for 70:30, I get 40:60.

Also I am unable to use this function on other similar images with different seeds proportion. Displaying both the binary images using imshowpair(bImg, gImg, 'falsecolor'), I get :

Please tell me where I have gone wrong and what is the solution? Thank you.

[KALYAN ACHARJYA]

Still remembered @Image Analyst Sir had advised you, have you followed?

[Shashank]

Sorry what? I don't get you.

[Image Analyst]

I do remember coding up a demo for this beans image recently. But I don't see it under Shashank's posts, so either it somehow got deleted or someone else with the same photo asked the same question (if so, is it a homework assignment?).

[Shashank]

No sir it is not an assignment. I am studying the degree of mixing of these seeds in a gas-solid fluidized bed. So to calculate and verify the mixing factor, I am using Image Analysis. Could you please let me know if this is the correct method to approach or is there some other better way? Thank you.

[Shashank]

@Image Analyst Sir, he is my group mate. We are together working on this project to study the degree of mixing. I couldn't understand the method that you have used in the demo, so I tried using a different method but still did not get the result. Please help me learn as I am new to Image Processing. Thank you.

[Image Analyst]

See attached demo on k-means. You should have a known number of classes, like 3 for white, black, and green. But after color segmentation you'll still have to do some spatial cleanup because of shadows, white specks on the seeds, specular reflections, etc.

[Shashank]

I used the code and got a result like this.

What I am thinking is, is there a way to superimpose "Class 2 Image Masking Original" and "Class 5 Image Masking Original" so that we can find the area occupied by each color seed? Then we could determine the ratio of their areas?

[Image Analyst]

Yes, you can OR the masks to combine classes

mask25 = mask2 | mask5;

[Shashank]

Is this the correct way to extract mask-2 and mask-5 from the for loop? If not, how should I extract them?

for c = 1 : numberOfClasses
  
  subplot(3, numberOfClasses, c + numberOfClasses);
  thisClass = allClasses(:, :, c);
  imshow(thisClass);
  caption = sprintf('Image of\nClass %d Indexes', c);
  title(caption, 'FontSize', fontSize);

  maskedRgbImage = bsxfun(@times, rgbImage, cast(thisClass,   
       'like', rgbImage));
  
  subplot(3, numberOfClasses, c + 2 * numberOfClasses);
  imshow(maskedRgbImage);
    if c==2
        x = maskedRgbImage;
    end

      if c==5
          y = maskedRgbImage;
      end

  caption = sprintf('Class %d Image\nMasking Original', c);
  title(caption, 'FontSize', fontSize);
  
  % Make indexed image
  indexedImageK(thisClass) = c;
end

mask25 = x | y;
imshow(mask25);

    if true
      % code
    end

[PeterHaultan]

Does this help?

   I = imread('lena.jpg');
      I = rgb2gray(I);
      [r,c] = size(I);
      L = kmeans(I(:),5);
      I = reshape(L,[r,c]);
      mask25 = (I==2)|(I==5);