How to get accurate clusters using k-means clustering in matlab - matlab

Question

I am doing a project on plant disease detection. I need to extract diseased parts from images of leafs but I'm not able to separate out diseased regions accurately using k-means. Specifically, the rest of the leaf is visible on the image with the diseased parts segmented. Here is the original image and image after extracting diseased parts:original image image after separating diseased parts
Here is the code I have written
b=imread('12.jpeg');
G=fspecial('gaussian',[200 250],1);
Ig=imfilter(b,G,'same');
figure,imshow(Ig);
conversionform = makecform('srgb2lab');
lab_img = applycform(Ig,conversionform);
figure,imshow(lab_img);
ab = double(lab_img(:,:,2:3));
nrows = size(ab,1);
ncols = size(ab,2);
ab = reshape(ab,nrows*ncols,2);
nColors = 2;
[cluster_idx, cluster_center] = kmeans(ab,nColors,'distance','sqEuclidean', ...,
'Replicates',3);
pixel_labels = reshape(cluster_idx,nrows,ncols);
figure, imshow(pixel_labels,[]), title('image labeled by cluster index');
segmented_images = cell(1,3);
rgb_label = repmat(pixel_labels,[1 1 3]);
for k = 1:nColors
color = lab_img;
color(rgb_label ~= k) = 0;
segmented_images{k} = color;
end
figure, imshow(segmented_images{1}), title('objects in cluster 1');
figure, imshow(segmented_images{2}), title('objects in cluster 2');
e=segmented_images{1};
figure,imshow(e);
conversionform = makecform('lab2srgb');
new_image=applycform(e,conversionform);
figure,imshow(new_image);
I want to extract only the diseased regions using K means clustering. I would be grateful if someone could help me with this. I am using matlab 2009a.

Answer 1

Here is a corrected code that will do what you expect:
function segmented_img = leaf_segmentation( original_img, nclusters )
original_img = im2double(original_img);
smoothed_img = imgaussfilt(original_img,1);
conversionform = makecform('srgb2lab');
lab_img = applycform(smoothed_img,conversionform);
ab_img = lab_img(:,:,2:3);
[nrows,ncols,~] = size(ab_img);
ab_img = reshape(ab_img,nrows*ncols,2);
cluster_idx = kmeans(ab_img,nclusters,'distance','sqEuclidean','Replicates',3);
cluster_img = reshape(cluster_idx,nrows,ncols);
%figure, imagesc(cluster_img), title('Clustering results');
segmented_img = cell(1,nclusters);
for k = 1:nclusters
segmented_img{k} = bsxfun( #times, original_img, cluster_img == k );
end
end
You can call it and visualise the results like so:
segmented = leaf_segmentation( original, 3 );
figure;
subplot(1,3,1), imshow(segmented{1}), title('Cluster 1');
subplot(1,3,2), imshow(segmented{2}), title('Cluster 2');
subplot(1,3,3), imshow(segmented{3}), title('Cluster 3');
Note that the order of the clusters may vary. You can order them a posteriori knowing that the leaf should be mostly green/yellow, and that the background should be mostly black.