How to find pixels of one cluster that are enclosed by pixels of another cluster?

I have this image that I have clustered using k-means method. Here, k = 3. Code provided by KSSV (contributor from matlab community). This is the code:

I = imread('image 2.jpg') ; 
[y,x] = find(I) ; 
idx = I>0 ; 
z = I(idx) ; 
% Use kmeans 
G = 3 ; % number of groups 
idx = kmeans(z,G) ; 
% plot
imshow(I)
hold on
for i = 1:G
    plot(x(idx==i),y(idx==i),'.')
end
% figure
% scatter(x,y,10,z,'s','filled') ; colorbar

So, after clustering I get 'image 3.jpg' attach. There are 3 clusters in the image: cluster 1 (light orange), cluster 2 (dark orange) and cluster 3 (blue).

I want to scan the image column by column to get the coordinates for the pixels which belong to cluster 2 and are enclosed by cluster 1 pixels.

Can someone please help with the code?

Answer 1

[KSSV]

What you need to use is boundary and inpolygon. You need not to scan through every column. Check the below code.

I = imread('image 2.jpg') ; 
[y,x] = find(I) ; 
idx = I>0 ; 
z = I(idx) ; 
% Use kmeans 
G = 3 ; % number of groups 
idx = kmeans(z,G) ; 
% Sort the groups 
[val,idx] = sort(idx) ; 
x = x(idx) ; y = y(idx) ; z = z(idx) ; 
idx = val ; 
% plot
imshow(I)
hold on
for i = 1:G
    plot(x(idx==i),y(idx==i),'.')
end

%% Get boundary of each cluster 
xx = cell(G,1) ; yy = cell(G,1) ; 
for i = 1:G
    xi = x(idx==i) ; yi = y(idx==i) ; 
    
    bd = boundary(x(idx==i),y(idx==i)) ; 
    xx{i} = xi(bd) ; 
    yy{i} = yi(bd) ; 
end
%% Get the points lying inside region 1
idx_1 = inpolygon(x,y,xx{1},yy{1}) ; 
figure
hold on
plot(xx{1},yy{1})
plot(x(idx_1),y(idx_1),'.r')
title('Points lying inside Region 1')

[Ashutosh Gowarikar]

I probably don't understand your code. But it doesn't really get what I am looking for. Please correct me if I am wrong. But I think it takes cluster boundaries individually for all three clusters and plots all the pixels irrespective of cluster group within a specified boundary (in this case, region 1).

If you see the image 2 attached, will notice that blue some blue pixels are bounded by cyan pixels. I only want to highlight such blue pixels.

I = imread('image 2.jpg') ; 
[y,x] = find(I) ; 
idx = I>0 ; 
z = I(idx) ; 
% Use kmeans 
G = 3 ; % number of groups 
idx = kmeans(z,G) ; 
% Sort the groups 
[val,idx] = sort(idx) ; 
x = x(idx) ; y = y(idx) ; z = z(idx) ; 
idx = val ; 
% plot
imshow(I)
hold on

    plot(x(idx==2),y(idx==2),'.b')
    plot(x(idx==3),y(idx==3),'.c')

Answer 2

[Image Analyst]

Get a binary image of each cluster by itself. Then to find which in cluster2 are also in cluster1, do an AND operation

inBoth = cluster2 & cluster1;