value changes when running each time

x=imread('im9.bmp');
HSV = rgb2hsv(x);

H = HSV(:,:,1); H = H(:);
S = HSV(:,:,2); S = S(:);
V = HSV(:,:,3); V = V(:);
idx = kmeans([H S V], 3,'distance','sqEuclidean');

tabulate(idx)

the percentage for each idx gets changed every time i run the program,whay i am getting like this,kindly assist

[Jan]

BTW., The creation of your [H S V] can be abbreviated, such that you do not need H, S and V explicitly:

idx = kmeans(reshape(HSV, [], 3), 3,'distance','sqEuclidean');

This behaviour of kmeans is explained in the dcoumentation. It is strongly recommended to read it, especially in cases of problems: doc kmeans

[kash]

I used

idx = kmeans(reshape(HSV, [], 3), 3,'distance','sqEuclidean');

even then i get same error each cluster changes and different color is applied each time

[Walter Roberson]

kmeans still uses random initialization when you use reshape(), unless you provide the starting centroids like I showed below.

Answer 1

[Youssef Khmou]

hi kash,

i quote , from Wikipedia : " Commonly used initialization methods are Forgy and Random Partition" so each time you run the code : Random partition is the cause of that small change , so here is what you can do :

you run the code 10 times , each time you store that variable, it is then a R.V calculate its mean and STD , and conclude .

[Walter Roberson]

Right. kmeans uses random initialization unless you provide initial values.

[kash]

walter can oyu tell how to set initial values plz

[Walter Roberson]

kmeans([H S V], 3, 'distance', 'sqEuclidean', 'start', MatrixOfCentroids)

MatrixOfCentroids would have to be 3 (number of clusters) by 3 (number of columns of [H S V])

[kash]

Walter i tried yours but i get error

Undefined function or variable 'MatrixOfCentroids'.

what should it be

[Walter Roberson]

What does the documentation for kmeans() say should be passed in after the word 'start' ?

[kash]

i passed [],got error ,passed 'uniform' stll same result,colour changes but dont which matrix to pass for

k-by-p matrix of centroid

[kash]

I ALSO TRIED

C=HSV(102:104,102:104); for k by p matrix,but still colour keeps changing

[Teja Muppirala]

Are you sure you have typed everything correctly?

x = imread('peppers.png');
HSV = rgb2hsv(x);
C=HSV(102:104,102:104); 
idx = kmeans(reshape(HSV, [], 3),3,'distance','sqEuclidean','Start',C);
tabulate(idx)

For me, no matter how many times I run this, I always get the same result:

Value    Count   Percent
    1    74013     37.64%
    2    99481     50.60%
    3    23114     11.76%

[kash]

sorry the mistake i made was placed C inside single quotes,thanks