how to define the starting point in gmmdistribution

Hi, I do not want to use the random sample as a starting point in gmmdistribution. Rather i want to define my initial guess. But when i am trying to do as per Mathworks instructions,it gives me erro. This is what my situation is:

• pixelList is 40*2 matrix
• k is equal to 2

C is a 2*2 integer matrix with initial guess. I believe,since i am fitting 2 clusters,i should define 2 starting points.Right?

 ob = gmdistribution.fit(pixelList,k,'Start',C,'Options',options,'Regularize',1e-5);
It gives me error:

Error using gmcluster (line 158)
'Start' must be 'randSample', a struct or a vector of integers.

Error in gmdistribution.fit (line 174)
    [S,NlogL,optimInfo] =...

Please reply and help

Answer 1

[kittu]

In your case the 'Start' should be a vector of 40 entries. These vector points define whether the corresponding pixel value is in cluster 1 or cluster 2. So it would look something like [1,2,2,1,2,1,2,1,2,1...]

[Niraj]

Thanks for the answer. But how do I generate this vector? Please reply.

Answer 2

[kittu]

you can generate this vector for instance by using Kmeans.

http://www.mathworks.nl/help/stats/kmeans.html