GMM & EM algorithm for mammogram classification

I try to estimate a mammogram by Gaussian mixture model and then obtain log-likelihood image.I initialize Mu & Sigma with kmeans but when I use gmdistribution command there is an error : "the shared diagonal covariance must be a row vector with the same number of columns as Mu"

or when I want to write code for EM function I face to an error at the first step:"undefined function or method... for input argument of type 'uint8','double','single',...and I don't know what type I should use(my input is an image). I will appreciate if someone guide me what to do. thank you

Answer 1

[Image Analyst]

When it says "undefined function or method... for input argument" you need to have the function name (which you omitted and replaced by dot dot dot) actually defined somewhere, either in your m-file, or in another m-file on the search path.

[bahar h]

Thank you, and what about the first problem?actually I don't know if I am right in my programming or not for example is it right to use gmdistribution code for estimating GMM model?or I am not sure which command should I use for EM algorithm?is there any exact code or I must write a special function? Thanks

[Image Analyst]

I don't know about that one. My copy of MATLAB doesn't have anything on GMM.

[bahar h]

OK Thank you

Answer 2

[Ilya]

For your first problem, type 'help gmdistribution/gmdistribution' (note that you need to type gmdistribution twice to get the help for the constructor). Or take a look at the doc page here http://www.mathworks.com/help/toolbox/stats/gmdistribution.html The error message is pretty clear.

I also note that you do not need to use kmeans. You can execute gmdistribution.fit to estimate the mixture (yes, by the EM algorithm).

[bahar h]

Thank you for your answer.