plot 2 gaussians obtained by gmdistribution.fit

[Marcos Belmonte]

Hi, I'm using gmdistribution.fit to obtain two gaussians in order to contrast my homogeneous image.The information I obtain is:

Mixing proportion: 0.461726
Mean: 195.8348

Component 2:
Mixing proportion: 0.538274
Mean: 68.9320

How could I represent the two gaussians in order to obtain the standard deviation of each gaussian at the bottom (base))?

Thanks

[Marcos Belmonte]

"Marcos Belmonte" <marco...@gmail.com> wrote in message <jjkdt0$9el$1...@newscl01ah.mathworks.com>...
> Hi, I'm using gmdistribution.fit to obtain two gaussians in order to contrast my homogeneous image. The data are from an histogram. The information I obtain is:

[Marcos Belmonte]

"Marcos Belmonte" <marco...@gmail.com> wrote in message <jjkehm$bdq$1...@newscl01ah.mathworks.com>...

The data are from an histogram so I have a unidimensional object. Is there a way to represent with pdf o cdf like ezsurf(@(x,y)pdf(obj,[x y]),[-10 10],[-10 10]) in http://www.mathworks.es/help/toolbox/stats/brklrj3.html#brklr93-1?

Thanks

[Tom Lane]

>> > Hi, I'm using gmdistribution.fit to obtain two gaussians in order to
>> > contrast my homogeneous image. The data are from an histogram. The
>> > information I obtain is:

..

>> > How could I represent the two gaussians in order to obtain the standard
>> > deviation of each gaussian at the bottom (base))?

If G is your fitted object, then G.Sigma is an array of the covariances. The
standard deviations are the square roots of the diagonal elements. Note this
is a 3-D array and you have 1-D data apparently, so you may find it
convenient to use squeeze(G.Sigma) to get a vector of the two variances.

> The data are from an histogram so I have a unidimensional object. Is there
> a way to represent with pdf o cdf like ezsurf(@(x,y)pdf(obj,[x y]),[-10
> 10],[-10 10]) in
> http://www.mathworks.es/help/toolbox/stats/brklrj3.html#brklr93-1?

I don't understand this question, but pdf(G,x) is the mixture pdf evaluated
at x. You could plot this on top of a histogram if you scale the histogram
to have total area 1.

-- Tom

[philb]

"Tom Lane" <tl...@mathworks.com> wrote in message <jjl2i5$gtm$1...@newscl01ah.mathworks.com>...

and how could I just plot one of the two gaussians with pdf ?

[Tom Lane]

> and how could I just plot one of the two gaussians with pdf ?

Take a look at the following, which plots the mixture distribution then the
two separate Gaussians individually:

load fisheriris
x = meas(:,1);
gm = gmdistribution.fit(x,2)
xx = linspace(4,8)';

plot(xx,pdf(gm,xx))

for j=1:2
line(xx,gm.PComponents(j)*normpdf(xx,gm.mu(j),sqrt(gm.Sigma(j))),'color','r')
end

-- Tom