Finding fitgmdist value range from histogram

Hello Altruists,

I have some data in 'rawdata.mat'

I want find the intensity distribution of the rawdata. For example, if we take the histogram of:

H = histogram(J);         

Looks like there are 2/3 Gaussian models. I want to take the one with maximum mean. But how do I find the range of values under that Gaussian distribution. For example the max Gaussian mean is at 234.

How do I choose the range of the Gaussian distribution?

GMModel = fitgmdist(double(J),2);

Answer 1

[Image Analyst]

See my File Exchange submission where you can fit the data to any number of Gaussians:

Fit Multiple Gaussians

[banikr]

what if the the number of Gaussians vary between 2 or 3? The function you gave has to specify 6 Gaussians. Existing fitgmdist works for me. I just need to find the range of values around mean to take the thresholding values.

[Image Analyst]

You didn't look at the code, did you? When you do, you'll see these lines right near the beginning:

% First specify how many Gaussians there will be.
numGaussians = 6;

Just change it to

% First specify how many Gaussians there will be.
numGaussians = 2;

and it will do 2 Gaussians.

[banikr]

Hi,

I read the code once you sent me.

But my ques was the histogram of data I have varies between 2 or 3 Gaussians for different samples. How do I define the data fits with 2 Gaussian or 3 Gaussian?

So far I am getting desired results from 'fitgmdist' and also looking through the codes you have sent. The other ques I had was to find the range/spread of the Gaussian.

Thanks.

[Image Analyst]

I'd probably start with the higher number and then sum the area under the curve of of each Gaussian. If the area is less than some small amount, you might want to ignore that Gaussian and recompute with one fewer number. In other words, start with 3 and if one of the 3 is too small, then repeat with only 2. If the area of one is still too small, then repeat with only 1 Gaussian.

[banikr]

Hi,

There is a variable called J in the rawdata.mat file I shared with this post. I changed:

y = double(J)
x = unique(double(J)) %there are 447 unique values in J -> [1, 483]
numGaussians = 6 % to check.
tFit = reshape(x, 1, []);
y = reshape(y, 1, []);

Kept the 'options' same.

I am getting the following error from 'fitgauss.m'

Undefined function or variable 'GetCallStack'.
Error in fitgauss (line 30)
	callStackString = GetCallStack(ME);
Error in @(lambda)(fitgauss(lambda,tFit,y))

Error in fminsearch (line 200)
fv(:,1) = funfcn(x,varargin{:}); 

[Image Analyst]

You're getting some sort of error. It's not being displayed because it seems some helper functions are not included. I'm including them here.

[Image Analyst]

Your data does not fit 3 or even 4 Gaussians well at all. Anyway, I'm attaching the code, test1.m, that works for 4 Gaussians and you can see that there is not a good fit for just Pure Gaussians.

There seems to be a ramp. You might try using fitnlm() and add in the ramp. I'm attaching a demo for fitnlm.

Adapt it to fit 3 Gaussians.