Fit 2D Gaussian Function

Hello,

I am having trouble with the code found on this site: https://www.mathworks.com/matlabcentral/fileexchange/37087-fit-2d-gaussian-function-to-data I made the modifications below so that I could use it to run my data:

%%---------User Input---------------------
MdataSize = 256; % Size of nxn data matrix
% parameters are: [Amplitude, x0, sigmax, y0, sigmay, angel(in rad)]
x0 = [1,128,30,128,30,0]; %Inital guess parameters
xs = x0;
% % x = [2,2.2,7,3.4,4.5,+0.02*2*pi]; %centroid parameters
% % noise = 10; % noise in % of centroid peak value (x(1))
noise = 0;
InterpolationMethod = 'nearest'; % 'nearest','linear','spline','cubic'
FitForOrientation = 0; % 0: fit for orientation. 1: do not fit for orientation

%%---Generate centroid to be fitted--------------------------------------
% % xin = x; 
% % noise = noise/100 * x(1);
[X,Y] = meshgrid(-MdataSize/2:MdataSize/2);
xdata = zeros(size(X,1),size(Y,2),2);
xdata(:,:,1) = X;
xdata(:,:,2) = Y;
% % [Xhr,Yhr] = meshgrid(linspace(-MdataSize/2,MdataSize/2,300)); % generate high res grid for plot
% % xdatahr = zeros(300,300,2);
% % xdatahr(:,:,1) = Xhr;
% % xdatahr(:,:,2) = Yhr;
%---Generate noisy centroid---------------------
% % Z = D2GaussFunctionRot(x,xdata);
% % Z = Z + noise*(rand(size(X,1),size(Y,2))-0.5);
Z = load('AVG PS of 17_12005_ROI1-4.txt');

My data (Z) is a 256x256 txt file with data at each point. I would like to fit a 2D gaussian, but am getting the following error message:

Error using lsqcurvefit (line 251)
Function value and YDATA sizes are not equal.

Error in Gaussian_Fit_PowerSpectrum (line 57)
    [x,resnorm,residual,exitflag] = lsqcurvefit(@D2GaussFunctionRot,x0,xdata,Z,lb,ub);

I have the “D2GaussFunctionRot.m” and “D2GaussFunction.m” codes loaded in the directory and cannot figure out how to proceed. Any help would be appreciated. Thank you!

Answer 1

[Jordan Lui]

I noticed in the original code that while MdataSize is 50, the xdata matrix ends up being 51x51x2. This causes many other resulting variables to be "n+1" dimensioned. Are you sure your code modifications reflect this?

[Ham Man]

does this code just work for square matrix?

Answer 2

[Trippy]

You can fix it by doing the following

MdataSize = 255

The idea is function @D2GaussFunctionRot when the input is x0 and xdata, will give out an output of size nXm, which is the exact size of your image/ Z.

[Ham Man]

I'm trying this code (Gero Nootz ) to fit Gaussian model to a 2d matrix. I'm getting this error:

Error using snls (line 47)
Objective function is returning undefined values at initial point. lsqcurvefit cannot continue.

Here is the data i'm using:

x0 = [1,0.2,0.5,0.2,0.5,0];
x = x0;
noise = 0;
InterpolationMethod = 'nearest';
FitForOrientation = 0;
xdata(:,:,1) =

       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000
       NaN   -1.0000   -0.6000   -0.2000    0.2000    0.6000    1.0000


xdata(:,:,2) =

       NaN       NaN       NaN       NaN       NaN       NaN       NaN
   -1.0000   -1.0000   -1.0000   -1.0000   -1.0000   -1.0000   -1.0000
   -0.7143   -0.7143   -0.7143   -0.7143   -0.7143   -0.7143   -0.7143
   -0.4286   -0.4286   -0.4286   -0.4286   -0.4286   -0.4286   -0.4286
   -0.1429   -0.1429   -0.1429   -0.1429   -0.1429   -0.1429   -0.1429
    0.1429    0.1429    0.1429    0.1429    0.1429    0.1429    0.1429
    0.4286    0.4286    0.4286    0.4286    0.4286    0.4286    0.4286
    0.7143    0.7143    0.7143    0.7143    0.7143    0.7143    0.7143
    1.0000    1.0000    1.0000    1.0000    1.0000    1.0000    1.0000
    
zdata =

   NaN   NaN   NaN   NaN   NaN   NaN   NaN
   NaN     0   123   388   125     0   NaN
   NaN    30   614   787   570    53   NaN
   NaN   133   765   897   744   180   NaN
   NaN   199   832   891   737   270   NaN
   NaN   143   810   903   782   210   NaN
   NaN    27   590   857   659    63   NaN
   NaN     0   100   442   176     0   NaN
   NaN   NaN   NaN   NaN   NaN   NaN   NaN
   
[x,resnorm,residual,exitflag] = lsqcurvefit(@D2GaussFunction,x0,xdata,zdata);

xdata are meshgrid outputs from r/R of pipe cross section and zdata represents the particle distribution in pipe cross section. I want to fit a 2D Gaussian model to the zdata.

I really appreciate any help or thought. Thanks in advance!