fitgauss.m

混合高斯概率密度模型,其参数估计可以通过期望最大化( EM) 迭代算法获得。

function R=FitGauss(im0,s,histO,K)
%R=FitGauss(im0,s,histO,K)
%fit quadratic function to the Bhattacharaya coefficient -rho
%for calculating rho we use:
% im0   -current RGB image
% s     -elliptical region s=[xpos,ypos, xscale, shear, yscale]
% histO -object histogram
% kernel-smooth kernel
%
%R assumed diagonal 5x5 matrix
%
%University of Amsterdam, The Netherlands
%Author:Zoran Zivkovic, www.zoranz.net
%Date: 17-5-2005

nStates=length(s);
R=eye(nStates);
fClip=0.005;

nSamples=2;
rSamples=zeros(1,nSamples);
%pos
[xt,V]=SToMeanVar(s);
rho0=Similarity(im0,xt,V,histO,K);

rD=[2 0 0 0 0;-2 0 0 0 0]';
[xt,V]=SToMeanVar(s+rD(:,1));
rSamples(1)=rho0-Similarity(im0,xt,V,histO,K);
[xt,V]=SToMeanVar(s+rD(:,2));
rSamples(2)=rho0-Similarity(im0,xt,V,histO,K);
rSamples(find(rSamples<fClip))=fClip;%clip negative
R(1,1)=mean(sum(rD.^2,1)./rSamples);

rD=[0 2 0 0 0;0 -2 0 0 0]';
[xt,V]=SToMeanVar(s+rD(:,1));
rSamples(1)=rho0-Similarity(im0,xt,V,histO,K);
[xt,V]=SToMeanVar(s+rD(:,2));
rSamples(2)=rho0-Similarity(im0,xt,V,histO,K);
rSamples(find(rSamples<fClip))=fClip;%clip negative
R(2,2)=mean(sum(rD.^2,1)./rSamples);

%var
rD=0.1*[0 0 s(3) 0 0;0 0 -s(3) 0 0]';
[xt,V]=SToMeanVar(s+rD(:,1));
rSamples(1)=rho0-Similarity(im0,xt,V,histO,K);
[xt,V]=SToMeanVar(s+rD(:,2));
rSamples(2)=rho0-Similarity(im0,xt,V,histO,K);
rSamples(find(rSamples<fClip))=fClip;%clip negative
R(3,3)=mean(sum(rD.^2,1)./rSamples);

rD=0.1*[0 0 0 s(4) 0;0 0 0 -s(4) 0]';
[xt,V]=SToMeanVar(s+rD(:,1));
rSamples(1)=rho0-Similarity(im0,xt,V,histO,K);
[xt,V]=SToMeanVar(s+rD(:,2));
rSamples(2)=rho0-Similarity(im0,xt,V,histO,K);
rSamples(find(rSamples<fClip))=fClip;%clip negative
R(4,4)=mean(sum(rD.^2,1)./rSamples);

rD=0.1*[0 0 0 0 s(5);0 0 0 0 -s(5)]';
[xt,V]=SToMeanVar(s+rD(:,1));
rSamples(1)=rho0-Similarity(im0,xt,V,histO,K);
[xt,V]=SToMeanVar(s+rD(:,2));
rSamples(2)=rho0-Similarity(im0,xt,V,histO,K);
rSamples(find(rSamples<fClip))=fClip;%clip negative
R(5,5)=mean(sum(rD.^2,1)./rSamples);

pi=rho0;