GMM.zip_GMM_Gaussian Mixture_image segmentation_mixture_高斯混合模型

function [iter, mask, P1, P2, P3] = EMGMM(I, k, m, n) ma = min(max(I)); % 取最大值 P1 = (1:k)*200/(k+1); % 初始化平均值数组 P2 = ones(1,k)*100; % 初始化方差数组 p3 = ones(m,n)/k; P3 = cell(1,k); % 初始化先验系数 k维矩阵 for i = 1:k P3{i} = p3; end iter = 0; % 迭代次数 % LOG = []; % 似然函数值 % loglik_p = -inf; % 初始似然函数值 % EM算法更新参数 while(1) iter = iter + 1; % E步 B = gaussian(I, P1, P2, k); % 高斯分布 for i = 1:k C{i} = cell2mat(P3(i)).*cell2mat(B(i)); end D = zeros(m,n); for i = 1:k D = D + cell2mat(C(i)); end for i = 1:k % 后验概率Z (t) Z{i} = cell2mat(C(i))./(D+eps); end % M步 for j = 1:k % 更新均值 E1 = cell2mat(Z(j)).*I; E2 = sum(sum(E1)); E3 = sum(sum(cell2mat(Z(j)))); P1(j) = E2/(E3+eps); end for j = 1:k % 更新方差 F1 = (I - P1(j)).*(I - P1(j)); F2 = cell2mat(Z(j)).*F1; F3 = sum(sum(F2)); F4 = sum(sum(cell2mat(Z(j)))); P2(j) = F3/(F4+eps); end G = zeros(m,n); for j = 1:k G = G + cell2mat(Z(j)); end for j = 1:k H1 = cell2mat(Z(j)); H2 = G; P3{j} = H1./(H2+eps); end % P1 = P11; % P2 = P22; % P3 = P33; if iter == 20 break; end end BB = gaussian(I, P1, P2, k); for i = 1:k CC{i} = cell2mat(P3(i)).*cell2mat(BB(i)); end DD = zeros(m,n); for i = 1:k DD = DD + cell2mat(CC(i)); end for i = 1:k ZZ{i} = cell2mat(CC(i))./(DD+eps); end % 计算每个点的归类 for i = 1:k ZZZ(:,:,i) = cell2mat(ZZ(i)); end mask = zeros(m,n); % 生成I大小的mask for i = 1:m for j = 1:n for t = 1:k M(t)= ZZZ(i,j,t); % 生成(i,j)位置的数组 end MA = find(M == max(M)); % 找到数组最大值的索引 mask(i,j) = MA(1); end end