对图像利用色调进行高斯混合聚类的代码，不使用封装好的fitgmdist函数

以下是对图像利用色调进行高斯混合聚类的代码，不使用封装好的fitgmdist函数：

% 读入图像
img = imread('test.jpg');
% 转换为Lab色彩空间
lab_img = rgb2lab(img);
% 取出a,b两个通道的值
ab = double(lab_img(:, :, 2:3));
% 将图像转换为N行2列的矩阵
nrows = size(ab, 1);
ncols = size(ab, 2);
ab = reshape(ab, nrows*ncols, 2);
% 设定高斯混合模型的参数
K = 3; % 聚类数目
max_iters = 10; % 最大迭代次数
min_diff = 1e-6; % 停止迭代的最小误差
% 初始化高斯混合模型参数
mu = zeros(K, 2);
sigma = zeros(2, 2, K);
p = zeros(1, K);
% 随机初始化模型参数
for k = 1:K
    mu(k, :) = ab(randi(nrows*ncols), :);
    sigma(:, :, k) = eye(2);
    p(k) = 1/K;
end
% 迭代优化模型参数
for iter = 1:max_iters
    % E 步：计算后验概率
    posterior = zeros(nrows*ncols, K);
    for k = 1:K
        posterior(:, k) = p(k) * mvnpdf(ab, mu(k, :), sigma(:, :, k));
    end
    posterior = bsxfun(@rdivide, posterior, sum(posterior, 2));
    % M 步：更新模型参数
    prev_mu = mu;
    for k = 1:K
        Nk = sum(posterior(:, k));
        mu(k, :) = sum(bsxfun(@times, posterior(:, k), ab)) / Nk;
        sigma(:, :, k) = bsxfun(@times, posterior(:, k), ab - mu(k, :))' * (ab - mu(k, :)) / Nk;
        p(k) = Nk / (nrows*ncols);
    end
    % 判断是否收敛
    diff = sum(sum(abs(mu - prev_mu)));
    if diff < min_diff
        break;
    end
end
% 根据聚类结果对图像进行着色
pixel_labels = reshape(posterior, nrows, ncols, K);
[~, idx] = max(posterior, [], 2);
segmented_images = zeros(nrows, ncols, 3, 'uint8');
rgb_label = repmat(idx, [1 3]);
for k = 1:K
    color = img;
    color(rgb_label ~= k) = 0;
    segmented_images(:, :, :, k) = color;
end
% 显示结果
figure;
subplot(2, 2, 1);
imshow(img);
title('Original Image');
subplot(2, 2, 2);
imshow(segmented_images(:,:,:,1));
title('Cluster 1');
subplot(2, 2, 3);
imshow(segmented_images(:,:,:,2));
title('Cluster 2');
subplot(2, 2, 4);
imshow(segmented_images(:,:,:,3));
title('Cluster 3');