集中质量算法matlab

集中质量算法（Centroid Method）是一种常用的图像分割算法，其基本思路是将一幅图像分成若干个区域，使得每个区域内部的像素灰度值相似，而不同区域之间的像素灰度值差异较大。下面是一份基于Matlab的集中质量算法代码实现，供参考：

function [segs, num_segs] = Centroid(img, T)
% img: input grayscale image
% T: threshold for segmentation
% segs: output segmented image
% num_segs: number of segments

[m, n] = size(img);
segs = zeros(m, n); % initialize segmented image
label = 0; % initialize label

for i = 1:m
    for j = 1:n
        if segs(i, j) == 0 % if pixel not labeled
            label = label + 1; % create new label
            segs(i, j) = label; % label current pixel
            neighbor_label = []; % initialize neighbor labels
            neighbor_val = []; % initialize neighbor values
            neighbor_count = 0; % initialize neighbor count
            c = [i, j]; % centroid position
            
            % find neighboring pixels and their values
            if i > 1 % north
                neighbor_count = neighbor_count + 1;
                neighbor_label(neighbor_count) = segs(i-1, j);
                neighbor_val(neighbor_count) = img(i-1, j);
            end
            if j > 1 % west
                neighbor_count = neighbor_count + 1;
                neighbor_label(neighbor_count) = segs(i, j-1);
                neighbor_val(neighbor_count) = img(i, j-1);
            end
            if i < m % south
                neighbor_count = neighbor_count + 1;
                neighbor_label(neighbor_count) = segs(i+1, j);
                neighbor_val(neighbor_count) = img(i+1, j);
            end
            if j < n % east
                neighbor_count = neighbor_count + 1;
                neighbor_label(neighbor_count) = segs(i, j+1);
                neighbor_val(neighbor_count) = img(i, j+1);
            end
            
            % compute centroid of neighboring pixels
            if neighbor_count > 0
                centroid = [0, 0];
                for k = 1:neighbor_count
                    centroid = centroid + [neighbor_label(k)*neighbor_val(k), neighbor_label(k)];
                end
                centroid = centroid / sum(neighbor_label.*neighbor_val);
                c = round(centroid(2:-1:1)); % convert to (i,j) format
            end
            
            % propagate label to neighboring pixels
            for k = 1:neighbor_count
                if abs(neighbor_val(k) - img(c(1), c(2))) <= T % if values similar
                    segs(c(1), c(2)) = neighbor_label(k); % propagate label
                end
            end
        end
    end
end

num_segs = max(max(segs)); % compute number of segments

该代码实现了Centroid算法的基本流程，包括：

1. 对图像中每个未标记像素进行遍历；
2. 对当前像素的相邻像素进行获取，并计算其质心位置；
3. 判断相邻像素的灰度值是否与当前像素的灰度值相似，如果相似则将其标记为当前像素所在的区域；
4. 重复1~3步骤，直到所有像素均被标记。

该算法的核心在于质心的计算和标记的传播，其中质心的计算采用了加权平均的方式，可以有效地减少噪声的影响，标记的传播则通过判断相邻像素的灰度值是否相似来实现。