slic算法的matlab代码

以下是MATLAB中实现SLIC算法的示例代码：

function [L, Am] = slic(I, k, m)

% I: 输入图像
% k: 超像素数目
% m: 超像素紧凑度参数

% 将RGB图像转换为LAB颜色空间
lab_I = RGB2Lab(I);

% 初始化超像素大小
[N, M, ~] = size(lab_I);
S = round(sqrt(N*M/k));

% 计算超像素中心
S_half = floor(S/2);
x_grid = S_half:S:(M-S_half);
y_grid = S_half:S:(N-S_half);
[LX, LY] = meshgrid(x_grid, y_grid);
C = [LY(:), LX(:)]; % 超像素中心

% 初始化超像素标签和距离
L = zeros(N, M);
D = inf(N, M);

% 迭代优化
for it = 1:10
    
    % 计算每个超像素的平均颜色和位置
    for i = 1:size(C,1)
        y_min = max(C(i,1)-S_half, 1);
        y_max = min(C(i,1)+S_half, N);
        x_min = max(C(i,2)-S_half, 1);
        x_max = min(C(i,2)+S_half, M);
        patch = lab_I(y_min:y_max, x_min:x_max, :);
        patch = reshape(patch, [], 3);
        C_lab(i,:) = mean(patch, 1);
        C_pos(i,:) = mean([x_min, y_min; x_max, y_max], 1);
    end
    
    % 计算每个像素到所有超像素中心的距离
    for i = 1:size(C,1)
        y_min = max(C(i,1)-2*S, 1);
        y_max = min(C(i,1)+2*S, N);
        x_min = max(C(i,2)-2*S, 1);
        x_max = min(C(i,2)+2*S, M);
        patch = lab_I(y_min:y_max, x_min:x_max, :);
        patch = reshape(patch, [], 3);
        D_patch = sum((patch-C_lab(i,:)).^2, 2) + m^2*((C_pos(i,1)-x_grid).^2 + (C_pos(i,2)-y_grid).^2);
        idx = sub2ind([N, M], y_min:y_max, x_min:x_max);
        D(idx) = D_patch;
        L(idx) = i;
    end
    
    % 更新每个超像素的中心位置
    for i = 1:size(C,1)
        idx = L(:) == i;
        C(i,:) = mean(lab_I(idx, :), 1);
    end
end

% 构造相邻矩阵
idx = reshape(1:(N*M), N, M);
adj = sparse([], [], [], k, k, 2*k);
for y = 2:N-1
    for x = 2:M-1
        c1 = L(y, x);
        c2 = L(y, x+1);
        if c1 ~= c2
            adj(c1, c2) = min(adj(c1, c2), D(y, x+1));
            adj(c2, c1) = adj(c1, c2);
        end
        c2 = L(y+1, x);
        if c1 ~= c2
            adj(c1, c2) = min(adj(c1, c2), D(y+1, x));
            adj(c2, c1) = adj(c1, c2);
        end
    end
end

% 计算超像素相似性矩阵
Am = full(adj);
Am(Am==0) = inf;
Am = exp(-Am/max(Am(:)));

% 显示超像素分割结果
figure;
imshow(I);
hold on;
boundary = L~=imfilter(L, [0 1 0; 1 0 1; 0 1 0]/4);
plot(find(boundary), 'r.');
end

function lab_I = RGB2Lab(I)

% RGB2Lab颜色空间转换
cform = makecform('srgb2lab');
lab_I = applycform(I, cform);

% 对L通道进行归一化
L = lab_I(:, :, 1)/100;
L(L>0.008856) = ((L(L>0.008856)+0.0992)/1.099).^0.45;
L(L<=0.008856) = L(L<=0.008856)/0.9033;
lab_I(:, :, 1) = L*100;
end

其中，输入参数`I`为待分割的RGB图像，`k`为超像素数目，`m`为超像素紧凑度参数。函数输出`L`为超像素标签，`Am`为超像素相似性矩阵。函数内部实现了迭代优化和相邻矩阵的构造，最终显示了超像素分割结果。