二分-k均值聚类算法进行图像分割并输出轮廓系数的完整代码

以下是使用Python实现二分-k均值聚类算法进行图像分割并输出轮廓系数的完整代码：

import numpy as np
from PIL import Image
from sklearn.metrics import silhouette_score

def load_image(filename):
    img = Image.open(filename)
    return np.array(img)

def save_image(filename, data):
    img = Image.fromarray(np.uint8(data))
    img.save(filename)

def kmeans(data, k, max_iter=100):
    centers = np.random.randint(256, size=(k,))
    for i in range(max_iter):
        clusters = [[] for _ in range(k)]
        for x in data:
            distances = np.abs(centers - x)
            idx = np.argmin(distances)
            clusters[idx].append(x)
        new_centers = [np.mean(cluster) for cluster in clusters]
        if np.allclose(new_centers, centers):
            break
        centers = new_centers
    return centers, clusters

def bisecting_kmeans(data, k, max_iter=100):
    clusters = [data]
    while len(clusters) < k:
        max_silhouette = -1
        for i, cluster in enumerate(clusters):
            centers, subclusters = kmeans(cluster, 2, max_iter)
            silhouettes = np.zeros(len(subclusters))
            for j, subcluster in enumerate(subclusters):
                if len(subcluster) > 1:
                    silhouette = silhouette_score(subcluster, np.zeros_like(subcluster))
                    silhouettes[j] = silhouette
            silhouette = np.mean(silhouettes)
            if silhouette > max_silhouette:
                max_silhouette = silhouette
                max_idx = i
                max_centers = centers
                max_subclusters = subclusters
        clusters.pop(max_idx)
        clusters.extend(max_subclusters)
    return clusters

if __name__ == '__main__':
    img = load_image('input.jpg')
    data = img.reshape(-1, 3)
    clusters = bisecting_kmeans(data, 4)
    for i, cluster in enumerate(clusters):
        centers, _ = kmeans(cluster, 1)
        distances = np.abs(cluster - centers[0])
        mask = np.argmin(distances, axis=1) == 0
        data[mask] = centers[0]
    img = data.reshape(img.shape)
    save_image('output.jpg', img)
    silhouettes = np.zeros(len(clusters))
    for i, cluster in enumerate(clusters):
        if len(cluster) > 1:
            silhouette = silhouette_score(cluster, np.zeros_like(cluster))
            silhouettes[i] = silhouette
    print('Silhouette coefficients:', silhouettes)

这个代码基于之前的二分-k均值聚类算法实现了轮廓系数的计算。首先，使用 bisecting_kmeans 函数将数据分成四个簇。然后，使用 kmeans 函数将每个簇压缩为一个点，并使用该点将簇中所有像素点赋值为该点的颜色。接下来，将数据数组转换回图像并保存输出图像。最后，使用 silhouette_score 函数计算每个簇的轮廓系数，并输出到控制台。

注意，这里使用了 scikit-learn 库中的 silhouette_score 函数来计算轮廓系数，如果没有安装该库，需要先安装。此外，轮廓系数的计算需要注意簇中至少有两个样本。