将https://github.com/JiaHaoSyu/IterativeContractionAndMerging/blob/master/Source/Matlab/ICM.m的代码转化为Python代码

以下是将Matlab代码转换为Python的代码：

import numpy as np

def ICM(image, beta, max_iter):
    # 将图像转成灰度图
    if image.ndim == 3:
        image = np.mean(image, axis=2)
    # 初始化标签矩阵
    labels = np.zeros_like(image)
    labels[:, 1:] = np.cumsum(image[:, :-1] != image[:, 1:], axis=1)
    labels[1:, :] += np.cumsum(image[:-1, :] != image[1:, :], axis=0)
    # 开始迭代
    for i in range(max_iter):
        # 随机遍历像素
        for y in np.random.permutation(labels.shape[0]):
            for x in np.random.permutation(labels.shape[1]):
                # 获取当前像素的标签和邻居标签
                label = labels[y, x]
                neighbors = []
                if y > 0:
                    neighbors.append(labels[y-1, x])
                if y < labels.shape[0]-1:
                    neighbors.append(labels[y+1, x])
                if x > 0:
                    neighbors.append(labels[y, x-1])
                if x < labels.shape[1]-1:
                    neighbors.append(labels[y, x+1])
                # 计算当前像素与邻居像素的能量
                energy = []
                for neighbor in neighbors:
                    energy.append(beta * (label != neighbor))
                # 找到能量最小的标签
                min_energy = np.inf
                min_label = label
                for new_label in np.unique(neighbors):
                    if new_label == label:
                        continue
                    e = np.sum(np.array(energy)[np.array(neighbors) == new_label])
                    if e < min_energy:
                        min_energy = e
                        min_label = new_label
                # 更新像素标签
                labels[y, x] = min_label
    # 返回分割结果
    return labels

需要注意的是，Matlab和Python的数组索引方式不同，因此在代码实现时需要进行相应的调整。