将网页https://github.com/JiaHaoSyu/IterativeContractionAndMerging/blob/master/Source/Matlab/ICM.m的ICM的matlab代码转化为Python代码以实现相同功能

以下是将该Matlab代码转换为Python代码的示例：

import numpy as np

def ICM(im, weights):
    # 设置参数
    num_neighbors = 4
    beta = 0.5
    max_iter = 10
    
    # 初始化图像分割
    h, w = im.shape
    labels = np.zeros((h, w), dtype=int)
    num_labels = 0
    label_hist = np.zeros((h*w), dtype=int)
    
    # 计算每个像素的权重和邻域信息
    weight_im = np.zeros((h, w, num_neighbors))
    for i in range(num_neighbors):
        if i == 0:
            weight_im[:, :, i] = weights[0] * np.exp(-beta * np.abs(im - np.roll(im, 1, axis=0)))
        elif i == 1:
            weight_im[:, :, i] = weights[1] * np.exp(-beta * np.abs(im - np.roll(im, 1, axis=1)))
        elif i == 2:
            weight_im[:, :, i] = weights[2] * np.exp(-beta * np.abs(im - np.roll(np.roll(im, 1, axis=0), 1, axis=1)))
        else:
            weight_im[:, :, i] = weights[3] * np.exp(-beta * np.abs(im - np.roll(np.roll(im, -1, axis=0), 1, axis=1)))
    
    # 迭代分割
    for iter in range(max_iter):
        for i in range(h):
            for j in range(w):
                # 计算像素i,j与邻域像素的不同标签数
                neighbor_labels = np.zeros(num_neighbors, dtype=int)
                for k in range(num_neighbors):
                    if k == 0:
                        neighbor_labels[k] = labels[i-1, j] if i > 0 else -1
                    elif k == 1:
                        neighbor_labels[k] = labels[i, j-1] if j > 0 else -1
                    elif k == 2:
                        neighbor_labels[k] = labels[i-1, j-1] if i > 0 and j > 0 else -1
                    else:
                        neighbor_labels[k] = labels[i+1, j-1] if i < h-1 and j > 0 else -1
                num_diff_labels = len(np.unique(neighbor_labels[neighbor_labels >= 0]))
                
                # 计算每个标签的能量
                energy = np.zeros(num_labels+1)
                for k in range(num_labels+1):
                    if k == num_labels:
                        # 新建标签
                        energy[k] = weight_im[i, j, :].sum()
                    else:
                        # 更新已有标签
                        pixels_in_label = (labels == k)
                        weight_sum = (weight_im[i, j, :] * pixels_in_label).sum()
                        energy[k] = weight_sum - beta * num_diff_labels
                
                # 分配标签
                new_label = energy.argmin()
                labels[i, j] = new_label
                if new_label == num_labels:
                    num_labels += 1
            
        # 更新标签直方图
        label_hist.fill(0)
        for i in range(h):
            for j in range(w):
                label_hist[labels[i, j]] += 1
        
        # 压缩标签
        labels_old = labels.copy()
        for i in range(h):
            for j in range(w):
                if label_hist[labels[i, j]] < 10:
                    neighbor_labels = np.zeros(num_neighbors, dtype=int)
                    for k in range(num_neighbors):
                        if k == 0:
                            neighbor_labels[k] = labels[i-1, j] if i > 0 else -1
                        elif k == 1:
                            neighbor_labels[k] = labels[i, j-1] if j > 0 else -1
                        elif k == 2:
                            neighbor_labels[k] = labels[i-1, j-1] if i > 0 and j > 0 else -1
                        else:
                            neighbor_labels[k] = labels[i+1, j-1] if i < h-1 and j > 0 else -1
                    neighbor_labels = neighbor_labels[neighbor_labels >= 0]
                    label_counts = np.zeros(num_labels, dtype=int)
                    for k in range(len(neighbor_labels)):
                        label_counts[neighbor_labels[k]] += 1
                    new_label = label_counts.argmax()
                    labels[i, j] = new_label
        
        # 如果标签没有发生变化，则提前结束迭代
        if (labels_old == labels).all():
            break
    
    return labels

请注意，这只是一个示例实现，可能需要根据具体情况进行调整。