Python图像连通域分析、图像分割

from skimage import io
import numpy as np
import matplotlib.pyplot as plt
from skimage import measure, color

if __name__ == '__main__':
    gt = io.imread('85484.png')

    gt = 1* (gt > 0)
    plt.figure(figsize=(10, 10))
    plt.imshow(gt)
    plt.show()



    # 【1】计算细胞和背景的像素频率
    c_weights = np.zeros(2)
    c_weights[0] = 1.0 / ((gt == 0).sum())
    c_weights[1] = 1.0 / ((gt == 1).sum())

    # 【2】归一化
    c_weights /= c_weights.max()

    # 【3】得到 class_weight map(cw_map)
    cw_map = np.where(gt == 0, c_weights[0], c_weights[1])
    plt.figure(figsize=(10, 10))
    plt.imshow(cw_map)
    plt.show()



    # 【4】连通域分析，并彩色化
    cells = measure.label(gt, connectivity=2)
    cells_color = color.label2rgb(cells, bg_label=0, bg_color=(255, 255, 255))
    plt.figure(figsize=(10, 10))
    plt.imshow(cells_color)
    # plt.imshow(cells)
    plt.show()

    import cv2

    # 【5】计算得到 distance weight map (dw_map)
    w0 = 10
    sigma = 5
    dw_map = np.zeros_like(gt)
    maps = np.zeros((gt.shape[0], gt.shape[1], cells.max()))
    if cells.max() >= 2:
        for i in range(1, cells.max() + 1):
            maps[:, :, i - 1] = cv2.distanceTransform(1 - (cells == i).astype(np.uint8), cv2.DIST_L2, 3)
        maps = np.sort(maps, axis=2)
        d1 = maps[:, :, 0]
        d2 = maps[:, :, 1]
        dis = ((d1 + d2) ** 2) / (2 * sigma * sigma)
        dw_map = w0 * np.exp(-dis) * (cells == 0)

    # dw_map=dw_map+cells
    plt.figure(figsize=(10, 10))
    plt.imshow(dw_map, cmap='jet')