图像边缘处find_contours的轮廓闭合问题

针对目标物体超出图像边界时，measure.find_contours无法自动闭合开放轮廓的问题，提供以下几种实用解决思路：

1. 手动检测并闭合边界轮廓 #

遍历每个提取到的轮廓，判断其起点和终点是否位于图像边界，若均在边界上，直接将轮廓首尾相连完成闭合：

import numpy as np
import matplotlib.pyplot as plt
from skimage import measure

r = masks[1]['segmentation']
h, w = r.shape
contours = measure.find_contours(r, 0.8)

# 处理开放轮廓
closed_contours = []
for contour in contours:
    start = contour[0]
    end = contour[-1]
    # 判断点是否在图像边界（上下左右边缘）
    def on_edge(point):
        y, x = point
        return y == 0 or y == h-1 or x == 0 or x == w-1
    
    if on_edge(start) and on_edge(end):
        # 首尾相连闭合轮廓
        closed_contour = np.vstack([contour, start])
        closed_contours.append(closed_contour)
    else:
        closed_contours.append(contour)

# 可视化结果
fig, ax = plt.subplots()
ax.imshow(r, cmap=plt.cm.gray)
for contour in closed_contours:
    ax.plot(contour[:, 1], contour[:, 0], linewidth=2)
ax.axis('image')
ax.set_xticks([])
ax.set_yticks([])
plt.show()

2. 扩展图像边界后提取轮廓 #

在原图像周围添加一圈背景像素，让跨边界的物体被完整包裹，此时find_contours会生成闭合轮廓，最后修正坐标对应原图像：

import numpy as np
import matplotlib.pyplot as plt
from skimage import measure

r = masks[1]['segmentation']
# 上下左右各扩展1像素背景（0值）
extended_r = np.pad(r, pad_width=1, mode='constant', constant_values=0)
contours = measure.find_contours(extended_r, 0.8)
# 修正轮廓坐标，减去扩展的1像素偏移
corrected_contours = [cnt - 1 for cnt in contours]

# 可视化结果
fig, ax = plt.subplots()
ax.imshow(r, cmap=plt.cm.gray)
for contour in corrected_contours:
    ax.plot(contour[:, 1], contour[:, 0], linewidth=2)
ax.axis('image')
ax.set_xticks([])
ax.set_yticks([])
plt.show()

3. 结合regionprops与凸包生成闭合轮廓 #

利用regionprops获取区域信息，通过凸包算法补全跨边界部分，生成闭合轮廓：

import numpy as np
import matplotlib.pyplot as plt
from skimage import measure, label
from skimage.morphology import convex_hull_image

r = masks[1]['segmentation']
# 标记图像中的区域
label_img = label(r)
region_props = measure.regionprops(label_img)

fig, ax = plt.subplots()
ax.imshow(r, cmap=plt.cm.gray)

for prop in region_props:
    # 生成区域的凸包
    hull = convex_hull_image(prop.image)
    # 提取凸包轮廓并转换到原图像坐标
    hull_contours = measure.find_contours(hull, 0.5)
    for hull_cnt in hull_contours:
        # 加上区域的bbox偏移量，还原到原图像位置
        contour = hull_cnt + prop.bbox[:2]
        ax.plot(contour[:, 1], contour[:, 0], linewidth=2)

ax.axis('image')
ax.set_xticks([])
ax.set_yticks([])
plt.show()

内容的提问来源于stack exchange，提问作者ewittry