使用 Pandas 探索和可视化区域属性

此玩具示例展示了如何计算一系列 10 张图像中每个标记区域的大小。我们使用 2D 图像，然后是 3D 图像。类 blob 的区域是合成生成的。随着体积分数（即 blob 覆盖的像素或体素的比率）增加，blob（区域）的数量减少，单个区域的大小（面积或体积）可能会越来越大。面积（大小）值以与 Pandas 兼容的格式提供，这使得数据分析和可视化变得方便。

除了面积之外，还有许多其他区域属性可用。

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns

from skimage import data, measure


fractions = np.linspace(0.05, 0.5, 10)

2D 图像

images = [data.binary_blobs(volume_fraction=f) for f in fractions]

labeled_images = [measure.label(image) for image in images]

properties = ['label', 'area']

tables = [
    measure.regionprops_table(image, properties=properties) for image in labeled_images
]
tables = [pd.DataFrame(table) for table in tables]

for fraction, table in zip(fractions, tables):
    table['volume fraction'] = fraction

areas = pd.concat(tables, axis=0)

# Create custom grid of subplots
grid = plt.GridSpec(2, 2)
ax1 = plt.subplot(grid[0, 0])
ax2 = plt.subplot(grid[0, 1])
ax = plt.subplot(grid[1, :])
# Show image with lowest volume fraction
ax1.imshow(images[0], cmap='gray_r')
ax1.set_axis_off()
ax1.set_title(f'fraction {fractions[0]}')
# Show image with highest volume fraction
ax2.imshow(images[-1], cmap='gray_r')
ax2.set_axis_off()
ax2.set_title(f'fraction {fractions[-1]}')
# Plot area vs volume fraction
areas.plot(x='volume fraction', y='area', kind='scatter', ax=ax)
plt.show()

在散点图中，许多点似乎在较低的面积值处重叠。为了更好地了解分布情况，我们可能希望在可视化中添加一些“抖动”。为此，我们使用 seaborn.stripplot（来自 seaborn 库，用于统计数据可视化），参数为 jitter=True。

fig, ax = plt.subplots()
sns.stripplot(x='volume fraction', y='area', data=areas, jitter=True, ax=ax)
# Fix floating point rendering
ax.set_xticklabels([f'{frac:.2f}' for frac in fractions])
plt.show()

/home/runner/work/scikit-image/scikit-image/doc/examples/segmentation/plot_regionprops_table.py:75: UserWarning:

set_ticklabels() should only be used with a fixed number of ticks, i.e. after set_ticks() or using a FixedLocator.

3D 图像

在 3D 中进行相同的分析，我们发现了一个更显著的行为：当体积分数超过 ~0.25 时，blob 会合并成一个巨大的整体。这对应于统计物理学和图论中的 渗流阈值。

images = [data.binary_blobs(length=128, n_dim=3, volume_fraction=f) for f in fractions]

labeled_images = [measure.label(image) for image in images]

properties = ['label', 'area']

tables = [
    measure.regionprops_table(image, properties=properties) for image in labeled_images
]
tables = [pd.DataFrame(table) for table in tables]

for fraction, table in zip(fractions, tables):
    table['volume fraction'] = fraction

blob_volumes = pd.concat(tables, axis=0)

fig, ax = plt.subplots()
sns.stripplot(x='volume fraction', y='area', data=blob_volumes, jitter=True, ax=ax)
ax.set_ylabel('blob size (3D)')
# Fix floating point rendering
ax.set_xticklabels([f'{frac:.2f}' for frac in fractions])
plt.show()

/home/runner/work/scikit-image/scikit-image/doc/examples/segmentation/plot_regionprops_table.py:107: UserWarning:

set_ticklabels() should only be used with a fixed number of ticks, i.e. after set_ticks() or using a FixedLocator.