python - 如何使用 OpenCV python 自动检测修剪和裁剪图像的一部分？

Question

我正在研究图像到文本的转换程序，我想删除周围不必要的图像、背景和文本，以便它强调我希望从中提取文本的主要区域！

请参阅药盒上的警告标签。如何仅提取图像的预期区域！？因为所有其他元素都会引起噪声并影响提取的准确性。

标签： pythonimageopencvimage-processing

Answer 1

我的建议是找到具有最大子轮廓数的轮廓。 具有最大子节点数的轮廓是其中包含最多字母的轮廓。

这只是粗过滤的第一阶段，还有更多阶段：

• 转换为灰度，并转换为二值图像。
• 查找轮廓和层次结构。
  用于RETR_TREE在轮廓内创建轮廓树。
• 找到具有最大子轮廓数的父轮廓。
• 从图像中裁剪上述轮廓的边界矩形。
• 查找连接的组件（集群）。
• 删除大、小、高和宽的簇——当然不是字母。
• 使用“关闭”形态学操作来合并文本区域。
• 找到轮廓，并裁剪轮廓的最大面积边界矩形。

这个问题具有挑战性，我的回答可能与您发布的图像过拟合。
您可以改进启发式方法以获得更通用的解决方案。

---

这是代码：

import numpy as np
import cv2

# Read input image
img = cv2.imread('dizziness.jpg')

# Convert to Grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Convert to binary image - use THRESH_OTSU for automatic threshold.
_, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

# Find contours and hierarchy, use RETR_TREE for creating a tree of contours within contours
cnts, hiers = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)[-2:]  # [-2:] indexing takes return value before last (due to OpenCV compatibility issues).

# https://docs.opencv.org/master/d9/d8b/tutorial_py_contours_hierarchy.html
# Hierarchy Representation in OpenCV
# So each contour has its own information regarding what hierarchy it is, who is its child, who is its parent etc. 
# OpenCV represents it as an array of four values : [Next, Previous, First_Child, Parent]
parent = hiers[0, :, 3]

# Find parent contour with the maximum number of child contours
# Use np.bincount for counting the number of instances of each parent value.
# https://docs.scipy.org/doc/numpy/reference/generated/numpy.bincount.html#numpy.bincount
hist = np.bincount(np.maximum(parent, 0))
max_n_childs_idx = hist.argmax()

# Get the contour with the maximum child contours
c = cnts[max_n_childs_idx]

# Get bounding rectangle
x, y, w, h = cv2.boundingRect(c)

# Crop the bounding rectangle out of img
img = img[y:y+h, x:x+w, :]
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# Convert to binary image (after cropping) and invert polarity
_, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)

#cv2.imshow('thresh', thresh);cv2.waitKey(0);cv2.destroyAllWindows()

# Find connected components (clusters)
nlabel,labels,stats,centroids = cv2.connectedComponentsWithStats(thresh, connectivity=8)

# Delete large, small, tall and wide clusters - not letters for sure
max_area = 2000
min_area = 10
max_width = 100
max_height = 100
for i in range(1, nlabel):
    if (stats[i, cv2.CC_STAT_AREA] > max_area) or \
       (stats[i, cv2.CC_STAT_AREA] < min_area) or \
       (stats[i, cv2.CC_STAT_WIDTH] > max_width) or \
       (stats[i, cv2.CC_STAT_HEIGHT] > max_height):
        thresh[labels == i] = 0

#cv2.imshow('thresh', thresh);cv2.waitKey(0);cv2.destroyAllWindows()

# Use "closing" morphological operation for uniting text area
thresh = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, np.ones((50, 50)))

# Find contours once more
cnts = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)[-2]

# Get contour with maximum area
c = max(cnts, key=cv2.contourArea)

# Get bounding rectangle
x, y, w, h = cv2.boundingRect(c)

# Crop the bounding rectangle out of img (leave some margins)
out = img[y-5:y+h+5, x-5:x+w+5]

# Show output
cv2.imshow('out', out);
cv2.waitKey(0);
cv2.destroyAllWindows()

---

结果：