如何从含表格的图像中提取线条，生成仅含表格线的黑白图像

问题背景 #

我有一张包含表格的图像：

我使用如下代码提取表格横竖线条，生成仅含表格线的图像：

import cv2 as cv
import numpy as np

img = cv.imread('1.jpg')
cv.imshow('Original', img)

gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
gray = cv.bitwise_not(gray)
bw = cv.adaptiveThreshold(gray, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 15, -2)

# 创建用于提取竖线的图像
vertical = np.copy(bw)
horizontal = np.copy(bw)

# 指定水平轴尺寸
cols = horizontal.shape[1]
horizontal_size = cols / 10
horizontal_size=int(horizontal_size)

# 创建形态学操作的结构元素，用于提取水平线
horizontalStructure = cv.getStructuringElement(cv.MORPH_RECT, (horizontal_size, 1))

horizontal = cv.erode(horizontal, horizontalStructure)
horizontal = cv.dilate(horizontal, horizontalStructure) 
# horizontal = cv.dilate(horizontal, (3, 3), iterations=1)

# 指定垂直轴尺寸
rows = vertical.shape[0]
verticalsize = rows / 10
verticalsize = int(verticalsize)

# 创建形态学操作的结构元素，用于提取竖直线
verticalStructure = cv.getStructuringElement(cv.MORPH_RECT, (1, verticalsize))

# 执行形态学操作
vertical = cv.erode(vertical, verticalStructure)
vertical = cv.dilate(vertical, verticalStructure)
# vertical = cv.dilate(vertical, (3, 3), iterations=1)

ret, thresh_v = cv.threshold(vertical, 127, 255, cv.THRESH_BINARY)
# cv.imshow('thresh vertical', thresh_v)

ret, thresh_h = cv.threshold(horizontal, 127, 255, cv.THRESH_BINARY)
# cv.imshow('thresh horizontal', thresh_h)

table_lines = cv.bitwise_or(thresh_h, thresh_v)
cv.imshow('table lines', table_lines)

cv.waitKey(0)

输出结果如下：

优化处理方案 #

当前输出杂乱是因为没有做噪点过滤、形态学核尺寸不合理导致文字笔画被误识别为表格线，按以下步骤调整即可得到和原图一致的完整表格线：

• 先加噪声过滤步骤，避免文字边缘噪点被误判为线条，在自适应阈值处理前增加高斯模糊操作

  gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
  gray = cv.bitwise_not(gray)
  # 新增高斯模糊
  gray = cv.GaussianBlur(gray, (3,3), 0)
  bw = cv.adaptiveThreshold(gray, 255, cv.ADAPTIVE_THRESH_MEAN_C, cv.THRESH_BINARY, 15, -2)
  

• 调整形态学核尺寸规则，原代码用图像宽高的1/10作为核长度，容易漏检短线条或者误判文字笔画，改为适配常规表格单元格尺寸的固定值即可

  # 替换原水平核长度计算逻辑，适配多数表格的单元格宽度
  horizontal_size = 25
  # 替换原垂直核长度计算逻辑
  verticalsize = 15
  

• 增加小噪点过滤步骤，合并横竖线后过滤掉面积过小的非表格线噪点

  table_lines = cv.bitwise_or(thresh_h, thresh_v)
  # 新增噪点过滤逻辑
  contours, _ = cv.findContours(table_lines, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
  for cnt in contours:
      if cv.contourArea(cnt) < 20:
          cv.drawContours(table_lines, [cnt], -1, 0, -1)
  

• 如有表格线断裂的情况，可增加一次闭运算补全缝隙

  kernel = cv.getStructuringElement(cv.MORPH_RECT, (2,2))
  table_lines = cv.morphologyEx(table_lines, cv.MORPH_CLOSE, kernel, iterations=1)
  

内容的提问来源于stack exchange，提问作者Meet Gondaliya