下面的代码在Python/OpenCV中的表格上绘制线条。

• 读取输入
• 转换为灰色
• 阈值（在220，但可能是图像相关的），以消除您的图像中的噪声。注意，您不会看到噪声，除非您在一个高阈值）
• 将阈值图像平均为一行
• 阈值为254
• 应用一个小的水平形态开放连接第5个文本列标题中的句号，这样就不会在那里画额外的线。
• 得到轮廓线、边界框和X轴中心。
• 在垂直线结果的输入副本和输入的第二个副本（最终结果）的X坐标处，将等值线绘制为垂直线
• 将阈值平均到一列
• 阈值为254
• 得到轮廓线、边界框和Y轴中心。
• 在水平线结果的输入副本和输入的前一个第二副本（最终结果）上的Y坐标处绘制等值线作为水平线
• 保存结果

输入：

import cv2
import numpy as np

# read the input
img = cv2.imread('table4b.png')
hh, ww = img.shape[:2]

# convert to grayscale 
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# threshold on white
thresh = cv2.threshold(gray, 220, 255, cv2.THRESH_BINARY)[1]

# average gray image to one row
row = cv2.resize(thresh, (ww, 1), interpolation = cv2.INTER_AREA)

# threshold on white
thresh1 = cv2.threshold(row, 254, 255, cv2.THRESH_BINARY)[1]

# apply small amount of morphology to merge period with column of text
kernel = cv2.getStructuringElement(cv2.MORPH_RECT , (5,1))
thresh1 = cv2.morphologyEx(thresh1, cv2.MORPH_OPEN, kernel)

# get contours
contours = cv2.findContours(thresh1, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]
result = img.copy()
result1 = img.copy()
for cntr in contours:
    x, y, w, h = cv2.boundingRect(cntr)
    xcenter = x + w // 2
    cv2.line(result1, (xcenter,0), (xcenter, hh-1), (0, 0, 0), 1)
    cv2.line(result, (xcenter,0), (xcenter, hh-1), (0, 0, 0), 1)

# average gray image to one column
column = cv2.resize(thresh, (1, hh), interpolation = cv2.INTER_AREA)

# threshold on white
thresh2 = cv2.threshold(column, 254, 255, cv2.THRESH_BINARY)[1]
# get contours

contours = cv2.findContours(thresh2, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours[0] if len(contours) == 2 else contours[1]
result2 = img.copy()
for cntr in contours:
    x, y, w, h = cv2.boundingRect(cntr)
    ycenter = y + h // 2
    cv2.line(result2, (0, ycenter), (ww-1, ycenter), (0, 0, 0), 1)
    cv2.line(result, (0, ycenter), (ww-1, ycenter), (0, 0, 0), 1)
    
# save results
cv2.imwrite('table4_threshold.png', thresh)
cv2.imwrite('table4_column_lines.png', result1)
cv2.imwrite('table4_row_lines.png', result2)
cv2.imwrite('table4_column_and_row_lines.png', result)
   
# show result
cv2.imshow('thresh', thresh)
cv2.imshow('result1', result1)
cv2.imshow('result2', result2)
cv2.imshow('result', result)
cv2.waitKey(0)

阈值图像：

行线图像：

列线图像：

行线和列线图像：

我试过了，但可能需要为输入图像调整一些阈值...
(The代码是C++，但我认为您可以从注解中了解它的作用。）

int main()
{
    //- Load Image as GrayScale
    cv::Mat Img = cv::imread( "Tbl.png", cv::IMREAD_GRAYSCALE );
    if( Img.empty() ){  std::cout << "imread() failed" << std::endl;    return 0;   }

    //- Count white pixels for each row
    std::vector<int> WhitePixCnts( Img.rows, 0 );
    {
        const unsigned char PixValThresh = 200; //Threshold 1
        for( int y=0; y<Img.rows; ++y )
        {
            const unsigned char *p = Img.ptr<unsigned char>( y );
            for( int x=0; x<Img.cols; ++x, ++p )
            {
                if( *p >= PixValThresh ){   ++WhitePixCnts[y];  }
            }
        }
    }

    //- Grouping consecutive white rows
    std::vector< std::pair<int,int> > WhiteBelts;
    {
        const int CntThresh = (int)( Img.cols * 0.95 ); //Threshold 2

        int BeltTopY = -1;
        for( int y=0; y<Img.rows; ++y )
        {
            if( WhitePixCnts[y] < CntThresh )
            {
                if( BeltTopY >= 0 )
                {
                    WhiteBelts.emplace_back( BeltTopY, y );
                    BeltTopY = -1;
                }
            }
            else if( BeltTopY<0 ){  BeltTopY = y;   }
        }
        if( BeltTopY >= 0 )WhiteBelts.emplace_back( BeltTopY, Img.rows );
    }

    //* (For simplification)
    //  In this sample, if the number of groups is insufficient here, give up.
    if( WhiteBelts.size() < 3 ){    std::cout << "give up" << std::endl;    return 0;   }

    //- Decide the threshold for belt thickness based on the found belt's thickness
    int ThicknessThresh = 0;
    {
        int MaxBeltThickness = 0;

        for( int i=1; i+1<WhiteBelts.size(); ++i ) //* Excluding top and bottom belt, because they will include margin.
        {
            const auto &Belt = WhiteBelts[i];
            MaxBeltThickness = std::max( MaxBeltThickness, Belt.second - Belt.first );
        }
        //Very simple decision...
        ThicknessThresh = (int)( MaxBeltThickness * 0.5 );  //Threshold 3
    }

    //- Accept top, bottom, and enoughly bold belts
    cv::Mat ShowImg;
    cv::cvtColor( Img, ShowImg, cv::COLOR_GRAY2BGR );
    
    for( int i=0; i<(int)WhiteBelts.size(); ++i )
    {
        const auto &Belt = WhiteBelts[i];
        if( i==0  ||  (i+1)==(int)WhiteBelts.size()  ||  ( Belt.second - Belt.first >= ThicknessThresh ) )
        {
            int BeltCenterY = ( Belt.first + Belt.second ) / 2;
            cv::line( ShowImg, cv::Point(0,BeltCenterY), cv::Point(ShowImg.cols,BeltCenterY), cv::Scalar(0,0,255) );
        }
    }
    cv::imshow( "Result", ShowImg );
    cv::waitKey();
    return 0;
}

这是结果：

并且该结果是从90度旋转的输入图像获得的：

(The通过简单地组合这些，可以获得完整的结果。）