OpenCV 实现图片的水平投影与垂直投影，并进行行分割

前言：对于印刷体图片来说，进行水平投影和垂直投影可以很快的进行分割，本文在OpenCV中如何进行水平投影和垂直投影通过代码进行说明。

水平投影：二维图像在y轴上的投影

垂直投影：二维图像在x轴上的投影

由于投影的图像需要进行二值化，本文采用积分二值化的方式，对图片进行处理。

---

具体代码如下：

//积分二值化
void thresholdIntegral (Mat inputMat, Mat& outputMat)
{

    int nRows = inputMat.rows;
    int nCols = inputMat.cols;

    // create the integral image
    Mat sumMat;
    integral (inputMat, sumMat);

    int S = MAX (nRows, nCols) / 8;
    double T = 0.15;

    // perform thresholding
    int s2 = S / 2;
    int x1, y1, x2, y2, count, sum;

    int* p_y1, *p_y2;
    uchar* p_inputMat, *p_outputMat;

    for (int i = 0; i < nRows; ++i)
    {
        y1 = i - s2;
        y2 = i + s2;

        if (y1 < 0)
        {
            y1 = 0;
        }
        if (y2 >= nRows)
        {
            y2 = nRows - 1;
        }

        p_y1 = sumMat.ptr<int> (y1);
        p_y2 = sumMat.ptr<int> (y2);
        p_inputMat = inputMat.ptr<uchar> (i);
        p_outputMat = outputMat.ptr<uchar> (i);

        for (int j = 0; j < nCols; ++j)
        {
            // set the SxS region
            x1 = j - s2;
            x2 = j + s2;

            if (x1 < 0)
            {
                x1 = 0;
            }
            if (x2 >= nCols)
            {
                x2 = nCols - 1;
            }

            count = (x2 - x1)* (y2 - y1);

            // I(x,y)=s(x2,y2)-s(x1,y2)-s(x2,y1)+s(x1,x1)
            sum = p_y2[x2] - p_y1[x2] - p_y2[x1] + p_y1[x1];

            if ((int) (p_inputMat[j] * count) < (int) (sum* (1.0 - T)))
            {
                p_outputMat[j] = 0;
            }
            else
            {
                p_outputMat[j] = 255;
            }
        }
    }
}
//垂直方向投影
void picshadowx (Mat binary)
{
    Mat paintx (binary.size(), CV_8UC1, Scalar (255)); //创建一个全白图片，用作显示

    int* blackcout = new int[binary.cols];
    memset (blackcout, 0, binary.cols * 4);

    for (int i = 0; i < binary.rows; i++)
    {
        for (int j = 0; j < binary.cols; j++)
        {
            if (binary.at<uchar> (i, j) == 0)
            {
                blackcout[j]++; //垂直投影按列在x轴进行投影
            }
        }
    }
    for (int i = 0; i < binary.cols; i++)
    {
        for (int j = 0; j < blackcout[i]; j++)
        {
            paintx.at<uchar> (binary.rows-1-j, i) = 0; //翻转到下面，便于观看
        }
    }
    delete blackcout;
    imshow ("paintx", paintx);

}
//水平方向投影并行分割
void picshadowy (Mat binary)
{  
   //是否为白色或者黑色根据二值图像的处理得来
    Mat painty (binary.size(), CV_8UC1, Scalar (255)); //初始化为全白
   
   //水平投影
   int* pointcount = new int[binary.rows]; //在二值图片中记录行中特征点的个数
    memset (pointcount, 0, binary.rows * 4);//注意这里需要进行初始化

    for (int i = 0; i < binary.rows; i++)
    {
        for (int j = 0; j < binary.cols; j++)
        {
            if (binary.at<uchar> (i, j) == 0)
            {
                pointcount[i]++; //记录每行中黑色点的个数 //水平投影按行在y轴上的投影
            }
        }
    }

    for (int i = 0; i < binary.rows; i++)
    {
        for (int j = 0; j < pointcount[i]; j++) //根据每行中黑色点的个数，进行循环
        {
            
            painty.at<uchar> (i, j) = 0; 
        }

    }

    imshow ("painty", painty);

    vector<Mat> result;
    int startindex = 0;
    int endindex = 0;
    bool inblock = false; //是否遍历到字符位置

    for (int i = 0; i < painty.rows; i++)
    {
      
        if (!inblock&&pointcount[i] != 0) //进入有字符区域
        {
            inblock = true;
            startindex = i;
            cout << "startindex:" << startindex << endl;
        }
        if (inblock&&pointcount[i] == 0) //进入空白区
        {
            endindex = i;
            inblock = false;
            Mat roi = binary.rowRange (startindex, endindex+1); //从而记录从开始到结束行的位置，即可进行行切分
            result.push_back (roi);
        }
    }

    for (int i = 0; i < result.size(); i++)
    {
        Mat tmp = result[i];
        imshow ("test"+to_string (i), tmp);
     }
    delete pointcount;

}
int main (int argc, char* argv[])
{
   
      Mat src = cv::imread ("test.jpg");

        if (src.empty())
        {
            cerr << "Problem loading image!!!" << endl;
            return -1;
        }

        imshow("in",src);
      
        Mat gray;

        if (src.channels() == 3)
        {
            cv::cvtColor (src, gray, CV_BGR2GRAY);       
        }
        else
        {
            gray = src;
        }
     

        Mat bw2 = Mat::zeros (gray.size(), CV_8UC1);
        thresholdIntegral (gray, bw2);

        cv::imshow ("binary integral", bw2);
  
        //picshadowx (bw2);
        picshadowy (bw2);
        waitKey (0);
    
    return 0;
}

 输入图片：

二值图片：

 水平投影：

 

垂直投影：

 行切割：

该处理方法，对印刷体有较好的效果，因为印刷体的行列区分明显，因此可以很快的进行行与列的分割。

---

by Shawn Chen，2017.12.13日，晚。