概率霍夫变化源码

void HoughLinesP(InputArray _image, OutputArray _lines,
                 double rho, double theta, int threshold,
                 double minLineLength, double maxGap )
{
   
    CV_INSTRUMENT_REGION();
	// _image,_lines若是UMAT，使用基于OpenCL的GPU计算
    CV_OCL_RUN(_image.isUMat() && _lines.isUMat(),
               ocl_HoughLinesP(_image, _lines, rho, theta, threshold, minLineLength, maxGap));
	
    Mat image = _image.getMat();
    std::vector<Vec4i> lines;
    HoughLinesProbabilistic(image, (float)rho, (float)theta, threshold, cvRound(minLineLength), cvRound(maxGap), lines, INT_MAX);
    Mat(lines).copyTo(_lines);
}

我们暂且不看GPU计算的部分，HoughLinesP直接调用了 HoughLinesProbabilistic(image,(float)rho,(float)theta,threshold,cvRound(minLineLength),cvRound(maxGap),lines,INT_MAX)，其源码如下

/*************************************************************\
*                              Probabilistic Hough Transform                             *
\*************************************************************/

static void
HoughLinesProbabilistic( Mat& image,
                         float rho, float theta, int threshold,
                         int lineLength, int lineGap,
                         std::vector<Vec4i>& lines, int linesMax )
{
   
    Point pt;
    float irho = 1 / rho;
    RNG rng((uint64)-1);	//随机数

    CV_Assert( image.type() == CV_8UC1 );

    int width = image.cols;
    int height = image.rows;

    int numangle = cvRound(CV_PI / theta);	//角向采样数
    int numrho = cvRound(((width + height) * 2 + 1) / rho);	//径向采样数

#if defined HAVE_IPP && IPP_VERSION_X100 >= 810 && !IPP_DISABLE_HOUGH
    CV_IPP_CHECK()
    {
   
        IppiSize srcSize = {
    width, height };
        IppPointPolar delta = {
    rho, theta };
        IppiHoughProbSpec* pSpec;
        int bufferSize, specSize;
        int ipp_linesMax = std::min(linesMax, numangle*numrho);
        int linesCount = 0;
        lines.resize(ipp_linesMax);
        IppStatus ok = ippiHoughProbLineGetSize_8u_C1R(srcSize, delta, &specSize, &bufferSize);