Convert data from cv::Mat to mxArray

Question

I want to send a Matrix to Matlab using "engine.h" in c++ code. The fact is that I have the data inside a cv::Mat, and I need to sent a mxArray. I tried to use this expression but it doesn´t work:

cv::Mat _priorP;
_priorP = Mat::eye(13, 13, CV_32FC1);
mxArray *mat;
mat = mxCreateDoubleMatrix(13, 13, mxREAL);

memcpy(mxGetPr(mat),_priorP.data, 13*13*sizeof(double));

Anybody knows the correct way to do the conversion? Any help would be apreciated. Thanks.

EDIT

I found this way: https://mcmap.net/q/1410210/-convert-data-from-cv-mat-to-mxarray

Tamarau

Answer 1

This thread shows how to convert a CvMat to mxArray. Even though it's not exactly the conversion code you are looking for, it's pretty close.

This is a simple conversion and you should be able to adjust the code to work with cv::Mat instead of CvMat. If you can't, a quick hack is to convert your cv::Mat data to CvMat and then use the code below as is (taken from the link I suggested):

mxArray* CvMat_to_new_mxArr (const CvMat* mat)
{
  const int TYPE = cvGetElemType (mat);

  // 2-d image
  if (CV_64FC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_64FC1> (mat);
  }
  else if (CV_32FC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_32FC1> (mat);
  }
  else if (CV_32SC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_32SC1> (mat);
  }
  else if (CV_16SC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_16SC1> (mat);
  }
  else if (CV_16UC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_16UC1> (mat);
  }
  else if (CV_8UC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_8UC1> (mat);
  }
  else if (CV_8SC1 == TYPE) {
    return helper_2dcvmat_to_mat<CV_8SC1> (mat);
  }

  //Multi-dimensional arrays not supported, yet.
  /*
  // 3-d image
  else if (CV_64FC3 == TYPE) {
    return helper_rgbimage_to_mat<IPL_DEPTH_64F> (img);
  }
  else if (CV_32FC3 == TYPE) {
    return helper_rgbimage_to_mat<IPL_DEPTH_32F> (img);
  }
  else if (CV_8UC3 == TYPE) {
    return helper_rgbimage_to_mat<IPL_DEPTH_8U> (img);
  }
  */

  // unsupported conversion, return null mxArray
  return mxCreateDoubleMatrix(0,0,mxREAL);    
}


template<int TYPE>
mxArray* helper_2dcvmat_to_mat (const CvMat* mat)
{
  void* pBeg;
  int pitch;
  cvGetRawData(mat, (uchar**)&pBeg,&pitch);

  CvSize size = cvGetSize (mat);
  const mxClassID cid = cvm_traits<TYPE>::CID;
  mxArray* pArrOut =
  mxCreateNumericMatrix(size.height,size.width,cid,mxREAL);
  void* pBegOut = mxGetData(pArrOut);

  typedef mc_traits<cid>::CT T;
  pix_iterator_2d<T,eRowWise> it_src1(static_cast<T*>(pBeg),
  size.width,size.height,pitch);
  pix_iterator_2d<T,eRowWise> it_src2(static_cast<T*>(pBeg),
  size.width,size.height,pitch);
  it_src2.end ();
  pix_iterator_2d<T,eColWise> it_dest(static_cast<T*>(pBegOut),
  size.width,size.height);

  std::copy (it_src1,it_src2,it_dest);

  return pArrOut;
}

Sapphire

Thanks. So it is always messy to use engine.h with openCV, isn't it? I was expecting something more direct, but if there is no simpler solution I will try this one. – Tamarau

One Question. Are you considering that opencv stores data by row and matlab by column? – Tamarau

I must insist that you open the thread I suggested and take a look at the complete source code available there. According to the other users, that code works. But if it doesn't, at least is a start for you. – Sapphire

Answer 2

There is a library developed by Kota Yamaguchi at http://github.com/kyamagu/mexopencv The package contains a C++ class (called MxArray) that converts between Matlab's native data type (mxArray) and OpenCV data types. The library directly supports C++ API for OpenCV (Open CV version 2.0 and higher), so there's no need to do an extra conversions (e.g. from cvMat to cv::Mat or from IplImage to cv::Mat). Sample usage:

#include "mexopencv.hpp" // include the library
#include "highgui.h"     

void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) 
{
    cv::Mat image;
    image = imread("filename"); // read an image from file
    plhs[0] = MxArray(image);   // convert from cv::Mat to mxArray 
}

That's it. Make sure to compile your mex function together with MxArray.cpp file from the library; you can do so in MATLAB command line:

    mex yourmexfile.cpp MxArray.cpp

Trichocyst

The source code is enough for me, I just want to make my own function. Thanks a lot, very usefull indeed.. – Tamarau

@Tamarau Glad to hear! MxArray.cpp was very helpful to me too. Consider up-voting this answer, so others can see this solution. – Trichocyst

Answer 3

I found a simpler method to do this convertion after some effort. What I do is creating a function like this:

void arithmetic::cvLoadMatrixToMatlab( Engine *ep, const Mat& m, string name)
{   
   int rows=m.rows;
   int cols=m.cols;
   //Mat data is float, and mxArray uses double, so we need to convert.   
   mxArray *T=mxCreateDoubleMatrix(cols, rows, mxREAL);
   double *buffer=(double*)mxGetPr(T);
   for(int i=0; i<rows; i++){
       for(int j=0; j<cols; j++){
           buffer[i*(cols)+j]= (double)m.at<float>(i, j);
       }
   }

   //memcpy((char*)mxGetPr(T), (char*)m.data, rows*cols*sizeof(double));
   engPutVariable(ep, name.c_str(), T);
   name=name+"="+name+"'";                    // Column major to row major (mat=mat')
   engEvalString(ep, name.c_str());

   mxDestroyArray(T);
}

Tamarau