A guide to connecting Matlab with OpenCV

This webpage provides a short guide to connecting Matlab with OpenCV. 

Matlab provides a MEX environment in order to write C functions instead of M-files. Recall that MEX (Matlab-EXecutable) files are dynamically linked subroutines from C/C++ code (or Fortran code) that, when compiled, can be run from within Matlab like M-files. Hence, MEX environment offers a way to call your custom C/C++ routines as if they were Matlab built-in function. A detailed MEX Guide is offered by MathWorks HERE

The following description presupposes that, except for Matlab, OpenCV is installed in your system. Note that connection has been verified in a 64-bit machine with Win7 (cl compiler of VC++ 2008), OpenCV2.1 and 32-bit Matlab R2009b. If you have installed a 64-bit Matlab version, try to use OpenCV 2.3 (or 2.3.1) that offers a prebuilt library for both 32-bit and 64-bit systems. Otherwise you sould re-build the libraries from source-code using the proper generator (32-bit or 64 bit). Note that 32-bit Matlab works in 64-bit machines as well.

I provide below two examples of MEX-files that use OpenCV functions. The first example is the simplest case where no input/output arguments are required. For example, image loading is called from within the function and the MEX file returns nothing (it just displays the result). The second case is more interesting since input and output arguments are Matlab variables. The Sun Peng's types convertor that provides bidirectional conversion between C/C++ (OpenCV)types and matlab matrices is used by this case.

Example 1: A simple MEX file without input/output arguments that uses simple OpenCV modules

/*******************************************************************
 This is a simple MEX file. It just calls openCV functions for loading,
smothing and diaplaying the results
********************************************************************/
 
#include <stdio.h>
#include <stdlib.h>
#include <opencv/cv.h>
#include <opencv/highgui.h>
 
#include "mex.h"
 
int smoothImage(char* filename){
 
    //load the image
    IplImage* img = cvLoadImage(filename);
    if(!img){
        printf("Cannot load image file: %s\n",filename);
    return -1;
    }
 
    //create another image
    IplImage* img_smooth = cvCloneImage(img);
 
    //smooth the image img and save the result to img_smooth
    cvSmooth(img, img_smooth, CV_GAUSSIAN, 5, 5);
 
    // create windows to show images
    cvNamedWindow("Original Image", CV_WINDOW_AUTOSIZE);
    cvMoveWindow("Original Image", 100, 100);
    cvNamedWindow("Smoothed Image", CV_WINDOW_AUTOSIZE);
    cvMoveWindow("Smoothed Image", 400, 400);
 
    // show the images
    cvShowImage("Original Image", img );
    cvShowImage("Smoothed Image", img_smooth );
 
    // wait for a key
    cvWaitKey(0);
 
    //destroy windows
    cvDestroyWindow("Original Image");
    cvDestroyWindow("Smoothed Image");
 
    //release images
    cvReleaseImage(& img);
    cvReleaseImage(& img_smooth);
 
    return 0;
};
 
void mexFunction(int nlhs, mxArray* plhs[], int nrhs, const mxArray* prhs[]){
    if (nrhs != 0) {
        mexErrMsgTxt("Do not give input arguments.");
    }
    if (nlhs != 0) {
        mexErrMsgTxt("Do not give output arguments.");
    }
    char *name = "cameraman.png";
    smoothImage(name);
}

Let us suppose that the above code is saved to  displayImage.cpp file. Since this file makes use of OpenCV header files, the compiler must be informed with the appropriate include directories. Moreover, in order to produce the executable file, the necessary libraries should be given.

There are two ways to define the required compilation flags:

1. by pre-editing the MEXOPTS.BAT file

2. by defining paths and libraries with the mex command

/*******************************************************************
 This is a simple MEX file that accepts as inputs an image and the
 size of a Gausian filter(two parameters). Then it applies the filter
 to the image by calling the OpenCV function cvSmooth and returns the 
 filtered image to a matlab variable.
********************************************************************/

#include <opencv/cv.h>
#include <opencv/highgui.h>
#include <opencv/cxcore.h>

#ifndef HAS_OPENCV
#define HAS_OPENCV
#endif

#include "mex.h"
#include "mc_convert.h"
#include "mc_convert.cpp"

void mexFunction(int nlhs, mxArray *plhs[], int nrhs,
        const mxArray *prhs[]) {
    
    //Read Matlab image and load it to an IplImage struct
    IplImage* inputImg = mxArr_to_new_IplImage(prhs[0]);
    
    //Read the filter parameters
    double filterHeight, filterWidth;
    mat_to_scalar (prhs[1], &filterHeight);
    mat_to_scalar (prhs[2], &filterWidth);
            
    //smooth the input image and save the result to outputImg
    IplImage* outputImg = cvCloneImage(inputImg);
    cvSmooth(inputImg, outputImg, CV_GAUSSIAN, filterWidth, filterHeight);
    
    //Return output image to mxArray (Matlab matrix)
    plhs[0] = IplImage_to_new_mxArr(outputImg);
    cvReleaseImage(&inputImg);
    cvReleaseImage(&outputImg);
    
}

Note that Matlab-to-OpenCV convertor needs to define the symbol name HAS_OPENCV to the C preprocessor. I define it in the source code, but one can either add the switch -DHAS_OPENCV to mex command or edit the COMPFLAGS definition in MEXOPTS.BAT (find the SET COMPFLAGS line) and add the /DHAS_OPENCV flag.