Collection and development kit of MATLAB MEX functions for OpenCV library.

The package provides MATLAB MEX functions that interface with hundreds of OpenCV APIs. Also the package contains a C++ class that converts between MATLAB's native data type and OpenCV data types. The package is suitable for fast prototyping of OpenCV application in MATLAB, use of OpenCV as an external toolbox in MATLAB, and development of custom MEX functions.

The current version of mexopencv is compatible with OpenCV 3.3. For older OpenCV versions, please checkout the corresponding branches (v3.2, v3.1, v3.0, v2.4, v2.3, and v2.1).

Consult the wiki for help.

• Structure
• Build
  • Linux
  • OS X
  • Windows
• Usage
  • Documentation
  • Unit Testing
• License

The project tree is organized as follows:

+cv/             OpenCV or custom API directory
+mexopencv/      mexopencv utility API directory
doc/             directory for documentation
include/         header files
lib/             directory for compiled C++ library files
samples/         directory for sample application codes
src/             directory for C++ source files
src/+cv/         directory for MEX source files
src/+cv/private/ directory for private MEX source files
test/            directory for test scripts and resources
opencv_contrib/  directory for sources/samples/tests of additional modules
utils/           directory for utilities
Doxyfile         config file for doxygen
Makefile         make script
README.markdown  this file

Prerequisite

• MATLAB or Octave (>= 4.0.0)
• OpenCV (3.3.0)

Depending on your platform, you also need the required build tools:

• Linux: g++, make, pkg-config
• OS X: Xcode Command Line Tools, pkg-config
• Windows: Visual Studio

Refer to the Makefile and make.m scripts for a complete list of options accepted for building mexopencv across supported platforms.

Refer to the wiki for detailed build instructions.

Currently, mexopencv targets the final 3.3.0 stable version of OpenCV. You must build it against this exact version, rather than using the bleeding-edge dev-version of opencv and opencv_contrib. UNIX users should consider using a package manager to install OpenCV if available.

• OpenCV
• OpenCV contributed modules

DO NOT use the "master" branch of opencv and opencv_contrib! Only the 3.3.0 release is supported by mexopencv.

First make sure you have OpenCV 3.3.0 installed in the system:

• if applicable, install OpenCV 3 package available in your package manager (e.g., libopencv-dev in Debian/Ubuntu, opencv-devel in Fedora). Sadly the OpenCV package provided in Ubuntu 17.04 is still behind with OpenCV 2.4, which can only be used with the v2.4 release of mexopencv.

• otherwise, you need to build and install OpenCV from source:

    $ cd <opencv_build_dir>
    $ cmake <options> <opencv_src_dir>
    $ make
    $ sudo make install
  

At this point, you should make sure that the pkg-config command can identify and locate OpenCV libraries (if needed, set the PKG_CONFIG_PATH environment variable to help it find opencv.pc):

$ pkg-config --cflags --libs opencv

If you have all the prerequisites, go to the mexopencv directory and type:

$ make

This will build and place all MEX functions inside +cv/. Specify your MATLAB directory if you installed MATLAB to a non-default location:

$ make MATLABDIR=/opt/local/MATLAB/R2017a

You can also work with Octave by specifying:

$ make WITH_OCTAVE=true

To enable support for contributed modules, you must build OpenCV from both opencv and opencv_contrib sources. You can then compile mexopencv as:

$ make all contrib

Optionally you can test mexopencv functionality:

$ make test

Developer documentation can be generated with Doxygen if installed:

$ make doc

This will create HTML files under doc/.

Refer to the wiki for detailed instructions on how to compile OpenCV for both MATLAB and Octave.

Currently, the recommended approach to install OpenCV in OS X is Homebrew. Install Homebrew first, and do the following to install OpenCV 3:

$ brew install pkg-config homebrew/science/opencv3
$ brew link opencv3

Otherwise, you can build OpenCV from source, similar to the Linux case.

If you have all the prerequisites, go to the mexopencv directory and type:

$ make MATLABDIR=/Applications/MATLAB_R2016a.app PKG_CONFIG_MATLAB=opencv3 LDFLAGS=-L/usr/local/share/OpenCV/3rdparty/lib -j2

Replace the path to MATLAB with your version. This will build and place all MEX functions inside +cv/.

Refer to the wiki for detailed instructions on how to compile OpenCV on Windows, and build mexopencv against it.

In a nutshell, execute the following in MATLAB to compile mexopencv:

>> addpath('C:\path\to\mexopencv')
>> mexopencv.make('opencv_path','C:\OpenCV\build')

Replace the path above with the location where OpenCV binaries are installed (i.e location specified in CMAKE_INSTALL_PREFIX while building OpenCV).

Contrib modules are enabled as:

>> addpath('C:\path\to\mexopencv')
>> addpath('C:\path\to\mexopencv\opencv_contrib')
>> mexopencv.make('opencv_path','C:\OpenCV\build', 'opencv_contrib',true)

If you have previously compiled mexopencv with a different configuration, don't forget to clean old artifacts before building:

>> mexopencv.make(..., 'clean',true)

Once MEX functions are compiled, you can add path to the project directory and call MEX functions within MATLAB using package name cv.

addpath('/path/to/mexopencv');
addpath('/path/to/mexopencv/opencv_contrib');

% recommended
out = cv.filter2D(img, kern);  % with package name 'cv'

% not recommended
import cv.*;
out = filter2D(img, kern);     % no need to specify 'cv' after imported

Note that some functions such as cv.imread will overload MATLAB's built-in imread function when imported. Use the scoped name when you need to avoid name collision. It is also possible to import individual functions. Check help import in MATLAB.

Check a list of functions available by help command in MATLAB.

>> help cv;              % shows list of functions in package 'cv'

>> help cv.VideoCapture; % shows documentation of VideoCapture

Look at the samples/ directory for examples.

mexopencv includes a simple documentation utility that generates HTML help files for MATLAB. The following command creates HTML user documentation under doc/matlab/ directory.

addpath('/path/to/mexopencv/utils');
MDoc;

On-line documentation is available.

You can test the functionality of compiled files by UnitTest class located inside test directory.

addpath('/path/to/mexopencv/test');
UnitTest;

Look at the test/unit_tests/ directory for all unit-tests.

The code may be redistributed under the BSD 3-Clause license.