For Developers

This describes how to build Matchbox from the source distribution.

Alexander Schindler alexander.schindler@ait.ac.at
Roman Graf roman.graf@ait.ac.at

Requirements

CMake

To build OpenCV and the Matchbox commandline tools Visual Studio solutions or
Makefiles for MinGW or gcc have to be generated using CMake.

Installing CMake

For Windows, download cMake from the cMake
website http://www.cmake.org and install it.

For Linux, cMake usually can be installed via the distribution package manager.
E.g. for Ubuntu

sudo aptitude install cmake

Open Computer Vision Library (OpenCV)

The Open Source Computer Vision (OpenCV) libarary opencv_library is a
library of programming functions for real time computer vision. It is released
under a Berkeley Software Distribution (BSD)
http://opensource.org/licenses/bsd-license.php license and thus is free for
commercial or research use. The library provides a wide range of implemented
algorithms on image analyses, image data and matrix manipulation, basic and
advanced image processing, object recognition and image feature extraction.

Most of Matchbox’es implementation is based on routines provided by OpenCV.

Installing OpenCV

The OpenCV source code can be downloaded from the OpenCV
website http://opencv.willowgarage.com/wiki/ for all platforms.
It is necessary to generate either Visual Studio solutions or Makefiles with
cMake to build it.

To do this, start CMake, browse to the directory where the downloaded OpenCV
source code resides, choose a directory where OpenCV will be built, click
‘Configure’ to choose which compiler you want to use, click ‘Finish’ and
then click ‘Generate’.

The Visual Studio solution or the Makefile can then be found in the build
directory.

Python

Python is a popular dynamic programming language. Python is used to model
diverse use cases and workflows of document image quality assurance.

Modules required by Matchbox:

  • Numpy – Numerical Python
  • Argparse – Parser for command-line options, arguments and sub-commands
Installing Python

Python comes with most linux distributions out of the box. For Windows, install
Python from the Python website http://www.python.org. The
additional libraries can also be found on the website or in the package
repositories of the Linux distribution of your choice.

C++ Compiler

Linux

Matchbox can be built with gcc, a compiler that is shipped with most Linux
distributions.

Windows

On Windows, either Visual Studio or MinGW (MinGW is open source software and can
be downloaded from their website http://www.mingw.org/ for
free) can be used to build the code.

Optional Packages

  • Intel Threading Building Blocks Library (TBB)

Installing Matchbox

Installation on Linux

Manual compilation:
Manual compilation should be straightforward:
After you made sure that the requirements are fulfilled, navigate to the source
code directory and enter

mkdir build 
cd build 
cmake .. 
make sudo 
make install 

In some cases it might be necessary to copy the files ‘MatchboxLib.py’ and
‘FindDuplicates.py’ manually from the ‘Python’ source directory to
‘/usr/bin’ or ‘/usr/local/bin’ and set the executable flag appropriately.

If the compilation succeeded, the commandline tools extractfeatures,
train, compare, FindDuplicates.py and CompareCollections.py are available.

Installation on Windows

Compiling Matchbox using MinGW

An environment variable called OPENCV_DIR, pointing to the directory where
OpenCV was built, has to be defined. This can be done by right-clicking My
Computer'', choosing
Properties” and changing to the Advanced'' tab. After
clicking the button
Environment variables”, an environment variable can be
added.

As a next step, a MinGW Makefile has to be generated using CMake.
Start CMake, click ‘Browse Source’ and find the directory containing the
Matchbox source. Click ‘Browse Build’ and choose a directory where the code
will be built. Click ‘Configure’ and choose ‘MinGW Makefiles’ from the
dropdown menu. Below the dropdown menu, ‘Use default native compilers’ has to
be selected. Click ‘Finish’.
After cMake finished configuring, click ‘Generate’. After cMake is done, the
Makefile can be found in the build directory.

Start the MinGW shell and navigate to the directory you specified as build
directory. Enter the command mingw32-make and press enter. The sourcecode
should now compile successfully. Issuing the command mingw32-make install
copies the executables and libraries to the right places in the filesystem.