ljubobratovicrelja/libcv

Library for basic computer vision tasks.

LibCV

Small library containing data structures and algorithms for most basic computer vision tasks, developed as learning project and coding practice, to be used for my faculty projects.

Contents

libcv contains following:

• Array structures - with reference counted data, and slicing operators.
  • template vector (with heap data), and vectorx (with statically defined size).
  • template matrix (m by n) designed for large matrices, that could hold images.
  • image array structure abstraction with type defined in runtime, which essentially is an 3D array.
• template bounded priority queue structure
• basic k-d tree structure implementation with k-nearest neighbour search.
• 2D contour, polygon and region structures
• basic image manipulation algorithms.
• basic linear algebra operations.
• lmdif integration from cminpack.
• image i/o
• basic gui support for showing images

Dependencies

Libcv depends on following libraries:

• Qt4 (4.8.6)
• libpng (libpng16)
• libjpg (jpegsr8)
• lapacke
• blas

Documentation

Proper documentation is not available, and it will not be if there is no interest from other people to use this library. I'll try to document code in doxygen style, so if anyone is up to it, he can build doc files himself.

Compilation

LibCV was mainly developed on Ubuntu with gcc 4.9.2. Ubuntu 15.04 has all needed libraries (and corresponding versions) on it's repositories, which can be installed using apt-get.

It was not tested on other *nix systems, but as long as your able to use gcc with c++11 support, you could easily download, build and link needed libraries. Also building blas and lapack should not be of any trouble on any *nix system.

Custom CMake build flags

• WITH_GUI - if FALSE, ignores gui module, and doesn't link Qt
• WITH_JPEG - turn on jpeg i/o support. If FALSE, does not link libjpeg.
• DOUBLE_REAL - if TRUE, real type in LibCV is double, if FALSE it is of single precision.
• COMPILE_EXAMPLES - compile example programs in ./examples/ directory.

Compilation on Windows

LibCV has been tested on Windows 8 with Visual Studio 2013. Instructions here will guide compilation using these tools, but for any older VC should not be much different.

NOTE: In ./thirdParty/ directory are built libs using Visual Studio 2013 (VC12) x64. If you need to compile it using some other toolchain, here are instructions I've followed to compile dependencies with vc12:

Qt

Download Qt 4.8.6 source here: http://download.qt.io/archive/qt/4.8/4.8.6/

Compilation steps:

1. Open VS Command line
2. configure -release -opensource -no-libpng -no-libjpeg
3. nmake
4. nmake install

NOTE: The -no-libpng and -no-libjpeg flags are entered since specified versions of png and jpeg libraries are used, Qt internals would not match, and image i/o would not work - png would compile ok, but runtime error would be thrown on image loading, for jpeg I'm not sure, haven't test it.

Png

libpng 1.6 can be downloaded at sourceforge: http://sourceforge.net/projects/libpng/files/

You'd need zlib also. Both are easily compiled using CMake, and with vc12 should not be any trouble.

Jpeg

I've skipped compiling jpeg on windows. There's some instructions on the web how to build it using nmake, but I've had no luck. If anyone can help with that, please mail me, I'd be grateful.

OpenBLAS

Easiest way to get blas and lapacke is through OpenBlas. Here are some nice instructions on how to install it: https://github.com/arrayfire/arrayfire/wiki/CBLAS-for-Windows

As guide suggests, you'd need cygwin with x86_64-w64-mingw32-gcc and x86_64-w64-mingw32-gfortran, and with cygwin terminal, compilation is easy - just make, and make install:

make BINARY=64 CC=/usr/bin/x86_64-w64-mingw32-gcc-4.8.3.exe FC=/usr/bin/x86_64-w64-mingw32-gfortran.exe
make PREFIX=./package install

Step 8., 9. and 10. can be skipped, since resulting lib file (libopenblas.dll.a) can be successfully linked with VC.

Release Versions

Version 0.1 is tagged because it was the version which camera calibration is developed with. But I can't guarantee the stability of library inside - it is only preserved because of the camera calibration project.

Contributions

If anybody is interested to make a contrubution, you're welcome, and thank you! Of course at this point, coding conventions or such rules are not firmly established - as long as you're willing to contribute, just make a pull request, and we'll work it out...

LICENSE

Library is under MIT license. See LICENSE file for details.