Open Source Computer Vision Library

Many functions and methods now take InputArray/OutputArray instead of "cv::Mat" references. LAPACK was replaced by an implementation of Jacobi SVD, decreasing footprint and compile time, and improving performance and accuracy. New BRIEF and ORB feature descriptors were added. A new calibration pattern "circles grid" was added. A new experimental variational stereo correspondence algorithm StereoVar was added. Support for Ximea cameras was added. Python modules were enhanced. Various other enhancements were made. Documentation was improved. Many bugs were fixed.

The library was reorganized into 12 smaller instead of 4 larger modules, while maintaining backward compatibility. An experimental CUDA acceleration module was added. Many improvements and enhancements were made. Bugs were fixed.

All of OpenCV now uses exceptions instead of the old libc-style mechanism. Parallel loops in OpenCV were converted from OpenMP to Intel TBB. SWIG-based Python bindings are no longer supported, in favor of the new interface. More complete support for 64-bit OSes was implemented, including HighGUI. Dynamic host CPU detection was implemented. The GrabCut image segmentation algorithm was added. A better stereo correspondence algorithm (SGBM) was added. The gstreamer backend was improved. For build, autotools were completely replaced by CMake. Tons of bug fixes and test enhancements were made.

A new C++ interface was added in addition to the existing OpenCV 1.x C API, bringing automatic memory management and many other advantages. A lot of new functionality, especially in the area of feature detectors and descriptors, was implemented. SSE2 optimization of many functions was made. The build system was modified to use CMake, and the directory structure was revised. The documentation was converted to LaTeX and improved.

This is the first stable version of OpenCV after
more than 6 years of development.