Ci dessous un mail d'un des développeurs de la team OpenCV envoyé sur la lite de diffusion OpenCV:


Dear all,

OpenCV team has just released beta version of the latest and greatest OpenCV v2.4.
Congratulations to everybody who took part in preparing the release and/or contributed new functionality, submitted bug reports, and everybody else who waited for it!

We put a lot of efforts to make it one of the most stable OpenCV releases, over 300 issues have been resolved in the past few months (http://code.opencv.org/projects/opencv/roadmap)
[BTW, everything, except for the packages themselves, is now hosted on our brand-new developer site http://code.opencv.org.]

Besides the numerous bug fixes, here are the major highlights of the release (for details, see http://code.opencv.org/projects/opencv/wiki/ChangeLog):

============

* We now have as much as 20 modules. The new modules are:

nonfree - SIFT and SURF feature detectors are moved there because of the potential patent issues. Good news is that SIFT is now ~4 times faster than in OpenCV 2.3.x.
videostab - experimental module for video stabilization. It's under active development, so treat it as some technology preview thing.
photo - another mostly "stub" module for computational photography. Currently, there is just inpainting there, but more stuff is expected to come.

[note that haartraining and traincascade apps are moved to the dedicated apps directory]

* A big part of C API was moved to legacy module, so you may need to update your #include's in order to build your apps. By 2.5 we plan to complete the transition to C++ API, though we will still retain the C API after that.

* A lot of new cool stuff in GPU module:

* 3 new optical flow algorithms
* 2 new features detectors
* displaying GpuMat in a highgui window
* improved performance in many functions

[BTW, there is also experimental OpenCL module by Yao Wang et al: http://code.opencv.org/svn/opencv/branches/ocl, which is not a part of the release]

* Significantly improved Android support, including support for Android x86. See http://code.opencv.org/projects/open..._Release_Notes for details.

* The current state-of-art detector for textureless objects, Line-MOD algorithm by S. Hinterstoisser, has been adopted by the algorithm author and Patrick Mihelich and included into objdetect module.

* Reworked features2d module with cleaner interface, including the base class Feature2D combines a feature detector with descriptor extractor for more efficient processing.

* 3 face recognition algorithms have been contributed by Philipp Wagner (contrib module).

* More advanced LogPolar algorithms contributed by Fabio Solari and Manuela Chessa (contrib module).

* New better EM implementation that should solve many issues of the previous implementation.

* Several algorithms, including color conversion, morphology, brute-force feature matcher, data type conversions, ... have been made multi-threaded using TBB and/or were optimized using SSE2.

* Also, we now have a comprehensive set of performance tests (which are google test-powered opencv_perf_<modulename> executables, similarly to existing opencv_test_<modulename> regression tests), so there is a solid foundation for the further code optimizations.

* New FFMPEG wrapper that utilizes multiple cores for video decoding was added.

* Added 2 new algorithms for 3d object pose estimation.

* Some nice touches to Canny and findContours have been made. The first one now handles color images; the second one can handle multi-level (i.e. labeled) images, not just black-n-white.

-----------------------

known issues:

* open issues at http://code.opencv.org/projects/opencv/issues
* failing tests at http://buildbot.itseez.com/ :[
* ^^^ in particular, the new ffmpeg wrapper, while being usable, is still been stabilized;
* ^^^ also, the old Python bindings (cv module) are quite problematic; it's recommended to switch to numpy-based cv2 bindings. The old bindings are to be dropped soon.

============

You are welcome to try it out and report any problems to our issue tracker (http://code.opencv.org).
The packages are located at SourceForge: http://sourceforge.net/projects/opencvlibrary/ - please, take the beta2 packages.

As usual, there is also a special branch (http://code.opencv.org/svn/opencv/branches/2.4/opencv) available, from which you can get the latest hot fixes.

For those who read until this point, here is a fancy visualization of OpenCV 2.4 creation process, done tonight by our team:
[ame="http://www.youtube.com/watch?v=ToD91PYaQOU"]OpenCV 2.4.0 - YouTube[/ame]

enjoy watching the video and using OpenCV!

Regards,
Vadim