Made small change which fixes the situation, when ORB does not find any descriptors on an image. Earlier ORB::operator() crashed in this case, since rowRange method was called for an empty matrix.

lot's of changes; nonfree & photo modules added; SIFT & SURF -> nonfree module; Inpainting -> photo; refactored features2d (ORB is still failing tests), optimized brute-force matcher and made it non-template.

minor build system changes (now cuda code in opencv_core is compiled using CUDA_ARCH* cmake variables)

 [~] Automatically tracked dependencies between modules
 [+] Support for optional module dependencies
 [+] Options to choose modules to build
 [~] Removed hardcoded modules lists from OpenCVConfig.cmake, opencv.pc and OpenCV.mk
 [+] Added COMPONENTS support for FIND_PACKAGE(OpenCV)
 [~] haartraining and traincascade are moved outside of modules folder since they aren't the modules

Refactored performance tests. TEST_CYCLE macro is renamed to TEST_CYCLE_N; SIMPLE_TEST_CYCLE is renamed to TEST_CYCLE; from now 100 iterations are default for performance tests

updated ORB: limit the number of output keypoints, use bi-linear interpolation between subsequent layers instead of much slower area-based interpolation between 0-th and i-th layers.

added improved ORB implementation, convex-convex polygon intersection, eigen2x2 low-level function ...

fixed problem with incorrect distance values returned by FlannBasedMatcher for hamming metric (int's were interpreted as floats)

updated feature_homography.py sample to use new features

wrapped StarFeatureDetector and GridAdaptedFeatureDetector

partially wrapped features2d framework
added feature_homography.py sample

fixed crashes in parallel HOG detector; avoid use of new POPCNT instruction in ORB (to make it compatible with older CPUs).