In CUDA 6.0 there was a bug in NPP LUT implementation (invalid results when
src == 255). In CUDA 6.5 the bug was fixed.

Replaced NPP LUT call with own implementation (ported from master branch)
to be independant from CUDA Toolkit version.
(cherry picked from commit eaaa2d27)

(cherry picked from commit 634ffed4)

(cherry picked from commit 62f27b28)

(cherry picked from commit 599f5ef5)

(cherry picked from commit 86e12b60)

(cherry picked from commit 5dff283b)

(cherry picked from commit e11333dd)

(cherry picked from commit b2790973)

(cherry picked from commit e650d87e)

(cherry picked from commit 60a5ada4)

fix output matrix allocation in cv::subtract(cherry picked from commit 629461c8)

call dst.create(...) before using it(cherry picked from commit 4c66614e)

2.4

Corrected bug #1437. Corrected the cast of two short values from an lParam. lParam contains two signed-short, the position can be negative when another mouse event happens with a mouse_move (mouse_click, etc.)
The Microsoft documentation specifies NOT to use LOWORD and HIWORD macros to extract the x/y positions as it won't work correctly with multiple monitors. See http://msdn.microsoft.com/en-us/library/windows/desktop/ms645607(v=vs.85).aspx.

Fixed spelling of CV_INTER_LINEAR flag in geometric_transformations.rst