add a 'split' train method to datasets/lfw_benchmark

[fix issue #3996] delete private.hpp include in sample code

fix issue for vs versions which aren't standarized for the c99 standard

corrected the testing data path for latentsvm detector and some typos

Link Correction

Adapt reg sample to changes in features detectors

Bug fixing

Adapt to changes in the interface of feature detectors, using now
Feature2D interface.

Wrap face module with python and java

The bug has the same origin as:
https://github.com/Itseez/opencv_contrib/pull/187
The problem was in "fast" version of function generating a binary code
for ensemble classifier posterior probability calculation. Bug was
eliminated by fixing the order of "values" in offsets calculating
function

Unused header

standard

Replaced property macros in headers

Corrected Link for corresponding dataset

Bug fix #1

Removed sphinx documentation

The bug appears on the initialization step, if the initial bounding box width was larger than height. The problem was in function calculating a binary code for ensemble classifiers. In this function “measurement” values were used in wrong order (h1,w1,h2,w2; instead of: w1,w2,h1,h2), generating a access violation error on “patch” data access by the reason that width>height and pointer to data is calculated linearly data.step*height+width

fix missing delete[] calls in or_mnist.cpp

Java and python wrappers for contrib modules

Function cuda::integral() does not require an extra buffer any more.

Remove algorithm factory

contrib_world module

updating github documentation structure

Fixed cvv module build