significantly improved parallel non-local means by using granularity parameter in parallel_for_ loop. Because the algorithm deals with sliding sums, it's essential that each thread has enough work to do, otherwise the algorithm gets higher theoretical complexity and thus there is no speedup comparing to 1-thread code (at best).

Changed parameter type of fastNlMeansDenoising[Multi][Abs] from std::vector<float> to const std::vector<float>&

Changed fastNlMeansDenoising and fastNlMeansDenoisingMulti back to sum of sq distances. Moved sq sum of abs distances to fastNlMeansDenoisingAbs and fastNlMeansDenoisingMultiAbs

Preparation for 16-bit colored denoising. Currently not working due to cvtColor not supportint 16-bit Lab conversion.

Basic 16-bit implmentation of fastNlMeansDenoising. Table-based exponetiation leads to high memory footprint and loss of precision in 16-bit mode.

Minimal image size is set to 5*5 for OCL path of fastNlMeansDenoising and fastNlMeansDenoisingColored since it breaks tests with test_loop_times=30 and in general doesn't make sense for computation on GPU.

This should allow many algorithms to take advantage of more parallelization
technologies.

e.g. <opencv2/core/core.hpp> become <opencv2/core.hpp>

Also renamed <opencv2/core/opengl_interop.hpp> to <opencv2/core/opengl.hpp>

Made all STL usages explicit to be able automatically find all usages of
particular class or function.

* reorder arguments
* rewrite accuracy tests
* replace doubles with integer arithmetic inside the main loop

All the functions from it are moved to the header
opencv2/photo/photo.hpp

enabled SSE3 by default; integrated SSE3-optimized bilateral filter (by Grigoriy Frolov); modified API of non-local means (use Input/OutputArrays)