- made some of dependencies explicit
- removed dependencies to highgui and some other modules where possible
- modified some samples to build without modules

* Added files from http://code.opencv.org/issues/1053

The code structure had to be changed because of the
features2d/xfeatures2d split and to avoid changing
multiple modules (gaussian_pyramid doesn't seem ready
for inclusion in the (x)imgproc module).
Except for changes related to this, the patches
remain unchanged, including the outdated
copyright notice at the top of the files.

* Hide Harris-Laplace implementation behind pimpl-creating wrapper.

* Add test for Harris-Laplace feature detector.

* Handle empty images in Harris-Laplace detector.

* Replace HarrisAffineFeatureDetector with more flexible AffineFeature2D

* collapse unnecessary HarrisLaplace class into HarrisLaplaceFeatureDetector

* fold DoG pyramid code into harris_laplace_detector.cpp

* tuck auxiliary functions for Harris-Laplace into anonymous namespace

* use mask to filter keypoints in Harris-Laplace detector

* remove unused value of differentiation scale from calcAffineAdaptation

* make descriptor copy in AffineFeature2D independent of the data type

* Document interface for Harris-Laplace and AffineFeature2D

* Make type conversions explicit and decorate float literals with f in HarrisLaplace and AffineFeature2D

* Replace usage of non-standard log2 in Harris-Laplace feature detector

* Fix virtual overload errors in AffineFeature2D

* Add basic tests for AffineFeature2D and fix what they uncover

* Make type conversions in Harris-Laplace feature detector explicit

* Change license header for Harris-Laplace detector and AffineFeature2D

* Use Matx for small matrices in AffineFeature2D

* Remove redundant attributes of Elliptic_KeyPoint

* Convert Matx22f to Matx22d for inverting in AffineFeature2D

Add CV_Assert for color image.

Implementation of PCT signature extractor (position-color-texture) and evaluation of their Signature Quadratic Form Distance (SQFD).

Minor clean up. There was no need to keep these lines.

fix for feature2d java wrappers as described in this post: http://answers.opencv.org/question/101675/surfsift-java-wrapper-for-opencv-3xosx-1011/

- with big images the int multiplication can overflow

Changed statements of type "#if __CUDA_ARCH__ >= 200" to
"#if defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 200" in order to
avoid warnings about __CUDA_ARCH__ being undefined.

- Saliency stl array instead of pointer array added
- Perf test in Extra Repository
- Epsilon change to 10E-3

This was already fixed here:
https://github.com/patricksnape/opencv_contrib/commit/8247a7a0f59ba6c1fcfbac6a660db3f8acdc9ff8

Now uses cvround

round does not exist in MSVC<2012 - added with the same code
as in line_descritor/binary_descriptor.cpp