GitLab

By default, VS2017 will cause CMake to build 32b.  This fixes that.  The permissive option was removed from non-windows builds as it could cause issues with non-gcc compilers.

I changed the `Readme.md` so that CMake auto-detects the architecture.
I added a switch in `CMakeLists.txt` so that the compiler options are chooses based on the architecture.
I tested the compilation on both Windows 10 and Ubuntu 20.04
Co-authored-by: Paul Bauriegel <paul.bauriegel@telekom.de>

Integral Image optimization has been replaced by the Chan algorithm, a cutting edge algorithm that is up to 3x faster while using only a fraction of the memory... and allows for processing large image files! 

The default grayscale algorithm has also changed from Qt to Mean.

Variance is now calculated using Population Variance instead of Sample Variance.  This only caused a 1 pixel difference with 2 algorithms.

ISauvola has been abstracted so that it is very easy to make "Improved" versions of any algorithm in the list, not just Sauvola.

Python bindings have been added along with a new set of demos for Python / PIL, C++ Qt and OpenCV, and JS with NodeJS and Web.

Updating the same memory has was proven safe in UpdateToBinary(...), so no temporary image is being used anymore. 

A BinarizationFactory class was added to help simplify using the library and for exposing new language bindings.

Adding WASM support to the Doxa framework.  This is driving our web demo.  The Bataineh algorithm was also added, along with other minor improvements.

Added Convolution and Edge Detector classes along with a color distance algorithm for the RGB color space.

Morphology

Cleaned up file existance checking using new filesystem::exists call, which also fixes a build issue with Clang.

Our new Morphology classes fixed an issue we had with calculating the min and max value of a window.  This would cause the wrong max value under certain circumstances.

Add gitattributes so that Git will not affect our image files by trying to change line ending character(s) by platform.

A very insidious issue crept up recently.  All of my Binarization tests were failing under Windows, but my Mac (using clang), generated identical images to what I was expecting.  This recently started happening all of the sudden, and even Otsu, my global thresholding algorithm, was effected.  It turns out that my base PPM image was not formatted correctly.  It used a Windows style return in the header which consists of 2 characters.  One of those characters was being read as image data, causing problems.  This has been corrected but the fact that it was working in Windows, and Clang builds on my Mac still work, leaves many questions.  I have some hunches.  This was no a code problem, just a bad sample image created by another tool.

Added an efficient algorithm for calculating the min and max of a local window.  Cleaned up some unit tests.

Fix missing Template and include Palette.hpp

Doxa 2.0 Pull Request

Doxa 2.0 - Slightly more OO-friendly with a more unified algorithm interface, new algorithms to the line-up, and other refinements.

Support for other compilers: Changed encoding to UTF-8 and resolved build errors expressed by Clang.  Added minor performance tweaks in regards to pixel traversal in Gatos.

8 bit gray scale

Storing all binary files as .pbm files.  Fixed the PBM read and write routine so it now works with Photoshop.  Updated all examples with the latest Kumite run since fixing the grayscale conversion issue.

Rectify a pixel processing problem when converting to grayscale.  This was found in the old code base.  The new method fixes this but creates different results, so all of the performance information needed to be updated.

Allow a calculator to be passed to the different algorithms to significantly speed up iterative tasks for the same image.  Broke up each algorithm into its own class.  Hopefully I didn't lose too much speed.