This project is test problem: image library.

To install application try following commands:
    ./configure
    make

To install from GIT repository run before
    autoreconf -vis



As for me performance is quite important for an image library. In C++
templates provide best performance for algorithms with abstract types.
So I selected a template class/function as internal implementation of
image algorithms. It gives to compiler opportunity to apply maximum 
optimization techniques.



From another side interface to user in my view might be as simplest as possible.
So I decided to remove any templates. Client should know nothing about internals 
and template magic inside. For user most simple way to set pixel format in class 
arguments and just to call functions without any magic. So we gives to client
only one class Image with common methods like create/cloneTo/convolve/scale with
getters and setters. This is wrapper on template algorithm implementation.



Library implementation may be divided into two parts. First part is algorithms
directly. Second part is proxy between simple class Image and algorithms. 



Algorithm's part is more simple. I did not put too many effort into optimal 
implementations. I tried to get the simplest one to illustrate.

Usually an algorithm has a declaration like it shown below:

template <class ImgData>
    void ConvolutionAlgorithm(
        ImgData * destination, 
        const ImgData * source, 
        const std::vector<ConvolutionElement> & convolutionVector
    );

where ImgData is a class which supports methods getPixel and setPixel to get/set
am image pixel. For pixels might be suported the following methods:
    pixelAdd         - add pixel;
    pixelScale       - multiplying pixel to coefficient;
    pixelSetDefault  - default value for pixel.
This is enought for existing image algorithms.



Proxy part is wrapper on actions. His name is ActionWrapper. It contains service 
code for creating temporary buffers, preserve data so on. At the end wrapper calls 
algorithm. ActionWrapper class is a template which supports three arguments. They
are pixel format, pixel type, pixel. Wrapper might call the appropriate algorithm
based on variable data. To make it simple some preprocessor magic is user. First
at all ActionWrapper supports method check to answer “Is it suitable template
instance for these arguments?” Simple macroses contains list of all supported
combination format/type and selecting best one. This is variadic macroses.