filter2D implementation details

edit

Hello,

Is it possible to have a more detailed description of the implementation of the filter2D() function?

Browsing and checking the source code I understood that in general it does the following:

if (kernel_size > some_threshold) then do correlation using dft(). else compute filtering using recursion.

But in both cases (using dft() or recursion), I'm unable to follow what the code actually does - what is the exact implementation.

So, can anyone provide a description of the implementation in each case using some high level pseudocode (or even a list of well defined steps)?

If the implementation in each case (using dft() or recursion) depends on some literature (books, papers), could anyone provide these literature sources?

Thanks in advance, Panos