117
methods for template calculation. We have then shown an example direct mapping
of PDE’s to templates for a selected image processing task. For a fast template
calculation we have shown that there is a way to implement GA on FPGA and
optimize any templates much faster than on computer/CPU.
Research question 5: How far can the advantages of "analog computing" be
used/gained through an emulation of analog computing on digital hardware
platforms like FPGA (for the benefit of an ultrafast image processing)?
The main advantage of analog computing is that we can “nearly” simultaneously get
the result of complex mathematical differential equations. All signals are generated
in parallel and in real-time and the electronic components do compute
simultaneously. For this research question we have developed and implemented an
emulation of analog computing on FPGA. The result is at least many thousand times
faster than desktop computers. This result has been a good step for implementing
complex models like CNN on FPGA.
Research question 6: How far can an efficient implementation of CNN on
