“…Nevertheless, due to the straightforward structural architecture of CNN that restricts itself to local couplings, hardware realizations as integrated circuits (e.g., in CMOS technology) of such networks are already available and offer the potentiality of real-time applications for a massively parallel signal processing. Recent studies have shown that robust templates, which do not only work on CNN simulators, but also on hardware implementations can be found using an on-chip optimization (Feiden and Tetzlaff, 2003). Further research will show whether our CNN-based classification approach can be transferred onto a hardware-CNN without reducing the performance.…”