Summary: An Efficient Gradient based Registration Technique for
Marco Reisert, Olaf Ronneberger, Hans Burkhardt
Georges Koehler Allee 52
This paper presents a coin recognition system based completely on the direction of the
gradient vectors. To optimally align two coins we search for a rotation such that as most
as possible corresponding gradient vectors point into the same direction. After discretizing
the gradient directions this can be done quickly by the use of the Fast Fourier Transform.
The classification is done by a simple nearest neighbor search followed by several rejection
criteria to meet the demand of a low false positive rate.
The goal of a coin recognition system is to automatically sort and classify high volumes of coins
with high accuracy within a small amount of time. Usually the classification is based on images
from both sides of the coin, its radius and thickness. With the changeover of 12 European cur-
rencies to the Euro this task has gained much more interest. In 2003 ARC Seibersdorf research
GmbH created the sorting device called Dagobert . The recognition unit of Dagobert is able