photocopied) refect IR wavelengths.
Once the hardware for the system has been
selected, images of the banknotes can be captured from the camera and image processing software can be developed. To facilitate
the programming of the camera, Qtechnology has developed a web-based interface that
enables the camera to be programmed remotely (Figure 5). The web-based tool accesses the
hardware of the camera through the Video-
4Linux interface and allows easy navigation
through tens of settings, such as the bit mode
of the sensor, the response curve, the triggering mode and the color map.
By using the web-based tools, a region of
interest (ROI) in the image of the Euronote can
be selected where details of a legitimate note
should not be visible when the note is illuminated by IR light. These tools also enable the frame
rate of the system to be set in software. It is possible, of course, to analyze the whole image of
the note in software. However, by using a smaller ROI, the maximum frame rate (up to more
than 1000 fps) of the system can be achieved.
Thanks to the PCIe Gen2 x4 link to the APU,
up to 2 GB/s of data can be received by the body.
Once the images have been sampled, they
can either be saved in a JPEG or in a lossless
format. To analyze the images, the developer can
use Jupyter Notebook ( http://jupyter.org). This
enables programmers to create and share documents that contain live code, equations, visualizations and explanatory text. Using the Notebook,
a developer can program the embedded system
in the Python programming language, which in
turn provides access to both Open CV (http://
opencv.org) and NumPy/SciPy ( www.numpy.
org) image processing libraries, enabling them
to be combined in a single program.
In the case of banknote identifcation, the algorithm was easily implemented using the Jupyter Notebook interface. The frst stage of the
image processing chain involved locating the
edges of the notes by applying a Canny flter
( http://bit.ly/VSD-CANNY) to the images of
the Euronotes. The numerical values for the
upper and lower thresholds were calibrated
using static interactive widgets, or, sliders.
With the edge detection performed, a fll-
ing algorithm was selected and applied to
fll out all closed regions in the images, after
which the contours of the notes were detected
to determine the specifc location of the notes
within the image (Figure 6).
Once the position of a note was determined, a ROI was located inside the image
corresponding to the location of features of a
legitimate bank note. These would be invisible when the note is illuminated by the IR
light source (Figure 7).
Because features on legitimate notes are
effectively invisible in the ROI, analyzing
this region can determine whether the note is
real or counterfeit. To do so, a simple threshold classifer was applied to determine the
number of pixels in the ROI. By setting the
thresholds of the classifer in software, it
is possible to determine the pixel intensity
values under the ROI and then determine the
legitimacy of the currency. Having done so,
the notes are then labeled to indicate whether they are real or counterfeit.
The system is light dependent, so it is important to
ensure that the samples of
the banknotes are well illuminated. Because of this, it
may be necessary to adjust
the threshold of the classifer
to accommodate the different
luminosity levels used to illuminate the banknotes.
Figure 4a and b: By illuminating the front of the banknote with IR light, only the emerald number, the right
side of the main image and the silvery stripe are visible. On the back, only the numerical value and the horizontal
serial number are visible.
Figure 5: Once the hardware for the system has
been selected, images of the banknotes can be
captured from the camera. To facilitate camera
programming, Qtechnology has developed a
web-based interface that enables camera features to be programmed remotely.