The present invention relates to a method and
device for controlling banknotes.
As is known, banknotes are produced from special
sheets (typically comprising watermarks and/or metal
bands) large enough to accommodate several finished
banknotes, and which are subjected to various printing
steps, using different printing methods, to obtain the
various graphic and alphanumeric characters. More
specifically, printing may comprise some or all of the
following steps:
a) Offset printing. This is performed out of line
with the edge of the sheet, which therefore cannot be
used as a reference by which to determine the
coordinates of the offset-printed details. Offset
printing is normally performed on both sides of the
sheet. b) Copperplate printing. This is performed at high
pressure, may be displaced with respect to the offset
printing, and slightly deforms the paper, thus possibly
resulting in inclination of the copperplate with respect
to the offset printing. Copperplate printing may only be
performed on the front or both sides (front-back) of the
banknote, and may comprise several steps, each of which
may be horizontally/vertically misaligned or inclined
with respect to the others and with respect to the
offset printing. c) Silk-screen printing. Like copperplate
printing, this may be displaced or inclined with respect
to the offset printing.
Following the above printing steps, the printed
sheet is quality controlled, and only the passed
banknotes are printed with serial numbers. Finally, the
sheet is cut to separate the banknotes, but cutting is
not performed in line with any of the printed details.
Quality control is currently performed manually to
ensure the various printed details conform closely
enough with an ideal value, and that there are no errors
in colouring (too much ink or none at all), no smudges,
etc.
At present, there is no way of automatically
controlling the print quality of banknotes, in that, to
take into account the numerous variables involved, the
deviation thresholds used to compare the banknote with a
specimen image would have to be so high that even
banknotes with serious errors in colouring would be
passed.
Automatic control systems do exist for validating,
discriminating between, or determining the deterioration
of banknotes already in circulation, but which provide
for examining only a very small portion of the note
(typically a narrow horizontal intermediate strip
through significant parts of the overall design). The
information supplied by such systems is therefore
insufficient for quality control purposes, in which case
the inking defects and smudges for detection are
normally localized.
It is an object of the present invention to
provide a reliable, automatic method and device for
controlling banknotes.
According to the present invention, there is
provided a method of controlling banknotes,
characterized by comprising the steps of acquiring an
image of a whole banknote, and comparing said acquired
image with a reference image of a whole specimen
banknote.
According to the present invention, there is also
provided a device for controlling banknotes,
characterized by comprising acquisition means for
acquiring an image of a whole banknote; and comparing
means for comparing said acquired image with a whole
reference image of a specimen banknote.
A non-limiting embodiment of the present invention
will be described by way of example with reference to
the accompanying drawings, in which:
Figure 1 shows an overall block diagram of a
device in accordance with the present invention; Figure 2 shows a flow chart of the method
according to the present invention; Figures 3 and 4 show block diagrams of details in
Figure 1; Figure 5 shows a plan view of a specimen banknote
indicating specific lines used in the method according
to the present invention.
Number 1 in Figure 1 indicates as a whole a device
for quality controlling banknotes 2 printed on a sheet
3.
More specifically, control device 1 comprises a
television camera 4 for picking up one banknote at a
time, and for generating and supplying a digitized
discrete grey-tone television signal to an image memory
5. Image memory 5 memorizes the image of banknote 2 in
the form of a matrix of dots (pixels), each of which is
assigned a value (also referred to hereinafter as a
shade value) related to the grey level (luminance) of
the pixel.
Image memory 5 is connected to an image processor
6 for performing a first processing operation of the
image of banknote 2, and for determining the coordinates
of the image with respect to a predetermined reference
system used later for comparison with a specimen
banknote. Image processor 6 is therefore connected to a
specimen memory 7, by which it is supplied with selected
portions of a specimen banknote, which are compared with
similar portions of the image to be repositioned. The
output of image processor 6 is connected to an edge
extractor 8, which receives the shifted image of the
banknote to be controlled, and processes the shifted
image to generate a processed image, the pixels of which
define the edges of the drawings and alphanumeric
characters on the banknote, and the portions of the
banknote having a brightness gradient with respect to
the adjacent portions.
Edge extractor 8 is connected to a local averaging
and comparing unit 9 for locally averaging the processed
image received from edge extractor 8, and for making a
local comparison with corresponding image portions of
the specimen banknote - also averaged - supplied by
specimen memory 7. Local averaging and comparing unit 9
also processes the local comparison results of all the
portions, and supplies, at an output 10, a signal S
accepting or rejecting the controlled banknote 2. A
control unit 11 is connected to units 5-9 to control the
operation sequence as well as any processing parameters.
With reference to Figure 2, the control method
shown comprises a first step of acquiring and memorizing
the image of a whole banknote by means of camera 4 and
image memory 5 (block 12). From the acquired image
(block 13), image processor 6 selects a number of small
predetermined regions containing predetermined
significant characteristics of the banknote, taking into
account any position inaccuracy resulting from
displacement of the banknote with respect to the
theoretical position, and from the printing deviations
described previously. For example, the predetermined
regions may be such as to definitely contain the edge
portions defined by lines A and B in Figure 5.
The selected predetermined regions of the camera
image are processed to extract the significant
characteristics (lines A and B) of the banknote (block
14), for example, as described in detail later on with
reference to Figure 3 showing edge extraction by edge
extractor 8.
Image processor 6 then redefines the position of
the banknote with respect to the reference system of
device 1 (used for the specimen banknote) using the
position of the extracted significant characteristics
and the reference position of the same significant
characteristics on the specimen banknote (block 15). For
example, image processor 6 may determine, in known
manner, the horizontal and vertical deviation of the
extracted predetermined significant characteristics with
respect to the same significant characteristics on the
specimen banknote, and calculate correct coordinates of
the banknote on the basis of the deviation, or may use
known rotation-translation algorithms.
The repositioned image of the banknote is then
supplied to edge extractor 8, which processes the image
by filtering it through an edge detection convolution
filter (block 16), e.g. a 3x3 kernel filter as shown in
Figure 3 and described in detail later on.
The processed image, by now only containing the
edges and pixels with brightness gradients with respect
to the adjacent regions, is then sent by extractor 8 to
local averaging and comparing unit 9, which processes
the image to add the values of pixels in predetermined
regions. More specifically, unit 9 divides the banknote
into a number of predetermined portions, and adds the
shade values of the pixels in each portion to obtain a
number of values, one for each portion and each
proportional to the mean shade value of that portion
(block 17). These values are then compared with
corresponding shade values of the specimen banknote
(processed beforehand in the same way as for the
banknote being controlled) to determine the deviation
(block 18); and the detected deviations as a whole are
processed according to predetermined criteria governing
acceptance or rejection of the banknote (block 19). For
example, the banknote may be passed if all the detected
deviations are below a predetermined threshold, or if a
significant portion (e.g. 90%) of the deviations is
below a first threshold, and the rest are anyway below a
second higher threshold.
If the banknote is passed (YES output of block
19), a pass signal is generated (block 20); conversely
(NO output), a reject signal is generated (block 21);
which signals may be used for printing the serial
numbers (which, as stated, are only printed on the
passed banknotes) and for separating the passed
banknotes from the rejects when sheet 3 is cut.
Figure 3 shows a diagram of the convolution filter
for extracting the edges in block 14 of Figure 2. More
specifically, the filter, which is substantially known
and indicated as a whole by 22, comprises two FIFO
registers 23, 24; a multiplication matrix 25 with nine
cells 26-34 arranged in three rows and three columns;
four adders 35-38; an input line 39; and an output line
40. Input line 39 is connected to the input of cell 26
and to the input of register 23; the output of register
23 is connected to the input of register 24 and to the
input of cell 29; the output of register 24 is connected
to the input of cell 32; each cell 26, 29, 32 has two
outputs, a first connected to the input of adder 35, and
a second connected to the cell on the right (27, 30, 33
respectively); each cell 27, 30, 33 has two outputs, a
first connected to the input of adder 36, and a second
connected to the cell on the right (28, 31, 34
respectively); each cell 28, 31, 34 has an output
connected to the input of adder 37; and the outputs of
adders 35, 36, 37 are connected to the inputs of adder
38, the output of which is connected to output line 40.
Cells 26-34 of filter 22 provide for multiplying
the input pixel value by a predetermined value (8 for
cell 30 and -1 for cells 26-29 and 31-34) and for
supplying it to the respective adders; and cells 26-31
supply the value of the same pixel (unchanged) to the
next cell in the same row. For each clock count,
therefore, a new-pixel value is supplied to cell 26 and
register 23; the "oldest" pixel in register 23 is
supplied to cell 29 and register 24; the "oldest" pixel
in register 24 is supplied to cell 32; cells 26, 29, 32
supply adder 35 with the result of multiplying the pixel
received in the previous clock count, and supply the
same pixel received previously (unchanged) to respective
next cells 27, 30, 33; similarly, cells 27, 30, 33
supply the multiplication result to adder 36, and the
unchanged pixel value to next cells 28, 31, 34; cells
28, 31, 34 simply supply the multiplication result to
adder 37; adders 35, 36, 37 supply the sum of the
previous values to adder 38; and adder 38 supplies the
total value to the output.
In the example shown, assuming the banknotes are
scanned in columns, and that each register 23, 24
memorizes a number of pixels equal to that of one column
(e.g. 128), cells 26-34 receive, at each clock count,
the values of a central pixel and the eight surrounding
pixels, and multiply them by the coefficients indicated;
and the multiplication results are then added so that
each pixel in the image is modified according to the
value of the eight adjacent pixels. Consequently, in the
case of uniform portions (same pixel values), each pixel
is assigned a zero or, at any rate, a low value, whereas
the pixels in edge or high-contrast portions are
assigned high values, thus transforming the original
image into a processed image containing practically only
edges, and in which the value assigned to the pixels
belonging to the edges indicates the degree of contrast
or gradient with the adjacent pixels.
An example embodiment of local averaging and
comparing unit 9 will now be described with reference to
Figure 4.
Unit 9 comprises a selecting element 41 having an
input 42 serially supplying the pixel values processed
by image processor 6, and a number of outputs 43, each
connected to a respective local section 44. Each local
section 44 comprises an adder 45 having a first input 46
connected to a respective output 43 of selecting element
41, a second input 47, and an output 48 connected to an
accumulator 49 having two control inputs 50, 51
respectively receiving an enabling signal EN and a reset
signal RES. Accumulator 49 also has an output 52
connected to input 47 of adder 45 and to a memorizing
element or latch 53 having an enabling input 54
receiving a respective control signal L, and an output
55 connected to a first input of a comparator 56, which
also has a second input connected to a reference buffer
57 for memorizing a local reference value and which is
enabled by a control signal B. Comparator 56 also has an
input receiving a control signal C, and an output 58
connected, like all the outputs of local sections 44, to
a logic unit 59, which, depending on the outcome of the
comparisons made in all the local sections 44,
determines acceptance or rejection of banknote 2 in
block 19 of Figure 2. In the example shown, control
signals EN, RES, L, B and C are supplied by control unit
11 in Figure 1.
In local averaging and comparing unit 9, the
pixels - scanned, for example, in columns - are supplied
to selecting element 41, which distributes them, in
predetermined groups, to sections 44. For example, if
local averaging and comparison are performed in 8x8
pixel regions, selecting element 41 sends the first 8
pixels in the first column to the first section 44, the
next 8 pixels in the first column to the second section
44, and so on up to the end of the column, and then
sends the first 8 pixels of the second column to first
section 44, the second 8 pixels in the second column to
second section 44, and so on. In each section 44, the
value of each pixel received at input 46 is added by
adder 45 to the previous total supplied at input 47, and
the sum is memorized in accumulator 49 enabled, at this
step, by signal EN. The sum is repeated for all the
received pixels of eight successive columns, and the
total is memorized in latch 53 enabled for the purpose
by signal L; accumulator 49 is reset by signal RES to
memorize the sum of the next region from zero; and
reference buffer 57 supplies a reference value REF
(corresponding to the sum of the specimen banknote pixel
values in the same region, the image of which has been
processed in the same way as described above to extract
the edges) to comparator 56, which, as stated, supplies
the local comparison value defining a local-error signal
E for use by logic unit 59.
Selecting element 41 may operate in different ways
to add the pixel values (local averaging) depending on
the control precision required and the characteristics
of the banknotes being controlled. For example, as
opposed to 8x8 pixel regions as described above, local
averaging and comparison may be performed in 16x16 pixel
regions. Alternatively, the banknote may be divided into
predetermined areas, even differing in size, so as to
contain whole copperplate details; in which case, the
banknote may be repositioned according to displacement
of the offset printing, and the areas for comparison
must be large enough for each to definitely contain the
respective copperplate detail, taking into account any
displacement of the copperplate with respect to the
offset printing. Alternatively, repositioning may be
performed according to displacement of the copperplate
printing, and the areas for comparison may be the same
size as the copperplate details, and therefore smaller
than previously.
In another solution, the local averaging and
comparison regions may be of predetermined size (e.g.
8x8) in the portions containing only offset or
copperplate printing, and of larger size in the combined
regions.
Finally, it should be pointed out that, as opposed
to being performed prior to extraction, the banknote
repositioning step may be performed after the edge
extracting step, using already extracted edge portions;
and convolution to extract the edges may be performed
using a software or hardware filter.