Field Of The Invention
This invention concerns an image processing device and
an image input device, which are equipped with a data synthesizing unit. More specifically,
it concerns a technique to effectively prevent the output of images which may not
Background Of The Invention
A number of image processing systems to prevent the illegal
duplication of banknotes and stock certificates have been proposed and implemented
in the prior art, see for example documents
EP 0 535 891A1
EP 0 506 469A2
mentioned below. These approaches involve installing a device in a system
which is closed from the input of the original image into the color copier until
the reproduced image is generated. In this sort of closed system, the image input
unit reads the original (a banknote or other document which may not legally be copied)
and uses the data to verify that it may not be copied. Based on this result, output
is prohibited in the image generating device.
In recent years, as image scanners, personal computers
and printers have become more sophisticated and cheaper, illegal copying in open
systems has become more of a social problem. An image scanner can be used to read
a banknote, and the image data which it reads can be stored in a personal computer.
The data can be sent from the computer to a color printer, which can then generate
and print out the image. In such a case, the image data stored in the personal computer
may have been sent directly in the form of signals from a scanner, or they may have
been acquired from a recording medium such as a floppy disk or a magneto-optical
disk. In this sort of open system, then, images are often generated from data of
unknown origin. In these cases, too, some means must be provided to prevent the
output of prohibited images.
The problem which arises in such cases is that the image
data which arrive at the printer's image generator are not necessarily identical
to those read from the original image. At the point when a decision must be made
in the image generator as to whether the image it is about to output is or is not
one which may not be copied, key features of the image may have already been lost.
In particular, in order to discourage tampering, the characteristic
mark printed on money or stock certificates to show that they may not be copied
is generally made so small that it is nearly imperceptible to the eye. If the computer
processes the image to modify it slightly, there is a high probability that some
of the data will be lost.
Some have proposed that the image be verified as legally
reproducible in the image scanner, so that the image will not even be read if it
may not be copied. However, many feel that prohibiting even the reading of the image
is excessive. It is extremely difficult to come up with an absolute way to prohibit
all illegal copying.
EP 0 535 891 A1
discloses an image processing apparatus (e.g. for a copying machine) which
has an image recognition unit or discrimination function to recognize selected patterns
in originals (e.g. bank notes and securities). If such a pattern in the original
was recognized, then a specific image pattern or pattern data are added to the image
data of the original data and merged data of image data of the original and inputted
pattern data are stored. Stored combined data then may be used to print a copy of
the original including the inputted pattern data in case of recognition of a target
pattern (the latter also called predetermined pattern). The inputted pattern data
are identifying the copying machine such that the counterfeiting user may be traced
The same way document
EP 0 506 469 A2
proposes an image processing apparatus which detects the presence of a
specific original and, depending on the degree of similarity between a target pattern
and the image to be analysed, the image data are changed such that the copying machine
or the user are identified in the case that the target pattern is detected. When
the target pattern is recognized, a copying machine identifying code is added to
the image data such that the faking may be traced back to the source.
It is an object of the invention to provide an image processing
device and an image processing method which improve the detectability of the presence
of a predetermined pattern in image data.
The invention is defined in claims 1 and 6, respectively.
Particular embodiments are set out in the dependent claims.
To accomplish the objectives outlined above, in particular,
the image processing device and image input device of this invention are designed
as follows. They have a unit to recognize a target pattern in the image data (the
target pattern will be recognized by an image recognition unit in the embodiment),
and a data synthesizing unit to synthesize the clear target pattern on the recognized
target pattern which is not fully clear when the recognition unit has detected a
pattern similar to the target pattern (which corresponds to a data synthesizing
unit in the embodiment). The entire device is realized as image processing device
in the embodiment.
When the recognition unit has found a pattern which resembles
the target pattern, a set of feature data is synthesized (i.e., inserted) into the
image data which were received. Thus the image data which pass through the image
processing device can later be accurately recognized as the detected image based
on the synthesized feature data. Then if, for example, the image data are subsequently
transmitted into an open system, it will be difficult to eliminate the target pattern
through image processing. Ultimately, when an image generator produces and attempts
to print out the said image data, they will be accurately recognized as the detected
image, and the required processing will be executed to prevent output.
The feature data is an enhanced version of the target pattern.
In this case, a pattern of the same sort as the target pattern printed on the bill
which is the object of detection, but enhanced, is written into the image data.
This makes it more difficult to eliminate the target pattern through subsequent
image processing. Ultimately, when the image generator attempts to print out the
image, the enhanced target pattern will be recognized as the object to be detected.
Even if an enhanced pattern is synthesized and imposed
on top of the original target pattern, it need not be conspicuous or intrusive.
Moreover, since any recognition processing executed by a latter stage device such
as an image generator would be able to use basically the same algorithm as the recognition
unit in the image processing device, this method has an obvious advantage. A different
recognition algorithm could, of course, also be used.
For illustrative purposes it should be noted that an appropriate
alternative feature data would be digital data such as an electronic watermark.
If an electronic watermark is synthesized in the image data, the subsequent recognition
processing could consist of detecting the said watermark. Such a watermark could
be detected with greater accuracy than other types of feature data. Moreover, if
invisible data such as an electronic watermark are written in, the presence of the
mark would not be detectable during normal use.
The image input device of this invention is an image processing
device defined in Claim 1. The image input device has the recognition unit execute
the recognition processing on the image data which it acquires. The image data synthesized
by the synthesizing device become the output data of the image input device.
If this method is used, when a document (a banknote, say)
is submitted to the image input device to be read, the read-in processing per se
will be executed. When the data are read in, new image data consisting of synthesized
(i.e., inserted) feature data are generated and output. These synthesized feature
data guarantee that when the user eventually attempts to print out the image, the
mark will be detected accurately, and the specified processing will be executed
to prohibit output.
An "image input device" according to this invention includes
not only a scanner with an image sensor as discussed in the embodiment, but also
transmission and signal communication devices which exchange image data with other
Brief Description Of The Drawings
Detailed Description Of The Invention
- Figure 1 shows the overall configuration of the image processing device according
to this invention.
- Figure 2 shows an example of marks and a pattern which might be used in this
- Figure 3 illustrates the principle by which the marks used in this invention
- Figure 4 shows the overall configuration of a preferred embodiment of the image
processing device according to this invention.
- Figure 5 shows the internal configuration of the mark shape extraction unit.
- Figure 6 shows the internal configuration of the color extraction unit.
- Figure 7 shows an example of a pattern which might be stored in first memory
- Figure 8 shows the window used in the pattern location detection unit.
- Figure 9 shows the internal configuration of the pattern location detection
- Figure 10 shows an example of a window which might be used in the pattern location
- Figure 11 shows an example of a decoder circuit which might be used in the pattern
location detection unit.
- Figure 12 illustrates how the data are stored in second memory unit 16.
- Figure 13 shows the internal configuration of the pattern location matching
- Figure 14 illustrates the concept used in the window element of the pattern
location matching unit.
- Figure 15 shows the actual hardware constituting the window element.
- Figure 16 illustrates the principle underlying the organization of the output
of the window element.
- Figure 17 shows an example of the data latching circuit used to compare the
marks on the outer squares. This circuit is a component of the pattern location
- Figure 18 shows an example of the data latching circuit used to compare the
marks on the inner squares. This circuit is a component of the pattern location
- Figure 19 shows examples of outputs of the pattern window.
- Figure 20 shows data and addresses stored in dictionary storage unit, the memory
for the outer square.
- Figure 21 (a) shows an example of the detected pattern.
- Figure 21 (b) shows an enhanced version of the detected pattern.
- Figure 22 (a) shows an example of the detected mark.
- Figure 22 (b) shows an enhanced version of the detected mark.
Figure 1 shows the basic configuration of an image processing
system in which the present invention has been implemented. As can be seen in the
figure, the document which is to be read is imaged by scanner 10 and taken in as
image data. The image data which are taken in are transmitted to personal computer
20 and stored in either computer 20's internal memory or in an external storage
device. At the time the data are taken in or after they are read out of the storage
device, an application installed in personal computer 20 which can process image
data executes the specified image processing or modification processing. The image
data are then output to an image generating device such as printer 30, which prints
the image on paper.
This configuration is well known in the prior art. Scanner
10 and personal computer 20 may be connected directly via a cable, or data may be
transmitted between them via a communication circuit. The image data read by scanner
10 may be stored in personal computer 20 via a storage medium such as a floppy disk
or a magneto-optical disk.
Printer 30, which must prevent the final output of the
copy of the banknote or other item which may not be reproduced (i.e., may not be
output) consists of printer 31, which performs the function of a printer, and image
recognition unit 32. Recognition unit 32 executes the specified processing on the
image data it receives from personal computer 20. It obtains the goodness of fit,
which indicates how similar the found pattern is to the one which is printed on
non-reproducible items, and it sends this goodness of fit to printer 31.
If the goodness of fit which it receives is above a given
threshold, printer 31 determines that the image data it has been asked to output
represent a non-reproducible item, and it executes the specified processing to prohibit
output. This processing may result in printing the entire page a given color, printing
the word "warning" in overlapping rows across the page, or cutting off the power
supply so that the output processing itself is prohibited.
Alternatively, instead of outputting the goodness of fit,
as in the previous example, first image recognition unit 32 can take the process
all the way through to determining whether output should be prohibited based on
the goodness of fit. If it determines that the image in question may not be output,
it outputs a "prohibit output" command. A third alternative would be that the image
recognition unit 32 and printer 31 be related in the same way as the recognition
unit and copying unit in a prior art copy machine.
With this invention, in addition to basic scanner unit
11, which fulfills the same function as a prior art scanner, scanner 10, the image
input device, has the image processing device 12 of the first embodiment as the
image processing proposed by this invention. Image processing device 12 has a second
image recognition unit 13, which executes the processing to detect a target pattern
in the image data, and a data synthesizing unit 14, which synthesizes data to enhance
the target pattern in the image data captured by scanner unit 11 when image recognition
unit 13 has detected a pattern resembling the target pattern.
With this method, if a target pattern is detected as the
image is being input or read in, data entry is allowed to continue, but data are
synthesized to enhance the target pattern. This insures that the pattern will not
be eliminated by any subsequent image processing. When a printer is eventually asked
to output the image, first image recognition unit 32 will be certain to recognize
the target pattern and prohibit output because of the enhanced target pattern. The
pattern may be enhanced by increasing the density of the corresponding pixels or
synthesizing a new perfect target pattern with no flaws.
Before explaining the detailed processing in data synthesizing
unit 14, we will now explain the detailed method to detect and recognized the target
In this embodiment, a target pattern 1 such as that shown
in Figure 2 is provided in a given location on the document to be detected, which
is something that may not legally be copied or output. A determination is made as
to whether target pattern 1 can be found in image data which is under image processing
by a printer and other image creating devices. If it can be found, a signal to prohibit
outputting the image data of the document is output according to this invention.
Target pattern 1 is composed of a number of marks 2 which
are placed in appropriate locations. In this embodiment, marks 2 are small triangles
which form the shape of a frame with nothing in the center. These triangles are
of a specified color. The principle underlying the arrangement of marks 2 in this
example is shown in Figure 3. The marks are placed along the sides of two squares.
In the example shown, there is a single mark 2 in a specified location on L1,
the line defining the inner square, and ten marks 2 in specified locations on L2,
the line defining the outer square. When marks 2 are placed along the sides of squares
(or rectangles) in this way, pattern 1 can be printed so that the sides of the square
are parallel to the horizontal and vertical axes of the image. The marks will then
correspond to the scanning path when the image is read, which will simplify the
detection hardware by allowing a smaller window to be used. When pattern 1 is actually
printed, of course, lines L1 and L2 will not be there, and
only the marks 2 shown in Figure 2 will be visible. This will make it difficult
for an uninformed person to figure out the principle of the arrangement and so make
it more difficult to tamper with the pattern.
According to this invention, all of marks 2 are formed
using a given color, as discussed above. The marks may be of identical or similar
colors. The color is one with a high density (numerical value of the color signal
is large) in one of the three color signals expressing the color data in an RGB
or other format. If the given color is yellow, the signal value of one signal is
large even when the color signals are expressed by RGB, Lab, or YMC. In other word,
B is larger than others in case of RGB, b in case of Lab, and Y in case of YMC.
This color choice of yellow makes it easier to detect the shape of the target pattern,
and the yellow target pattern is not easy to detect by the human naked eyes.
We will now explain how to recognize or detect the target
pattern according to the embodiment of this invention. Target pattern 1 is composed
of a number of marks 2. The first step is to detect these marks 2 in the image and
determine their locations. The color of marks 2 is selected to the specific color
which gives the large contrast, or different in a color component signal. The density
of one of the three color component signals constituting the marks, therefore, will
be higher than that of the other two. By executing threshold processing in the image
signals with respect to this color component signal and binarizing the image data,
we can extract marks 2 and recognize the pattern.
In this embodiment, we can achieve greater accuracy of
detecting the marks by extracting the pixels constituting marks 2. In addition to
pattern recognition based on a single color component signal, we can determine whether
each pixel in the mark has a given color component density. When a single color
component signal is subjected to threshold processing independently of the other
signals, it is possible for wrong color (i.e., a color different from the color
of the marks) with a density above the threshold to be extracted as well. For this
reason, the image data obtained are filtered with respect to the density of the
three color component signals. By extracting pixels whose density is within a given
range, we can extract only pixels of the desired color.
We can execute pattern shape recognition using a single
color component signal and then determine whether the pattern is the specified color.
By finding the logical sum (AND) of these two processes, we can accurately extract
just the pixels constituting the marks. This concludes the detection process of
We next obtain the locations of the marks we have extracted.
By determining whether they are in a given spatial relationship with each other,
we can find how closely they resemble the specified target pattern (i.e., their
goodness of fit). Since the marks 2 which constitute pattern 1 in the example shown
in Figure 2 are placed on the outlines of two squares, two windows can be set up
which correspond to these squares. We could then determine whether, for example,
the marks 2 found in these windows while scanning in the raster direction are in
a given spatial relationship with each other.
We shall next discuss an embodiment of the image processing
device of this invention which is ideally suited for implementing the processing
method described above. Figure 4 shows the overall configuration of such an image
processing device. In this example, the image processing device according to this
invention is installed in a image input device such as a scanner, which allows to
receive the prohibited image to copy (such as banknote) but does not allow to output
the received image as it is, In other words, the image processing device according
to this invention will modify the received image so that it can not be printed out.
As can be seen in Figure 4, the image data read by the
image sensor in the copy machine are transmitted to image processing device 12 (a
buffer IC) in image input device 10. As these image data are scanned by an image
sensor such as a CCD, they are transmitted successively in real time, region by
region, to the image processing device. The actual data which are sent are 8-bit
color data for each of the red (R), green (G) and blue (B) components.
The RGB color signals pass through image input unit 42
and are transmitted to binary processing unit 43. The binarized image data (i.e.,
a binary image) are stored in first memory unit 44. The binary image stored in unit
44 is sent to mark location detection unit 45. The marks 2 constituting target pattern
1 are extracted from the binary image, their locations are specified, and they are
stored in second memory unit 46. Everything before unit 46 constitutes the mark
detecting unit to detect the marks.
The data representing the locations of marks 2 which are
stored in second memory unit 46 are transmitted to pattern location matching unit
47. A window of a specified shape and size is used to determine how well the locations
of the marks match a specified arrangement (i.e., the arrangement constituting pattern
1 (see Figure 2). The result of matching level of the marks is transmitted to goodness
output unit 48. Goodness output unit 48 will calculate the goodness of fit of the
received pattern based on the number of the pixels which are matched with the predetermined
arrangement, and the goodness of fit will be forwarded to data synthesizing unit
Data synthesizing unit 14 judges if the target pattern
should be synthesized or not based on the goodness of fit outputted from goodness
of fit unit 48. If yes, the target pattern 1 is synthesized on the received pattern
stored in third memory unit 49 using the pattern location data outputted from pattern
location matching unit 47, the angle of rotation, and mark location data outputted
from mark location detecting unit 45 (stored in second memory unit 46).
Third memory unit 49 is used to hold the received pattern
temporary during the time delay while the image data are processed in first memory
unit 44, and second memory unit 44. The received pattern is directly from the image
input unit 42. In Figure 1, the outputted image from scanner 11 is directly inputted
to data synthesizing unit 14 for the purpose of showing in the drawing and the explanation
of the invention. It is , however, necessary in an actual configuration that the
received image (the outputted image from the scanner 11) is stored in third memory
unit 49 for synchronizing with marks location data outputted from to second memory
unit 46, and pattern location data outputted from pattern location matching unit
47, all of which are inputted to data synthesizing unit 14. It is, of course, possible
to input the received image directly to data synthesizing unit 14.
As mentioned above, since the target pattern is preferably
yellow, it is hardly recognized by the human naked eyes. It gives, therefore, only
little influence to the synthesized pattern even after the synthesizing the target
patter on the received pattern. It is not synthesizing the target pattern on other
pattern, but on the pattern which was originally identical to the target pattern.
Because of this reason, the synthesizing the target pattern makes further little
changes on the received pattern which no body can recognize the synthesizing.
The example shown is the configuration when the target
pattern is yellow. It is, however, no problem if the target pattern is different
color from yellow if one of the color components forming the color has a distinct
difference from other color components. It is because it is easy to detect the marks
using the one color component. The selected color can be the color which is easy
to recognize, because the target pattern is synthesized on the original pattern,
which does not give any effects to the human naked eyes.
Because of the synthesizing, the next image processing
device and the image generating device which are connected to the image processing
device according to this invention can recognized the synthesized target pattern
easily, and the pattern recognition can be performed without fails.
We shall next discuss the detailed function and configuration
of each unit of the image processing device according to this invention. Binary
processing unit 43 comprises mark shape extraction unit 43a, mark color extraction
unit 43b and AND element 43c, which finds the logical sum (AND circuit) of the outputs
of units 43a and 43b. A single color component signal with a high density, whichever
of the RGB signals is best suited to extracting the mark, is sent to mark shape
extraction unit 43a. The other color component signals, or, alternatively, all the
color component signals, are sent to mark color extraction unit 43b. The marks we
detected in this example are yellow, so the B signals are sent to mark shape extraction
Mark shape extraction unit 43a, shown in Figure 5, executes
threshold processing. It has a comparator with eight bits to match the eight bits
constituting the color component signals. The B signal received from image input
unit 42 is input into input terminal a; a fixed threshold value is input into input
terminal b. When a is less than b, the output will be "1". (The higher the density,
the lower the numerical value.)
In this embodiment, the threshold value is fixed. The use
of a floating binary circuit would allow the threshold value to be changed according
to the density of the document. Such a configuration would yield a more accurate
pattern. In mark shape extraction unit 43a, 8 bits type is used for the high accuracy,
but a fewer bits type can be used as long as the accuracy can be maintained. In
this example, B color component corresponding to yellow is used, but other color
component can be also used according to the selection of the mark color.
Processing each RGB signal, mark color extraction unit
43b, shown in Figure 6, extracts any pixel whose density falls within a fixed range
(between an upper and a lower limit value). Unit 13b consists of three 4-bit window
comparators which input RGB signals one by one and compare them with upper and lower
limit values (when b > a > c, the output will be "1") and a three-output AND
element which finds the logical product of the outputs of the three comparators.
The upper and lower limit values used by the comparators have a specified margin
with respect to the RGB signal expressing the color of the marks 2 which are to
be detected. When a pixel is found for which each of the R, G and B components has
a density within the fixed range, the output of the AND element will go to "1",
and the pixel will be extracted as a prospective constituent of a mark.
Mark color extraction unit 43b uses a 4-bit window comparator
because printed materials in general suffer from extreme non-uniformity (variation)
of color. Higher color resolution would be pointless. However, as pattern recognition
requires accuracy, mark shape extraction unit 43a uses an 8-bit comparator, as mentioned
above. Separating pattern recognition from color extraction in this way allows us
to extract the mark accurately and absorb errors due to variation among the colors
constituting the mark so that the mark can be specified correctly.
A binary processing unit 43 configured as described above
outputs a binary image in which pixels which are the same color as mark 2 are expressed
as "1" and pixels of other colors are expressed as "0". This image is stored in
first memory unit 44, which in this embodiment consists of a given number of line
buffers. Mark location detection unit 45 can use the binary image to detect marks
2 and extract their locations. Once the locations of the marks are known, there
is no need to store all the image data representing the document. Only a few dozen
lines of buffer space are needed to store the data needed for subsequent processing.
An example of a binary image which would be stored in first
memory unit 44 (the line buffer) is shown in Figure 7. Each square in the drawing
represents one pixel. In the example we have been using, marks 2 are triangles.
In order to make them distinguishable from dirt smudges, we have left them white
in the center. The mark shown in Figure 7 is a minimal triangle. In this embodiment,
the marks have been made as small as possible to enable the hardware needed to detect
them to be made smaller as well as to make it more difficult for an uninformed person
to find them. The mark of pixels which constitutes the actual mark is that shown
in Figure 7.
Mark location detection unit 45 reads out the data stored
in first memory unit 44 and detects the marks of pixels which constitute triangles
like the one shown in Figure 7. In this example, it uses a window of 5 x 4 pixels
(Q00 through Q34) like that shown in Figure 8. When the binary
data for pixels Q02, Q11, Q13 and Q20
through Q24 are "1" and those for the other pixels are "0", the HIT output
goes to "1", and the location data are stored in second memory unit 46.
Since the mark 2 is rotated in an actual case, a plurality
of mask data shown in Figure 8 are prepared along with the rotation angle. The location
data and the rotation angle data are stored in second memory unit 46. If the mark
is a circle, the mask data and the rotation angle data are not necessary. In Figure
8, the mask has 4 x 5 size, but the 5 x 5 size can be used for detecting the detailed
location of the top portion of the triangle .
The result of this detection (mark has been found: 1; no
mark has been found: 0) is stored in second memory unit 46. In this embodiment,
a location data of the central white portion within the triangle will be used to
represent the window. If necessary, shrunk window such as 2 x 2, or 4 x 4 can be
used which is a gradated data. This gradated data does not seriously affect to detect
the target pattern at pattern location matching unit 47, because the received pattern
1 is much larger than the individual mark 2, so gradated mark data will be sufficient
for the relatively large received pattern.
Mark location detection unit 45, which executes the processing
discussed above, is shown in Figure 9. Unit 45 comprises window unit 45a, which
measures 5 x 4 pixels, and decoder circuit 45b, which receives the pixel data (1/0)
detected by unit 45a and determines whether they are in a given arrangement.
More specifically, as is shown in Figure 10, each Qij
in the window consists of a flip-flop. The window has four rows in each of which
five flip-flops are connected in series. A row of flip-flops corresponds to the
pixel data (1/0) stored on a single line in the line buffer which serves as first
memory unit 44. The data are input into the flip-flops, starting with the head flip-flop,
in order from the first pixel. Each flip-flop simultaneously receives a clock (CLK)
signal. Once synchronized, it transmits the data to the next flip-flop.
When clock signals are input one by one, the input of data
to the flip-flops assumes the same form as scanning pixel by pixel in the scanning
direction. When the final pixel data for a given line (1/0) have been input, we
begin to input data to the head flip-flop of the next line down. This is identical
to moving down one pixel in the feed direction. Consequently, the output of the
flip-flops when five pixels' worth of data have been input will be as shown in the
drawing and will be equivalent to the window pictured in Figure 8. The output MWQij
(i = 0 through 3, j = 0 through 4) of each flip-flop is sent to decoder circuit
Decoder circuit 45b, which is pictured in Figure 11, comprises
a 20-input AND element. When a mark is input, the input terminals constituting white
pixels have their input reversed. Thus when black pixels "1" are lined up in a given
shape, all the inputs of the AND element which constitutes the decoder circuit go
to "1", and the HIT output of the decoder circuit goes to "1". If the value of a
single pixel is different from the that of the others, one of the inputs to the
AND circuit will be "0", and the HIT output of the decoder circuit will go to "0".
Based on the output of decoder circuit 45b, data representing the locations of all
the valid marks 2 are stored in second memory unit 46.
If we are to map out the internal data stored in second
memory unit 46 when the pattern has been detected, they would be arranged as shown
in Figure 12. Each square in the drawing corresponds to a 5 x 4 pixel chunk of the
image data captured when the document is imaged.
As stated above, mark location detection unit 45 outputs
a compressed binary image in which portions where marks 2 are found go to "1" and
pixels of other colors are expressed as "0" This image is stored in second memory
unit 46, which in this embodiment also comprises a given number of line buffers.
Since the target pattern is detected based on this compressed binary image, there
is no need to store all the image data representing the document. Only a few dozen
lines of buffer space are needed to store the data needed for subsequent processing.
Pattern location matching unit 47, which detects target
pattern 1, is configured as shown in Figure 13. It has a window 47a of a given size,
which it uses to scan the mark data (1/0) stored in second memory unit 46.
In this example, target pattern 1 comprises a number of
marks 2 placed along the outlines of two squares, as shown in Figures 1 and 2. For
this reason it is desirable to extract simultaneously all the "1"s stored in second
memory unit 46, which can be seen in Figure 12.
Leaving a margin one chunk wide around the outer square
on which the marks are placed, we use a window of 10 x 10 chunks, as shown in Figure
14. The hatched portions of the drawing (inner square W1 and outer square
W2) are the regions where data will actually be collected.
The actual configuration of window unit 17a is shown in
Figure 15. Ten rows of flip-flops with ten serial inputs and parallel outputs are
provided. Data in a given line buffer of second memory unit 46 are input into the
corresponding row of flip-flops in order starting with the head flip-flop. The operation
is identical to that of the aforementioned 5 x 4 pixel window unit 45a in mark location
detection unit 45. Data synchronized by a clock signal CLK are transmitted in order.
The output of the flip-flops at any given moment, as shown in Figure 16, functions
as a 10 x 10 unit window.
The units in the output which correspond to the hatched
portions of Figure 14 are AWQ33 through 36, 43, 46, 53, 56 and 63 through 66 (inner
square W1) and AWQ11 through 18, 21, 28, 31, 38, 41, 48, 51, 58, 61,
68, 71, 78 and 81 through 88 (outer square W2). These data can be used
to detect target pattern 1.
The data stored in the units of outer square W2
(the outputs of the specified flip-flops stated above) are sent to latching circuit
47b; those stored in the units of inner square W1 (the outputs of the
specified flip-flops stated above) are sent to latching circuit 47c. These circuits
latch to conform to the working of window unit 47a.
Latching circuit 47b comprises 28 D latches (flip-flop
groups with parallel input and parallel output), as shown in Figure 17. These latches
maintain the input data as they are when the latch signal is received. In this example,
AWQ15 is the head of the unit outputs constituting the window in Figure 16. The
units which correspond to the location of outer square W2 are arranged
in order in a clockwise circuit. Accordingly, as is shown in Figure 17, AWQ15 is
input to the head D latch and AWQ14 is input to the final D latch.
Similarly, latching circuit 47c comprises 12 D latches
as shown in Figure 18. AWQ35 is the head of the unit outputs constituting the window
in Figure 16. The units which correspond to the location of inner square W1
are arranged in order in a clockwise circuit. AWQ34 is input to the final D latch.
As is shown in Figure 13, the output of latching circuit
47c is sent to address decoder circuit 47d. The 12-bit data latched in circuit 47c
are used as address data to access the corresponding portion of dictionary storage
unit 47e. Dictionary data for 0 deg., 90 deg., 180 deg. or 270 deg. on the outer
square are output to comparison circuit 47f. The 28-bit data latched in latching
circuit 47b are sent to comparison circuit 47f, where the data are compared bit
by bit. The result, match or no match, is transmitted to counter circuit 47g.
Generally when a document is copied, it must be placed
parallel to the sides of the glass. Although the document may be parallel, it sometimes
happens that it has been placed on the glass with its top and bottom or left and
right sides reversed. Thus a target pattern on a document may be oriented in any
one of four directions, as shown in Figures 19 (A) through (D). When the marks 2
constituting target pattern 1 are stored in a specified area of second memory unit
46, the data latched in latching circuits 47b and 47c by way of latching unit 47a
are arranged in four ways, as shown in Figure 20.
The data arrangement is stipulated in this way, and the
relationship of the data latched by a mark on inner square W1 at a given
angle of rotation to the data latched by a mark on outer square W2 is
one to one. If the marks have been arranged appropriately, the data will be different
at the four angles of rotation shown in Figures 19 and 20. If we find a match while
trying to determine which data arrangement a certain region (in this example, inner
square W1, which has fewer data) corresponds to, there is a high probability
that we have found the target pattern. Since we now know its angle of rotation,
we can use the data latched from outer square W2 as reference values
(i.e., data which are valid if this is the target pattern). We compare these reference
values with the actual data obtained from outer square W2. A high coincidence
indicates that the pattern has been recognized. Pattern location matching unit 47
uses this same principle.
In this embodiment, efficient use is made of the data from
inner square W1 and outer square W2. As discussed above, the
less voluminous data from inner square W1 are used as addresses. The
data strings in the units in Figure 20 are stored at specified addresses in outer
dictionary storage unit 47e.
The data latched by the circuit for the inner square, 47c,
are used as addresses. If there is a match with one of the four addresses shown
in Figure 20, the data stored in outer dictionary storage unit (ROM) 47e which correspond
to this address are read out and compared by comparison circuit 47f. By checking
whether there is an address which corresponds to the mark data on inner square W1,
we determine whether the marks we have found are arranged in target pattern 1. Comparing
the mark data on outer square W2 with reference values which have been
read out also allows us to determine whether the marks we have found are arranged
in target pattern 1.
Comparison circuit 47f is composed of D latch unit D1,
which stores the data collected from the outer square; D latch unit D2,
which stores the dictionary data (i.e., the reference data) it receives from outer
dictionary storage unit 47e for the outer square at specified angles of rotation;
a dual-input coincidence circuit CC, which compares the data stored in latches D1
and D2; and parallel input-series output shift register SR, which receives
each output of coincidence circuit CC and outputs it in the same order. Units D1,
D2, CC and SR each have 28 stages to correspond to the number of data
(28) comprising outer square W2.
When this configuration is used, the data transmitted by
latching circuit 47b and those received from dictionary storage unit 47e (the reference
data) are stored in their respective latching units, D1 and D2.
They are input into coincidence circuit CC in response to a specified timing signal.
If the reference data and the data indicating the actual locations where the marks
are extracted are identical, the output of coincidence circuit CC will go to "1".
If they are not, it will go to "0". The result of this determination is stored in
the corresponding shift register SR. A shift clock is applied to shift register
SR, and the data stored in it are input to counter circuit 47g in order, bit by
Counter circuit 47g counts the number of data which are
"1". If all the data acquired from outer square W2 coincide with the
reference data stored in the dictionary, the 28 bits output by the shift register
will all be "1". If none of them coincide, the 28 bits will all be "0". The higher
the count value, then, the higher the goodness of fit of the data to the target
pattern. In this example, the count value is the goodness of fit of the detected
pattern with respect to the target pattern. When the goodness of fit exceeded a
predetermined threshold value, it would judge the document to be non-reproducible
and output a signal indicating detection of the target pattern.
In this example, four outer dictionary storage units corresponding
to four angles when the document is aligned to the line of scanning stand. More
dictionaries can be prepared for more detailed rotation of the document.
We will now explain the essential part of this invention.
From the goodness of fit it receives from output unit 48, data synthesizing unit
14 determines whether to synthesize another set of data. Based on the position data
from second memory unit 46 and pattern location matching unit 47, it synthesizes
target pattern 1 and marks 2 in the original image from storage unit 20. These synthesized
data represent an enhanced version of the target pattern. Concrete examples are
This synthesis may, for example, can create the marks whose
color saturation has been maximized (to FF). If the threshold value is, say, 80,
and the density of the marks comprising the target pattern in the image which is
read is 85, the marks will be recognized when the image is read, but they can be
eliminated through subsequent processing. In particular, when marks have a density
near the threshold value, as in the example above, they can readily be eliminated.
By increasing the density of the appropriate pixels to
FF, we give them a density value far greater than the threshold. They can no longer
be eliminated through subsequent image processing, and they will be certain to be
detected in the final image recognition processing.
As can be seen in Figure 21 (a), if the pattern which is
recognized is missing a mark where one is supposed to be (the marks drawn in broken
lines indicate places where marks are missing), a perfect pattern with no missing
marks can be synthesized, as is shown in Figure 21 (b).
If the density is to be increased as described above, two
alternatives may be used. One is to enhance only some of the marks which have been
detected. Suppose that a pattern like that shown in Figure 21 (a) is detected. Here
the first option would be to maximize the density of only those pixels comprising
the marks 2 which were detected, i.e., those shown by solid lines, so that the overall
pattern itself would still be defective. The second option would be to fix the target
pattern as shown in Figure 21 (b) by maximizing the density of all the pixels comprising
each of the marks.
In this embodiment, the precise state of each pixel comprising
each mark is not stored with the location of the mark. If it were, and some of the
pixels were missing, as in Figure 22 (a), we would have the option of regenerating
the image as we found it (i.e., maximizing the density of those pixels which had
been detected while ignoring the missing pixels, as in Figure 22 (a)) or restoring
the pattern to its proper condition with no missing pixels, as in Figure 22 (b).
In the embodiment we have been discussing, the target pattern
1 that is in the original image is synthesized; however, the invention is not limited
to this approach only. It would be equally acceptable, for example, to synthesize
data constituting an electronic watermark. If this were to be done, the configuration
would need to be that shown in Figure 4. Data associated with the electronic watermark
would be stored in data synthesizing unit 14. When the goodness of fit received
from goodness of fit output unit 48 exceeded a threshold value, an electronic watermark
based on the stored watermark data could be inserted into the image data which had
If an electronic watermark is to be synthesized, there
is no particular reason that it must be synthesized in the exact location of target
pattern 1 in the image data. In this case, then, third memory unit 49 may be dispensed
with. Since the principle underlying an electronic watermark and the method used
to synthesize it involve technology which is well known in the prior art, we shall
not present a detailed explanation here.
The image processing and image generating devices which
receive electronic data from the image input device do not need the sorts of image
recognition units discussed above, since they need to read only the data which represent
the electronic watermark. Thus this method has the effect that the size of the recognition
unit and the load imposed on it will be smaller.
In all of the embodiments and modified examples discussed
above, the processing unit in the image input device (i.e., the device which reads
the image) has consisted of hardware. It would, of course, be equally acceptable
to realize this unit through software. Alternatively, the image input unit (the
image reader) could be eliminated altogether and its function realized within an
image processing device such as a personal computer.
As has been discussed, each of these embodiments makes
the target data on the original document easier to distinguish when the data are
input into an image generating device (i.e., a printer). This prevents documents
from being copied illegally.
As has been described above, with the image processing
and input devices according to this invention, targeted features are recognized
when the image is read in. These features are enhanced so that they will be more
difficult to eliminate through subsequent image processing. As a result, the features
will be easier to recognize when the image is to be generated by an image generating
In particular, if the device is configured as described
in Claim 2, an enhanced version of the target pattern will be superimposed on the
original pattern. The result will be a pattern indistinguishable to the human eye
which, even if seen, will not be noticed. If the device is configured as described
in Claim 3, digital data will be added which can be detected with great accuracy
in subsequent recognition processing. The use of digital data makes it much easier
to determine that the detected pattern is the target pattern.