JPH0562045A - Paper sheet recognition device - Google Patents

Abstract

PURPOSE:To obtain the high-speed and highly reliable device excellent in the prevention of positional deviation and oblique deviation by reading the image of the entire area requiring the recognition of a paper sheet, extracting a certain amount of statistics from the gradation image and determining a binarized threshold. CONSTITUTION:An oblique deviation correction part 3, when it reads the data of a paper sheet 1 with a scanner 2 while carrying the paper sheet 1 in a prescribed direction, removes the relative distortion between the paper sheet 1 and the scanner 2. After the inclination of the read image is corrected by the oblique deviation correction part 3, necessary and minimum gradation image data are segmented by an effective area determination part 5 and the segmented data becomes the object of a processing. The gradation image data are binary coded and the image of its border point is obtained. The obtained border point image is compared with a border point image preliminarily prepared for identification to recognize the type of the paper sheet 1 and whether or not the paper sheet 1 is genuine. With such recognition by means of the border point image, high recognition rate can be obtained. And since check is performed for the segmented absolute minimum data, high speed processing can be obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a paper sheet recognition apparatus used for recognizing types and authenticity of bills, securities, bonds and the like.

[0002]

2. Description of the Related Art In a financial institution, an automatic deposit / withdrawal machine (ATM), an automatic transaction device, etc. are installed in order to automate deposit / saving and transfer processing. In this automatic depositing / dispensing machine, when a customer inserts a bill, the type and authenticity of the bill is recognized. This processing is performed by the bill validator, but here, first, an optical or magnetic pattern is read along one or more lines parallel to the bill transport direction. The optical pattern has certain characteristics for each denomination due to the pattern of the bill. The magnetic pattern also has the same characteristics. The detected pattern, which is an analog signal, is compared with a predetermined threshold value and binarized at many points on the read line. The pulse train thus obtained is counted by the counter circuit, and the count value is compared with the dictionary data.
If the count value is close to any of the dictionary data for all the detected lines, it is possible to recognize the type, authenticity, front / back, direction, etc. of the bill.

[0003]

By the way, in the conventional method as described above, the feature of a partial region along a specific line on a bill is usually extracted and recognized.
Therefore, when a small-area forgery exists outside the detection area, it may be recognized as a genuine note. This occurs when, for example, banknotes are partially pasted together. In addition, in order to increase the reliability of recognition,
There is also a method of detecting the outer size and thickness of a bill and recognizing the type and authenticity of the bill from various angles. However,
In such a method, if even one parameter is out of the standard range, it is determined to be a fake bill. Therefore, in such a case, the rate of determination as a fake bill increases, which is a problem in practical use. Of course, such a problem is not limited to the appraisal of banknotes, and is similarly applicable to the appraisal of securities and other various paper sheets.

In order to solve these problems, first,
It is preferable that the widest possible area of the paper sheet is read at a high resolution and used as a reference for recognition so that counterfeiting or bonding of a small area can be detected. Further, it is preferable to recognize the type and authenticity by using one criterion having the highest reliability in order to improve the recognition rate. But,
If the amount of data to be read is large, it takes a long time to process the data, making it difficult to perform collation at high speed. Therefore, there is a problem in adoption in an automatic transaction device provided in a bank or the like. Further, in the method of reading a part of the paper sheet, if the paper sheet is misaligned in its transportation, the detection data may fluctuate, resulting in erroneous recognition. In order to solve this, a recognition process is desired that is not affected by any positional deviation of paper sheets. The present invention has been made in view of the above points, and an object of the present invention is to provide a paper sheet recognition apparatus that is resistant to positional deviation, skew deviation, and the like, and has high speed and high reliability.

[0005]

A paper sheet recognition apparatus according to the present invention includes a scanner for reading data on a paper sheet, and a skew angle of the paper sheet when the data is read from the transportation state of the paper sheet. The skew correction unit that detects the skew of the read signal of the scanner and writes the skew signal in the buffer memory, and the grayscale image data written in the buffer memory is analyzed to detect the edge of the paper sheet in the buffer memory. A gray level image data of the valid area is read out from the valid area determining unit that extracts the partial position and the central position and determines the valid area of the gray level image data, and from the buffer memory. A binarization processing unit that determines a binarization threshold value and binarizes the grayscale image data that the paper sheet occupies, and a boundary point image in which white / black boundary points are extracted from the binarized image data. Boundary point to get An output unit, the boundary point image, and a comparison unit that compares a plurality of boundary point images that are prepared in advance to create a matching rate distribution, and the paper sheet based on the output of the matching unit. It is characterized in that it is provided with a recognition unit for recognizing the type and authenticity.

[0006]

In this apparatus, the skew correction unit corrects the inclination of the read image, and the effective area determination unit cuts out the minimum necessary grayscale image data for processing. The grayscale image data is binarized, and the boundary point image is obtained. This is compared with a boundary point image prepared in advance for identification, and the type and authenticity of the paper sheet are recognized. According to the recognition by the boundary point image, the recognition rate is high, and the minimum necessary data is cut out and collation is performed, so that high-speed processing is also possible.

[0007]

The present invention will be described in detail below with reference to the embodiments shown in the drawings. FIG. 1 is a block diagram showing an embodiment of a paper sheet recognition apparatus of the present invention. First, the paper sheet 1 to be recognized by the present invention is, for example, a banknote, a securities, a bond, or the like. Optical or magnetic data is recorded on such a paper sheet 1. The apparatus of the present invention is configured to read such data using the scanner 2. In the following embodiments, an example will be described in which the paper sheets are bills and the optical information printed on the bills is read using the scanner 2. In this case, the scanner 2 is composed of, for example, an image line sensor or the like.

The skew feeding correcting unit 3 detects the relative inclination between the sheet 1 and the scanner 2 when the data is read by the scanner 2 while the sheet 1 is conveyed in a predetermined direction. This is the part that removes data distortion. Although the specific configuration will be described later, the skew feeding correction unit 3 includes the buffer memory 4
On the other hand, the skew-corrected data is written. The buffer memory 4 stores the grayscale image data obtained by reading the paper sheets as described above. However, in this case, an image such as a background arranged around the paper sheet 1 is also read at the same time. Therefore, the effective area determination unit 5 is provided to remove these and extract the minimum necessary data.
The operation of the buffer memory 4 will be described later in detail.
The effective grayscale image data stored in the binarization section 2 is binarized by the binarization unit 6. Then, the boundary point extraction unit 7 obtains a boundary point image described later. This boundary point image has certain characteristics depending on the type of banknote, front and back, direction, and the like. Therefore, in the collation unit 9, the prepared reference boundary point image 8 is compared with the boundary point image output by the boundary point extraction unit 7, and the result is output to the recognition unit 10 to determine the type of the paper sheet 1, Authenticity, front and back,
Recognize the direction etc. The above is the schematic configuration and operation of the apparatus of the present invention.

The specific construction and operation of each part of the apparatus of the present invention will be described below in order. FIG. 2 is a block diagram showing a specific embodiment of the scanner 2 and the skew feeding correction unit 3 of the apparatus of the present invention. In FIG. 2, the image line sensor 21 arranged on the conveyance path of the paper sheet 1, the photo sensor 22, the A / D conversion unit 23, the skew feeding angle detection unit 24, the skew feeding correction unit 3, the buffer memory 4, and the position. The detector 27 is shown. The image line sensor 21 is a well-known CC
It is composed of an optical image reading element group such as D. The output is digitized by the A / D conversion unit 23 and input to the skew feeding correction unit 3. Photo sensor 22
Is configured to detect the inclination of the paper sheet 1 on the conveyance path, and the skew angle detection unit 24 calculates, based on the output, how much the paper sheet 1 is inclined with respect to the conveyance path. ing.

FIG. 3 shows an operation explanatory diagram of the skew feeding correcting portion. In the figure, the sheet 1 is conveyed to the image line sensor 21 in the direction indicated by the arrow 22. In this case, the sheet 1 may have a predetermined inclination with respect to the image line sensor 21 as shown in the figure due to variations in the bill insertion direction of the customer. In order to detect this inclination, a pair of photodetector elements, P1 and P2, are arranged in parallel with the image line sensor 21. The photodetectors P1 and P2 are assumed to be arranged, for example, separated by a width Pw.

When the paper sheet 1 is inserted into the image line sensor 21 with a predetermined inclination as shown in FIG. 3, a time difference occurs between the detection signals of the two photodetection elements P1 and P2. .. If the conveyance speed is added to the detection time difference in this case, the conveyance amount difference Pt is obtained. By using these data, the skew angle can be obtained by using the formula S1 shown in FIG. 12 to 18, the arithmetic operation of each part of the device of the present invention, which will be described in order below,
The role and the specific arithmetic expression are used for illustration. The image line sensor 21 reads an optical analog pattern of reflected light or passing light of the paper sheet 1.
This signal is converted into grayscale image data in digital format by the A / D conversion unit 23 shown in FIG. The skew angle detection unit 24 shown in FIG. 2 calculates the skew angle in the manner described above, and inputs the control signal to the skew correction unit 3. As a result, the skew-corrected grayscale image data is stored in the buffer memory 4. In this case, the main scanning address of the read grayscale image data is i0 and the sub-scanning address is j.
If 0, the main scanning address to be written in the buffer memory 4 is i1, and the sub-scanning address is j1, the conversion can be performed by the arithmetic expression S2 shown in FIG.
In this way, the grayscale image data obtained by reading the sheet 1 is stored in the buffer memory 4 shown in FIG.

It is preferable that the scanner 2 has a readable length equal to or larger than a diagonal line of the sheet 1 to be recognized so that the entire sheet 1 can be read. By the way, the grayscale image data thus read into the buffer memory 4 includes the data of the periphery of the paper sheet or the background portion thereof. If an attempt is made to perform subsequent processing including these data, the processing speed cannot be increased. Therefore, first, the position detection unit 27 shown in FIG.
The position of the corresponding grayscale image data in the buffer memory 4 is detected.

FIG. 4 is a diagram for explaining the operation of the position detecting section. As shown in the drawing, the grayscale image data 51 is stored in the buffer memory 4 in the range shown in the drawing. Then, the white portion 52 at the center of the grayscale image data 51 is the grayscale image data area of the paper sheet to be recognized. The position detecting unit 27 shown in FIG. 2 has a structure shown in FIG. 4 when the skew feeding correcting unit 3 shown in FIG. 2 writes the grayscale image data into the buffer memory 4.
Addresses Is, Ie, Js of the end position of the paper sheet shown in
Detect Je. Further, by further calculation, the center position coordinates Ip and Jp of the grayscale image data corresponding to the paper sheet are detected. As this method, the position detecting section obtains the density waveform based on the grayscale image data for the total width Iw in the main scanning direction and the total width Jw in the sub scanning direction of the buffer memory. This waveform is shown on the right side and the lower side of FIG.
Then, the threshold value Mt is set based on these waveforms. This threshold Mt is set to an intermediate value between the maximum density and the minimum density. Then, the initial values of Is, Ie, Js, and Je are respectively Iw as shown in the arithmetic expression S3 of FIG.
, 0, Jw, 0. Then, each time the skew feeding correction unit 3 sequentially writes the grayscale image data in the buffer memory 4, the update formulas of S4 to S7 shown in FIG. 13 are sequentially updated.

For example, the update formula of S4 is such that when the output value M (i, j) of the pixel (i, j) becomes smaller than the threshold value Mt, the main scanning coordinate address i is compared with Is and the main scanning coordinate address i is compared with Is. If the scanning coordinate address i is smaller, the value of the main scanning coordinate address i is assigned to the end position variable Is. In the following, exactly the same corresponding operations are performed for the update formula of Ie, the update formula of Js, and the update formula of Je.

By the above-described processing, the skew feeding correction unit 3
When all the grayscale image data have been written to the buffer memory 4, the upper right and lower left coordinates of the paper sheet are detected as (Is, Js) and (Ie, Je), respectively. After that, the position detector 27 shown in FIG. 2 extracts the center position of the paper sheet. The arithmetic expression in this case is as shown in S8 of FIG. The center position coordinates (Ie, Je) thus obtained correspond to the reading position shift of the paper sheet 1. Therefore, the effective area in the buffer memory 4 is determined on the basis of the center position coordinates.

FIG. 5 is a diagram for explaining the operation of the effective area determining section.
The effective area determination unit 5 shown in FIG. 1 determines the effective area in the buffer memory 4 based on the output result of the position detection unit 27 shown in FIG. In this case, first, the center position coordinates Ic and Jc of the grayscale image data of the paper sheet are used as a reference. And
Assuming that the white area shown in FIG. 5 is an area including the grayscale image data of the actually read paper sheet, an area having a width Bw and a vertical width Bh, which is slightly wider than this, is determined as the effective area. This width is set to, for example, the area occupied by the largest paper sheet actually recognized by the apparatus. For example, when recognizing banknotes, three types of banknotes, 1,000-yen, 5,000-yen, and 10,000-yen, will be recognized. In this case,
Always cut out the area where the largest 10,000 yen note can be recognized.

The grayscale image data thus cut out is
Binarization processing is performed by the binarization unit 6 shown in FIG. Figure 6
A block diagram of the binarization unit is shown in FIG. This circuit is based on the data read from the buffer memory 4, a density distribution creating unit 61 for creating a density distribution, a threshold value determining unit 62, and a binary value for binarizing the grayscale image data. Conversion unit 6
It consists of 3. Here, in the buffer memory 4, FIG.
4, Iw × Jw grayscale image data M arranged on the grid shown in S9 is stored. Concentration distribution creating unit 61
Reads the grayscale image data of the effective area described above from the buffer memory 4 and creates a density distribution series representing the frequency of occurrence of the grayscale value. That is, the density distribution creating unit 61 reads only the grayscale image data M shown in S10 of FIG. In order to binarize the grayscale image data, an appropriate threshold value is determined so as not to be affected by changes in the paper quality, print density, etc. for each sheet, and changes over time in the image line sensor of the scanner. Is desirable. In order to determine the threshold value, in this embodiment, the average value of the first moment of concentration distribution series is calculated, and this is determined as the optimum threshold value MthI. Figure 1 shows the average value of the first moment
4 is calculated by the calculation shown in S11.

FIG. 7 is a diagram for explaining the operation of the binarizing unit. In FIG. 7, the vertical axis represents the concentration distribution series Mhst, and the horizontal axis represents the moment m. In this case, the first-order moment average value MthI is determined as a threshold value, and when the distribution shape of this concentration distribution series Mhst has two peaks, the valley portion of the peak, Mst, is set as the threshold value. You may do it. Further, the threshold value may be determined by obtaining a variance value between the gray value and the number of pixels to be read. Which of these methods is used to determine the threshold value is empirically determined depending on the nature of the paper sheet that is actually the target.

The threshold value thus obtained is first used in the authenticity determination of the paper sheet as the first step. That is, when the counterfeit paper sheet created by copying the paper sheet is read, the threshold value thereof is significantly different from the threshold value of the genuine paper sheet. That is, the variation in the threshold value caused by various variations such as the paper quality and print density of the paper sheet falls within the range of Mst and Msp shown in FIG. Therefore, the formula of S12 shown in FIG. 14 is used for authenticity determination in the first stage of the paper sheet, and if the threshold value is not included in this range, the paper sheet is treated as a counterfeit. Suspend. In addition, it is preferable to appropriately select the wavelength of the irradiation light source by the scanner so that the threshold value falls within a certain range.

The values of Mst and Msp are determined by a threshold value obtained from the density distribution series by performing the above processing on thousands or tens of thousands of real paper sheets. The binary conversion unit 63 shown in FIG. 6 binarizes the grayscale image data based on the threshold value thus obtained. In this case, the read pixel B has the content shown in S13 of FIG. As described above, this is in the effective area calculated and calculated from the center position coordinates of the actually read paper sheet. This range is shown in S14 of FIG. The binarization process is performed according to S15 of FIG. That is, the read grayscale image data is compared with the previously determined threshold value, and if the threshold value is larger than the threshold value, 1 is output, and if it is smaller, the output is 0.

Next, noise is removed from the signal after the binarization process as described above. The noise removing unit 64 is provided on the output side of the binary conversion unit 63 in FIG. This noise removal is a process of removing isolated points and the like existing in the binarized image data. This processing itself has been established by a conventionally known image processing technique, and an example thereof is shown below. First, the pixel (x, y) of the binarized image data B
Whether the output value of the pixel (x, y) is noise is determined from the sum of the output values of the peripheral 8 pixels ηB (x, y) around the pixel (x, y) corresponding to the output value of To judge. This calculation is based on S16 and S17 of FIG. While sequentially scanning the image data in the sub-scanning direction in the main scanning direction, the output value of the pixel ZY of the image data B is 0, and
When the sum of the output values of the peripheral pixel group centered on the pixel (x, y) shown in S18 of 5 is 5 or less, 1 (black) is output.
When it is larger than 5, 0 (white) is output. This determination is as shown in S18 of FIG. Further, when the output value of the pixel (x, y) in the image data B is 1 (black), 1 (black) when the sum of the output values of the peripheral pixel groups is 3 or more according to the formula shown in S19 of FIG. Is output, and when it is smaller than 3, 0 (white) is output.

In the binary image from which the noise has been removed as described above, the black area is painted black and the straight lines and curves are smoothly expressed. The image data from which the noise has been removed in this way is output to the boundary point extraction unit 7 shown in FIG. This binarized image data is composed of main scanning width Bw × sub scanning width Bh pixels arranged in a grid as shown in S20 of FIG. Here, the boundary point extraction unit 7
Extracts white and black boundary points from this image data.

FIG. 8 shows a boundary point extracting operation explanatory diagram. The upper right of the figure shows the image data 7 that is the target of boundary point extraction.
It is 1. The image data 71 includes, for example, a rectangle 72
And a circle 73 are drawn. Boundary points are extracted from the left side of the image data. In this case, the number of black pixels on the scanning line is counted while scanning the image in the figure in the X direction. The results are plotted in the graph on the left. When this operation is repeated in the Y direction in the drawing, the graph shown on the left side of FIG. 8 is drawn. That is, in this graph, the horizontal axis represents the number of black pixels and the vertical axis represents the position in the Y-axis direction. For example, if the line for drawing the rectangle 72 and the circle 73 in the figure has a thickness of one pixel, the rectangle 7
100 pixels are extracted for the upper side and the lower side of 2, and 2 pixels are extracted for the other portions. Also, four pixels are extracted in the portion where the rectangle 72 and the circle 73 overlap when viewed from the left side. Boundary points are extracted in the same manner also in a direction that is just perpendicular to this direction, that is, when viewed from the lower side of the figure. That is, the graph shown on the lower side of FIG. 8 shows the number of black pixels on the vertical axis and the position in the X direction on the horizontal axis. In this case, the boundary points are extracted so that they can be drawn with a graph having a shape similar to that seen from the left side.

Note that this operation is specifically performed by the following arithmetic processing. First, the 4-pixel group ηN (x, y) including the pixel (x, y) of the target binary image N can be expressed by the formula shown in S21 of FIG. The sum of the output values has the content shown in S22 of FIG. From this, it is determined whether or not the output value of the pixel (x, y) in the binary image N is a black and white boundary point. This judgment is as shown in FIG.
The determination is made as shown in S23 of 6. Here, while sequentially scanning in the sub-scanning direction in the main scanning direction, the sum of the output values of the 4-pixel group ηN (x, y) including the pixel (x, y) of the target binary image N is added. Is larger than 0 and 2 or less, 1 (black) is output. On the other hand, if the sum of the output values is 0 or larger than 2, 0 (white) is output.

When the boundary points are extracted by such processing, the output is output to the collating unit 9 shown in FIG. In the collation unit 9, the output of the boundary point extraction unit 7
A plurality of registered boundary point images that have been registered in advance, that is, a reference boundary point image, is superimposed, and the matching rate is calculated from the number of matching pixels, and the weight ratio is calculated from the number of boundary point pixels of the input boundary point image. It is calculated, and the denomination, front / back, direction, and authenticity of the paper sheet are identified from the mapping relationship of each value. The boundary point image E is composed of main scanning width Pw × sub scanning width Ph pixels arranged in a grid pattern, like the binary image N.
This is shown in S24 of FIG. Now, a method of creating a registered boundary point image to be compared with the input boundary point image will be described. The registered boundary point image D serves as a reference for searching the correctness of the boundary point occurrence position of the target input boundary point image. This is to insert the same type of paper, for example, 10,000 yen bills into this device in a certain direction, and perform the processing up to the boundary point extraction described so far for thousands or tens of thousands of sheets. It is obtained by superimposing the results obtained by. When such boundary point images are obtained for a large number of 10,000-yen tickets of the same type, the occurrence rate distribution of the boundary points of the images is concentrated in a certain range.

FIG. 9 is an explanatory diagram of the boundary point image occurrence rate distribution which should serve as such a reference. The x-axis and the y-axis in the figure correspond to the vertical or horizontal direction of the ten thousand yen ticket. The two-axis EP (x, y) shown vertically is an axis showing the occurrence rate distribution of boundary points. As shown in the figure, for example, if the curve 82 drawn on the ten-thousand-yen ticket 81 is obtained as a boundary point image by the boundary point extraction, a mountain 83 having a very high probability of occurrence is obtained at this portion. Be done. In addition, this boundary point occurrence rate EP
(X, y) can be obtained as shown in S25 of FIG. 17 when the number of superposed sheets is s. As is clear from the figure, the boundary points commonly obtained for the same type of paper sheets are obtained with a high incidence, but the boundary points due to, for example, local stains or density changes appear with a low incidence. In order to actually register as a comparison target and use it as a reference boundary point image, such boundary point images with a low occurrence rate are excluded, and the boundary of the portion with the highest occurrence rate shown in FIG. 9 is excluded. Register the point image. As a result, when the same type of paper sheet is compared with the boundary point image, an accurate matching process can be performed with an extremely high matching rate.

The boundary point image as such a reference is, for example, in the case of identifying a ten-thousand-yen bill, a five-thousand-yen bill, and a one-thousand-yen bill. Register in advance. The matching unit 9 shown in FIG. 1 performs such matching between the registered boundary point image and the input boundary point image. Here, the registered boundary point image D, like the input boundary point image E, is the main scanning width Bw arranged in a grid pattern.
B. Sub-scanning width Bh. This state is shown in S26 of FIG. Here, when the two are superposed, the collation is performed with the center position obtained in the processing from the skew feeding correction unit 3 to the effective area determination unit 5 as a reference as described above.

FIG. 10 is an explanatory view of the reference boundary point image selection operation of the collating unit. As shown in the figure, there are four types of boundary point images that can be used as a reference, as described above, one for the front side, one for the back side, and one for which the insertion direction is the forward direction and the other for the reverse direction. Is registered. Reference numeral b in FIG. 10 is a classification number determined according to the inserted state of the bill.
Further, a in FIG. 10 is a classification number determined in accordance with the features such as the outer dimensions of the bill. Therefore, in FIG. 10, it is assumed that there are A types of paper sheet widths L, and A sets of boundary point images are prepared. Here, if the width of the inserted banknote in the longitudinal direction is measured and this is used as a reference first, FIG.
It is possible to select one set or several sets corresponding to the optimum a from the many reference boundary point images shown in FIG. Thereby, unnecessary collation processing can be omitted, the processing time can be shortened, and the recognition accuracy can be improved by utilizing the distinctive features such as the outer dimensions of the paper sheets.

Corresponding to the output value of the pixel (x, y) of one or several sets of reference boundary point images D thus selected,
The pixel (x, y) of the input boundary point image E to be compared and a peripheral eight pixel group ηE (x, y around the pixel x, y)
y) can be expressed by the formula shown in s27 of FIG.
The sum of the output values is as shown in FIG. 17S28. From this result, the correctness of the output value at E (x, y) is determined. When the output value of the pixel (x, y) in the registered boundary point image Dk is 1 (black), the output value of the pixel (x, y) in the input boundary point image E and the pixel (x, y) are centered. The sum of the output values of the peripheral 8 pixel group ηE (x, y) is searched. Also, the pixel (x, y) in the registered boundary point image Dk
Is 0 (white), the output value of the pixel (x, y) in the input boundary point image E and the peripheral 8 pixel group ηE (x, y) centered on the pixel (x, y) Do not search the sum of output values. In this way, the number of matching pixels AGR by superimposing the four types of registered boundary point images Dk
k is obtained according to S29 of FIG.

That is, while sequentially scanning in the main scanning direction in the sub-scanning direction, when the output value of the pixel of the registered boundary point image Dk is 1 (black), the output value of the pixel of the input boundary point image E is 1 (black) or the output value of the pixel of the input boundary point image E is 0 (white), and the peripheral 8 pixel group η
The output value 1 of the pixel of the registered boundary point image Dk is added to the matching pixel number AGRk only when either of the cases where the sum of the output values of E (x, y) is not 0 is satisfied. The matching unit 9 shown in FIG. 1 obtains the number of matching pixels AGRk by superimposing the four types of registered boundary point images Dk registered in advance, and outputs them to the recognition unit 10 shown in FIG. The recognition unit 10 calculates the ratio of the number of matching pixels AGRk output from the collation unit 9 to the number of boundary point pixels of the registered boundary point image Dk, that is, the matching rate. The recognition unit 10 also calculates the ratio of the number of boundary point pixels of the input boundary point image E to the number of boundary point pixels of the registered boundary point image Dk, that is, the weight ratio. Thus
The recognition unit 10 determines the type, front / back, authenticity, and direction of the target paper sheet based on the mapping relationship between the coincidence rate and the weight ratio.

FIG. 11 shows an operation explanatory diagram of the recognition unit. In this figure, the horizontal axis represents the weight ratio Pm and the vertical axis represents the concordance rate Pr, and the above mapping relationship is illustrated graphically. The coincidence rate Pr (k) of the input boundary point image E on the vertical axis is the coincidence rate pixel number AG by the superposing means with respect to the number of boundary point pixels of each of the k types of registered boundary point images Dk registered in advance.
It is the ratio of Rk. This is shown in S30 of FIG. Also, the weight ratio Pm of the input boundary point image E on the abscissa is shown.
(K) is a registration boundary point image Dk of k types registered in advance
Is the ratio of the number of boundary point pixels of the input boundary point image E by the matching unit 9 to the respective number of boundary point pixels. This is shown in S31 of FIG.

In FIG. 11, the hatched portion is defined by the mapping relationship between Pr and Pm of the target paper sheet.
It is the permissible range for identifying the type, front / back, authenticity, and direction. When the input boundary point image E and the registered boundary point image Dk extracted from the target paper sheet are of the same type, from the equations shown in S30 and S31 of FIG.
Theoretically, the values of Pm and Pr are close to 1, respectively. This is shown in S32 of FIG. In actual recognition, the output values of Pm and Pr
For the value of, the range of allowable values as described above must be defined.

FIG. 19 shows a conceptual diagram of such an allowable range determining method. In the block 121 of the figure, first, the 0th registered boundary point image D0 and several thousand or several tens of thousands of sheets of the same type (this number is S sheets) are input respectively and extracted. Superimposition with the boundary point image E is performed. From the equations of S30 and S31 in FIG. 18, the target sample number S = (0,
Corresponding to S), the values of the coincidence rate Pr (0, S) and the weight ratio Pm (0, S) are output. Similarly to block 121, in block 122, the first registered boundary point image D1 and S sheets of the same type of paper sheet are overlapped with the respective extracted input boundary point images E.
Then, in the same manner as in block 121, the matching rate Pr
The values of (1, S) and the weight ratio Pm (1, S) are output. In this way, similar processing is performed corresponding to the k kinds of registered boundary point image numbers registered in advance. Next, from the respective coincidence rates Pr and weight ratios Pm thus obtained,
A frequency distribution HST (Pr) of the coincidence rate and a frequency distribution HST (Pm) of the weight ratio are prepared, respectively. The result is shown in FIG. The result is shown in FIG. A section in which the frequency distribution of the coincidence rate and the weight ratio as shown in these figures does not become 0 is set to an allowable range. This is shown in S33 of FIG.

The allowable range IVP as described above allows a slight divergence or error of the value obtained by the equations of S30 and S31 shown in FIG. 18 for thousands of genuine sheets or tens of thousands of sheets. It is a range. If the input boundary point image E and the registered boundary point image Dk are of different types, the matching rate P obtained in S30 and S31 of FIG.
The mapping relationship between r and the weight ratio Pm is a mapping relationship outside the allowable range shown in FIG.

For example, for a banknote that is counterfeited by partial pasting, the threshold value determined in S11 of FIG. 14 changes, which changes the ideal threshold value of a real banknote, and the input boundary point is changed. The black and white boundary point generation position of the image E shifts over the entire screen. Therefore, in the case of fake paper sheets that are partially stuck together, the matching rate Pr
And the weight ratio Pm have a mapping relationship other than the allowable range shown in FIG. This makes it possible to reliably determine the authenticity of the paper sheet. In the case of a fake paper sheet created by copying the entire paper sheet, the above-mentioned FIG.
The threshold value obtained by the equation 11 of 4 is a threshold value that can be calculated by the fluctuation of the paper quality or print density of each target paper sheet, or the temporal change of the image line sensor used for reading the same. It is very different from the change of. Therefore, the processing can be interrupted at the threshold value determining stage. The recognizing unit 10 determines whether or not the mapping relationship between the coincidence rate Pr and the weight ratio Pm by superposition with the registered boundary point image Dk falls within the allowable range. Then, the mapping relations between the coincidence rate Pr and the weight ratio Pm corresponding to the numbers are sequentially compared in the order of the registered boundary point image number k, and the recognition result R is finally obtained when the mapping relations fall within the allowable range.
ES is output. This is shown in S34 of FIG.

As a result, with respect to the paper sheets read by the scanner 2 shown in FIG. 1, the boundary point image is extracted after recognizing the center position of the paper sheets, which processes many paper sheets of the same type in advance. By comparing with the boundary point image obtained by the above, highly reliable recognition is performed within a certain allowable range. The present invention is not limited to the above embodiments.
The scanner 2, the skew feeding correction unit 3, the buffer memory 4, the effective area determination unit 5, the binarization unit 6, the boundary point extraction unit 7, or the collation unit 9 and the recognition unit 10 which have been introduced in the above embodiments, respectively.
Each block such as may be replaced with a block having various configurations having the same function, and the arithmetic processing and the like executed in each block may be processed by one control computer. It may be processed by a separate control circuit.

[0037]

The above-described paper sheet recognition apparatus of the present invention reads the image of the entire area that requires the recognition of the paper sheet, extracts a certain statistical amount from the grayscale image, and binarizes it. Since the threshold value is determined, the data targeted for recognition is not affected by the quality and print density of the paper of the paper sheet, and stable recognition is always possible. Furthermore, by using such a threshold value, the accuracy of the boundary point generation position is improved, and a simple matching process can be realized with high reliability. Also, if the size of the area to be read by the scanner is set to be equal to or greater than the maximum diagonal length of the target paper sheet, all the necessary images will not be affected by the displacement of the target paper sheet and the skew displacement. Can be extracted.

Further, before binarizing the image data, the center position of the paper sheet is detected from the target gray image, and the area occupied by the paper sheet is set from the gray image. The amount of subsequent signal processing can be reduced, and the processing speed can be increased. In addition, the process for creating the reference registration boundary point image and the process for extracting the input boundary point image may include fluctuations in paper quality, print density, etc., environmental temperature fluctuations, and scanner output fluctuations. However, since the method is not affected by this, it is possible to perform highly reliable collation processing that asks whether the boundary point generation position is correct. Also,
By removing the isolated points and noise from the output of the binarization processing means, the reliability of the boundary point image can be improved.

Further, the difference between the longitudinal width of the target paper sheet and the previously registered longitudinal width is detected, and based on this difference, the registered boundary point image used for superimposition is selected. By performing the above, it is possible to omit unnecessary collation processing and shorten the collation processing time. Further, the boundary point image serving as a reference is targeted at a large number of sheets of the same type, and the image of the portion having a low occurrence rate is removed from the boundary point image obtained through the same operation as the recognition of the sheets. Since this is done, it is possible to perform collation with an extremely high matching rate. In addition, the scale used as a reference for identification can be only the mapping relationship between the coincidence rate and the weight ratio, and since it is possible to use only one reference, the reliability of recognition can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of a paper sheet recognition apparatus of the present invention.

FIG. 2 is a block diagram of a scanner and a skew feeding correction unit of the apparatus of the present invention.

FIG. 3 is an operation explanatory diagram of a skew feeding correction unit of the device of the present invention.

FIG. 4 is an explanatory view of the operation of the position detection unit of the device of the present invention.

FIG. 5 is an operation explanatory diagram of an effective area determination unit of the device of the present invention.

FIG. 6 is a block diagram of a binarizing unit of the device of the present invention.

FIG. 7 is an explanatory diagram of the operation of the binarizing unit of the device of the present invention.

FIG. 8 is an explanatory diagram of a boundary point extracting operation of the device of the present invention.

FIG. 9 is an explanatory diagram of a reference boundary point image occurrence rate distribution of the device of the present invention.

FIG. 10 is an explanatory diagram of a reference boundary point image selection operation of the matching unit of the device of the present invention.

FIG. 11 is an operation explanatory diagram of a recognition unit of the device of the present invention.

FIG. 12 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 13 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 14 is an explanatory diagram of the arithmetic operation of the device of the present invention.

FIG. 15 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 16 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 17 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 18 is a diagram illustrating a calculation operation of the device of the present invention.

FIG. 19 is a conceptual diagram of a method for determining an allowable range of the device of the present invention.

20 is an explanatory diagram of a result of FIG.

FIG. 21 is an explanatory diagram of a result of FIG.

[Explanation of symbols]

 1 paper sheet 2 scanner 3 skew correction unit 4 buffer memory 5 effective area determination unit 6 binarization unit 7 boundary point extraction unit 8 reference boundary point image 9 collation unit 10 recognition unit

Claims (7)

[Claims] 1. A scanner for reading data on a paper sheet, and at the time of reading the data of the paper sheet, the skew angle is detected from the transport state of the paper sheet, and the skew of the read signal of the scanner is corrected. And a grayscale image data written in the buffer memory is analyzed to extract the edge position and the center position of the paper sheet in the buffer memory, and the grayscale image data is written. Of the effective area determining unit that determines the effective area of the paper, and reads out the grayscale image data of the effective area from the buffer memory, determines the binarization threshold value from the frequency distribution of the density, and 2 grayscale image data
A binarization processing unit that performs binarization, a boundary point extraction unit that obtains a boundary point image in which white / black boundary points are extracted from the binarized image data, the boundary point image, and a reference prepared in advance. And a recognition unit that recognizes the type of the paper sheet and the authenticity based on the output of the matching unit. A paper sheet recognition device characterized by the above. 2. The paper according to claim 1, wherein the readable length of the scanner is set to be equal to or more than the diagonal length of the paper sheet having the largest diagonal line among the paper sheets to be recognized. Leaf recognition device. 3. The paper sheet according to claim 1, wherein the effective area of the grayscale image data written in the buffer memory is set to an area occupied by the largest one of the paper sheets to be recognized. Class recognition device. 4. The paper sheet recognizing apparatus according to claim 1, further comprising noise removing means for removing isolated points and noise from the output of the binarization processing means. 5. A boundary point image serving as a reference is a boundary point image obtained through the same operation as the recognition of paper sheets for a large number of sheets of the same type, and a portion having a low occurrence rate is detected. The paper sheet recognition apparatus according to claim 1, wherein the image is removed. 6. A feature amount other than the boundary point image of a paper sheet to be matched is used to select a corresponding one of the reference boundary point images prepared in advance, and then a matching rate is selected. The paper sheet recognition apparatus according to claim 1, wherein the distribution is created. 7. A boundary point image of a paper sheet to be collated,
2. The paper sheet is recognized according to whether or not the mapping relationship between the coincidence rate and the weight ratio by superimposing the boundary point images, which is a reference prepared in advance, is within a certain range. Paper sheet recognition device.