JPH08194857A - Paper sheet discriminating device - Google Patents

Abstract

PURPOSE: To easily calculate similarity and to shorten processing time. CONSTITUTION: The total number of picture elements in reference data for discrimination is set up to 40×18=720, and with respect to all sorts of paper sheets, the number of picture elements of '1' e.g. is set up to 100 picture elements. Namely patterns are mutually different in accordance with the sorts of paper sheets, and when respective patterns are binarized by the same threshold, the numbers of picture elements of '1' are different in accordance with the sorts of paper sheets. However the numbers of picture elements of each sort can be equalized to be 100 picture elements by adjusting the threshold in accordance with the sorts of paper sheets. When a paper sheet 3 to be discriminated is read out by a CCD sensor 5, similarity with the reference data is calculated by finding out the number of picture elements of '1' in the reference data superposed with picture elements of '1' in the paper sheet 3 in a product sum computing element 14. When the data of the paper sheet 3 completely coincide with the reference data, the number of picture elements becomes '100', and when both data do not completely match with each other, similarity of '100' to '0' can simply be found out in accordance with the number of superposed picture elements.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to the types of paper sheets, the front and back sides,
The present invention relates to a paper sheet discriminating device for discriminating a direction.

[0002]

2. Description of the Related Art Generally, paper sheets such as banknotes are printed with patterns and characters, and the type and authenticity of the paper sheets can be discriminated. For this reason, the conventional sheet discriminating apparatus includes a photoelectric converter such as a CCD and detects a pattern or a character pattern as an electric signal. This electric signal is layered and processed as multi-valued digital data. On the other hand, reference data obtained by previously reading and binarizing the pattern of each paper sheet is stored in the apparatus. Then, the paper sheet is discriminated by checking which reference data the data obtained by reading with a CCD or the like matches (for example, Japanese Patent Laid-Open No. 58-056088).
No. 02-167490, No. 02-186.
493).

[0003]

However, the above-mentioned conventional techniques have the following problems. That is, for example, when identifying banknotes of 1,000 yen, 5,000 yen, and 10,000 yen,
Standard data representing patterns of each banknote are prepared. Such reference data is, for example, an approximate value by the least squares method excluding an average value of pattern densities and an abnormal value. Then, the density data of the image read by the sensor is compared with this reference data. In this case, the print density of the paper sheets varies. In addition, the light amount of the light source changes due to changes in environmental temperature and the like. Further, the output of the amplifier that amplifies the output of the sensor changes slightly. Therefore, due to these factors, the image read by the sensor changes slightly. As a result, the density data read by the sensor does not completely match the reference data for that category.

Therefore, when discriminating the types of banknotes of 1,000 yen, 5,000 yen, and 10,000 yen, the data of the banknote to be judged and the reference data of 1,000 yen, 5,000 yen, and 10,000 yen. The type of banknote is determined by obtaining the degree of similarity with and the reference data that is most similar to the degree of the degree of similarity. However, there is a problem in that the calculation of the degree of similarity dealing with multi-values (8 bits, etc.) is generally complicated, the calculation process takes time, and it becomes a factor that the processing time for discriminating the type of banknote becomes long.

[0005]

In order to solve the above-mentioned problems, the paper sheet discriminating apparatus of the present invention is characterized by the following points. (1) An integration / averaging processing unit is provided that integrates and averages the patterns of paper sheets that have been position-detected and cut out for each predetermined region. A binarization processing unit that binarizes the data processed by the integration and averaging processing unit using a certain fixed threshold value is provided. A product-sum calculation unit that obtains a product sum of the feature vector and each type of reference vector stored in advance as a feature vector using the pattern vector binarized by the binarization processing unit is provided. Each reference vector is binarized so that all of them have the same value, and a reference pattern storage unit is provided for pre-storing the value of each element of each reference vector. A discrimination unit that discriminates the type of the reference pattern having the maximum value among the calculated values obtained by the sum of products calculation unit as the type of the paper sheet to be discriminated is provided.

[0006]

[Action]

(1) The total number of pixels of the reference data for discrimination is 40 × 18 = 7
For example, "1" for all 20 types of paper sheets
The number of pixels of is 100. That is, the pattern differs depending on the type of paper sheet, and when binarized with the same threshold value, “1”
Although the number of pixels of No. varies depending on the type of paper sheet, the number of pixels of "1" is made equal, for example, 100 by adjusting the threshold value according to the type of paper sheet. In this way, when the sheet to be discriminated is read by the CCD, the similarity with the reference data is calculated by calculating the number of pixels of "1" of the reference data that overlaps the pixel of "1" of the sheet to be discriminated. Just ask. As a result, when the data of the paper sheet to be discriminated is completely matched with the reference data, it is "100", and when it is not completely matched, "100"-
The similarity of “0” is calculated. In this way, the degree of similarity can be easily obtained.

[0007]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a paper sheet discriminating apparatus of the present invention. The illustrated device is a position detection unit 1
0, position correction cutout unit 11, integration and averaging processing unit 12,
A binarization processing unit 13, a sum of products calculation unit 14, a reference pattern storage unit 15, and a determination unit 16 are provided. The position detector 10
The entire position of the paper sheet 3 stored in the buffer memory 9 is detected. The position correction cutout unit 11 removes the paper sheet 3 from the image stored in the buffer memory 9 excluding the background portion.
Cut out the whole pattern of. Integral averaging processor 12
Is to integrate and average the entire pattern of the paper sheet 3 for each predetermined area.

The binarization processing unit 13 is an integration / averaging processing unit 1.
The data processed by 2 is binarized using a fixed threshold value. The sum-of-products calculation unit 14 includes a binarization processing unit 13
By using the binarized pattern vector as a feature vector, the product sum of the feature vector and each type of reference vector stored in advance is obtained. That is, the sum of the products of the corresponding components of the pattern vector obtained by the binarization processing unit 13 and each reference vector stored in the reference pattern storage unit 15 is calculated. The reference pattern storage unit 15 is a ROM or R
Binary data of each element of each reference vector is stored in advance. If there are n types of reference paper sheets, all necessary number of reference vectors are stored.

The determination unit 16 determines that the type of the reference pattern having the maximum value among the product sums calculated by the product sum calculation unit 14 is the type of the sheet 3 to be determined.
In FIG. 1, the image of the paper sheet 3 is stored in the buffer memory 9 as follows. LED array 2
Illuminates a region of the paper sheet 3 in the longitudinal direction (direction perpendicular to the transport direction M) from the lower side in the drawing. Here, the paper sheet 3 has a peculiar pattern printed on both sides like a bill, for example. When the paper sheet 3 is conveyed in the arrow direction by the conveying rollers 1a and 1b and passes through the illumination area, transmitted light corresponding to the shade of the printed pattern is obtained. CCD sensor 5 is lens 1
The transmitted light collected at 9 is received and an analog signal is generated. The operation of the CCD sensor 5 is performed by the paper edge detection sensor 1
The sampling pulse T1 generated by the control circuit 18 is controlled by the pulse T0 obtained in step S7.

The amplifier 6 amplifies the analog signal output from the CCD sensor 5. The A / D converter 7 controls the control circuit 1 according to the pulse T0 obtained by the paper sheet leading edge detection sensor 17.
It is controlled by the timing signal T2 created in 8. The timing signal T2 is generated by the number of effective pixels of the CCD sensor 5, and a signal for one line in the longitudinal direction of the paper sheet 3 is sequentially sent to the buffer memory 9 via the shading correction circuit by the control signal T3. . These operations are performed at a constant interval in the transport direction M by the sampling pulse T1 in the control circuit 18, and finally an area including all print patterns of paper sheets is stored in the buffer memory 9. The value stored in the buffer memory 9 is the paper sheet 3
Since it includes both the portion and the background portion other than that, it is necessary to accurately cut out only the portion of the paper sheet 3.
This processing is performed by the position detection unit 10 and the position correction cutout unit 11 as described above.

Next, the operation of the above-mentioned device will be described. First, the operation of the sheet discrimination process will be described. FIG. 2 is a flowchart of the determination processing procedure. First, step S1
Then, the image data of the paper sheet 3 is stored in the buffer memory 9 as described above. Then, in step S2, this image data is corrected by the shading correction circuit for the variation in the light amount of the light source of the illumination means at the reading position. Then, in step S3, the position detection unit 10 detects the position and orientation of the paper sheet.

Next, in step S4, the position correction cutout unit 11 performs position correction based on the information detected by the position detection unit 10, and cuts out the outer frame of the paper sheet. And
In step S5, the integration / averaging processing unit 12 integrates several pieces of pattern data printed on the paper sheet cut out by the position correction cutout unit 11 in each of the main scanning direction and the sub-scanning direction to form an area of several mm square. Integral averaging process is performed.
By this integration and averaging process, variations in image data due to expansion and contraction of paper sheets, print misalignment, stains, wrinkles, variations in print density, etc. are reduced, and uniform data can be obtained. Next, in step S6, the integrated average data of the print pattern obtained by the CCD sensor is further binarized by a binarizing unit with an appropriate threshold value that can obtain the features (character and graphic portions) of the paper sheet, Get the feature vector.

Then, in step S7, the degree of similarity between this feature vector and a reference vector described later is calculated.
The reference vector with the highest degree of similarity determines the type of paper sheet 3 to be discriminated. Therefore, in step S8,
The determination unit 16 makes this determination and sends the result to the upper interface. As described above, the type, front / back, direction, etc. of the paper sheet 3 are determined.

Next, the procedure for calculating the degree of similarity will be described in detail. First, the corresponding pixels of the paper sheet are represented by attaching respective numbers as shown in FIG. Then, the total image data of the paper sheets is 72 from X0 to X719.
It consists of 0 binary pixel data. Then, consider these as one pattern vector. The total number of pixel data is not limited to 720 and may be changed as appropriate. If the feature vector to be discriminated is X, the reference vector of the first type of paper sheet is X (1), and the reference vector of the second type of paper sheet is X (2), these vectors The correlation can be expressed schematically as shown in FIG. At this time,
The angle formed by these vectors is θ as shown in FIG.
Assuming (1) and θ (2), the smaller the θ, the more similar the correlation of the vectors. For example, in the example of FIG. 4, since θ (2) <θ (1), the feature vector X is more similar to the reference vector X (2).

This θ (i) is calculated by using the inner product of the vectors,
It is calculated by the following formula. cos θ (i) = (X · X (i)) / {| X | · | X (i) |} (1) where X (i) is the reference vector of the paper sheet of type i, θ (i) is the feature vector X and the reference vector X
It is an angle with (i). Cos θ (i) in the above equation (1)
Takes a value of 0 or more and 1 or less, and if these are applied to the pattern of FIG. 3, the degree of similarity is obtained as in the following expression (2). S (i) = Σ (Xj · Xj (i)) / (ΣXj ² · ΣXj (i) ² ) ^1/2 (2) where Σ means the sum from j = 0 to j = 719 .

This is calculated for all reference vectors, and the one that is closest to "1" is the most similar to that type of paper sheet. Then, an appropriate threshold value for rejecting the paper sheet 3 is set, and if it is equal to or larger than this threshold value, it is officially determined to be a paper sheet of that type. On the other hand, if it is less than or equal to this threshold value, the CPU of the host device is notified that the paper sheet to be discriminated should be rejected. Such similarity is determined by the angle formed by the vectors and is not affected by the difference in the size of the vectors. Therefore,
For example, even if the output of the sensor amplifier slightly changes and the density level changes overall, the similarity is not affected.

Next, a method of creating the reference vector will be described with an example. FIG. 5 shows a procedure for creating a reference vector. First, in step S11, n is set as an initial value.
= 1; N = N0 (N0 indicates the number of samples), and in steps S12 to S15, steps S1 to S in FIG.
The same process as 4 is performed. Then, in step S16, integral averaging processing is performed on the feature data, and the feature data is added one after another. Then, this processing is performed in step S
It is executed until n = N0 in the judgment formula of 17. That is, this process is executed for N0 sheets of paper, and step S1
Create a reference vector at 8.

In step S18, first, the value obtained in step S16 is divided by the sample number N0 to obtain the sample number N0.
Calculate the average value in. Then, this average value is used as a representative value of the sample number N0. It should be noted that statistics other than the average value may be used as the representative value. Next, this representative value is binarized using the p-tile method with an appropriate area ratio from the dark side. In this case, it is determined that the closer the A / D converted pixel value is to “0”, the more characteristic. By doing so, the size (square root of the sum of squares) of all reference vectors of each type of paper sheet can be made equal.

As a result, it becomes unnecessary to calculate the magnitudes of the feature vector X and the reference vector X (i) in the above-mentioned equation (2). Therefore, the algorithm for calculating the degree of similarity becomes simpler than the case of directly using the equation (2). For example, in the case of banknotes, the size of the reference data is the same between denominations of 1,000 yen, 5,000 yen, and 10,000 yen. As a result, even if the size of the bill to be discriminated differs from each other, only the numerator of the formula (2) is calculated, so that the relatively similar reference vector is obtained. That is, the similarity is
It is calculated by the following equation (3). S (i) = Σ (Xj · Xj (i)) (3) Here, Σ means the total sum from j = 0 to j = 719.

Since the feature vector is binarized in advance, as is apparent from the equation (3), the calculation with the reference vector is only a simple product-sum calculation of "0" and "1". From a different point of view, this is the same as determining the number of common parts between the set of "1" components of the feature vector and the set of "1" components of the reference vector. As a result, the time for calculating the degree of similarity is significantly shortened. In this way, the type, front and back, direction, etc. of a paper sheet can be discriminated only by the sum of products of two values, and it is very fast, and the high feature extraction ability can be effectively utilized to make a highly reliable discrimination. can do.

Further, as described above, even when the light amount of the light source or the amplifier output of the circuit for amplifying the output of the reading sensor slightly changes due to the temperature change, the change causes the overall density of the pattern of the paper sheet. Change, the magnitude of the obtained vector only changes, and the direction remains unchanged.
Therefore, there is no problem that the ability to discriminate the type, front and back, and direction of the paper sheet is lowered, and the rejection rate is increased.
Furthermore, when changing the illumination system or lens system,
Since the reference data can be easily created as described in detail above, the development man-hours can be shortened as compared with the conventional case.

The present invention is not limited to the above-mentioned embodiments, and it goes without saying that various modifications can be made. For example, in the above-described embodiment, the case of discriminating the denomination of banknotes has been described, but the present invention is not limited to this, and if the total number of the types is finite, the paper sheets such as other securities can be booked. By applying the invention, it is possible to determine which type it is.

[0023]

As described above, according to the paper sheet discriminating apparatus of the present invention, since the magnitudes of the reference vectors are made equal to each other, the calculation of the degree of similarity can be simplified and the bill It is possible to shorten the processing time for determining the type.

[Brief description of drawings]

FIG. 1 is a block diagram of a first embodiment of a paper sheet discriminating apparatus of the present invention.

FIG. 2 is a flowchart illustrating a paper sheet discrimination processing procedure.

FIG. 3 is a flowchart illustrating a procedure for creating a reference vector.

FIG. 4 is an explanatory diagram of vector correlation.

FIG. 5 is an explanatory diagram of corresponding pixels in a bill.

[Explanation of symbols]

 12 integration / averaging processing unit 13 binarization processing unit 14 product sum calculation unit 15 reference pattern storage unit 16 determination unit

Claims (1)

[Claims] 1. An integral averaging processing unit that integrates and averages a pattern of a paper sheet that has been position-detected and cut out for each predetermined region, and data processed by the integral averaging processing unit,
A binarization processing unit that performs binarization using a certain fixed threshold, and a pattern vector binarized by the binarization processing unit as a feature vector, and the feature vector and the reference vector of each type stored in advance. The sum-of-products calculation unit for obtaining the sum-of-products and the binarization process are performed in advance so that the magnitudes of the respective reference vectors are all the same, and the values of the respective elements of the respective reference vectors obtained as a result are stored in advance. A reference pattern storage unit for determining the type of the reference pattern having the maximum value among the calculated values obtained by the product-sum calculation unit as the type of the paper sheet to be determined, and A paper sheet discriminating apparatus comprising: