KR100599913B1 - Device, method and program for identifying paper sheet - Google Patents

Abstract

The image processing unit 18 performs meandering correction and density correction on the piece of paper image data obtained by the optical sensor unit 13. The librarian comparison unit 18 creates the comparison paper image data for each gold species from the paper image data created by the image processing unit 17 and the reference data of each gold species stored in the reference data unit 19. Then, the processing for pattern matching with the same kind of librarian data stored in the data section 20 for purchase of the comparison bill image data is performed for all kinds of gold species, and the gold species having the minimum distance between patterns is regarded as the paper type of the paper. do. The discrimination result is stored in the storage unit 21.

Paper money discrimination device, ATM, withdrawal and withdrawal of bills, bill differentiation, kind discrimination of partial bills, bill processing device

Description

Paper discrimination device, paper discrimination method and program {DEVICE, METHOD AND PROGRAM FOR IDENTIFYING PAPER SHEET}

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a paper sheet discriminating apparatus of a paper sheet processing apparatus, and more particularly, to a paper sheet discriminating apparatus and a paper sheet discriminating method for discriminating the gold species of a partial bill (paper sheet) in which at least half of paper sheets are missing.

In a paper sheet device in which papers such as paper money are put and withdrawn, such as banknotes (ATM), paper sheets are stored in the device, and the deposit and withdrawal process is performed in accordance with a customer's instruction input. In this paper sheet processing apparatus, the differentiation process of the kind (paper seed | species etc.) of paper sheets (paper | money etc.) to be inputted and received is performed.

This discrimination process is generally performed by the following method. The entire paper sheet is divided into a plurality of minute regions, and the shade and appearance of each of the minute regions are measured by an optical sensor, a thickness sensor, or the like, and the measured values are converted into gradation values or the like by quantization processing and stored. Then, a predetermined image processing (tilt correction or density correction processing) is performed on the image data composed of the gradation values, and the paper data (paper money) is obtained by comparing the obtained image data with the purchased data (image data created from true winding). It is supposed to determine the type (gold type) of the.

By the way, the conventional paper sheet processing apparatus makes discrimination object only that the external shape is normal, and rejects the paper sheet having an unusual appearance as it does not discriminate as a discrimination object, and considers it as discrimination abnormality. This is because in the conventional paper sheet processing apparatus, the type of paper sheets is discriminated by buying image data of the entire sheet paper and pattern matching with the data. In such a conventional paper sheet processing apparatus, as a matter of course, for partial bills (paperboards) in which more than half of the paper sheets are missing, the area comparable to the librarian data is too small to specify the gold type.

By the way, as operation | movement and 24 hours operation at the unmanned store of paper sheet processing apparatus have become common in recent years, it has been strongly requested | required to always determine the banknote balance (amount of stock) in the bill storage in a paper sheet processing apparatus. . In the paper sheet processing apparatus, the banknotes inserted from the input port or the banknotes for the means money taken out from the banknote storage may be cut off in the middle of being conveyed to the discriminating unit in the apparatus. In such a case, the holder of the paper sheet processing apparatus has a large difference between being rejected with known gold species and being rejected with unknown gold species, even if it is a cut piece. However, in the discriminating part of the conventional paper sheet processing apparatus, the cut paper sheets (paper fragments) were rejected without specifying the type (gold species).

It is an object of the present invention to provide a paper sheet discriminating device and a paper sheet discriminating method capable of reliably specifying a paper type of a partial bill (so-called paper sheet) in which at least half of paper sheets are missing.

The finger discrimination apparatus of the present invention includes reference data storage means, librarian data storage means, sensor means, and librarian comparison means.

The reference data storage means stores reference data, which is image data of a true bill, with respect to at least one gold type. The librarian data storage means stores librarian data for discrimination of true bills by the same number of gold species as the reference data. The sensor means acquires image data of a piece of paper, which is a piece of paper sheet. The librarian comparison means synthesizes the piece of paper image data obtained by the sensor means with the reference data of each gold species stored in the reference data storage means to create the comparison bill image data for each gold species, and the comparison bill image data Pattern is matched with the librarian data of the same gold species stored in the librarian data storage means to discriminate the gold species of the paper.

Since the reference data is image data of a true bill, the comparison bill image data obtained by affixing and combining a piece of paper image data with the reference data is the image data of the true book that the portion other than the piece of paper image data 50 has the librarian data. To match. Therefore, by pattern matching the comparison paper money image data and the librarian data of the same paper money, it is possible to discriminate whether the paper money image data, that is, the paper money is part of the true hand, and estimate the paper money of the paper money based on the result of pattern matching ( Can be discriminated).

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram showing the system configuration of a finger discriminating apparatus as a first embodiment of the present invention.

FIG. 2 is a view for explaining the outline of the processing of the paper discrimination apparatus of the first embodiment. FIG.

Fig. 3 is a flowchart for explaining the entire paper discrimination process of the paper discrimination apparatus of the first embodiment.

4 is a flowchart for explaining concentration correction processing in the first embodiment.

FIG. 5A is a diagram showing an entire data structure of librarian data, and FIG. 5B is a diagram showing a data structure of pixels of librarian data.

FIG. 6A is a diagram showing an entire data structure of comparison bill image data, and FIG. 6B is a diagram showing a data structure of pixels of comparison bill image data.

7 is a schematic diagram of a pattern matching method between comparison bill image data and librarian data.

8 is a schematic diagram of pattern matching processing in a front image.

9 is a detailed flowchart of pattern matching processing in the front image.

10 is a schematic diagram of a pattern matching process in a horizontal line image.

Fig. 11 is a detailed flowchart of a pattern matching process of horizontal line image data.

12 is a schematic diagram of a pattern matching process in a vertical line image.

Fig. 13 is a detailed flowchart of the pattern matching processing in the vertical line image.

14 is a schematic diagram of a pattern matching process in a left oblique line image.

15 is a schematic diagram of a pattern matching process in a right oblique line image.

Fig. 16 is a detailed flowchart of pattern matching processing in the left inclined line image.

Fig. 17 is a detailed flowchart of the pattern matching processing in the right inclined line image.

Fig. 18 is a block diagram showing the system configuration of the finger discriminating apparatus according to the second embodiment of the present invention.

19 is a diagram illustrating a configuration of a neuro calculator.

20 is a diagram illustrating a configuration of a unit of a neuro calculator.

21 is a diagram for explaining data input to a neuro calculating section.

Fig. 22 is a diagram for explaining an input pattern when the neuro calculator performs front image pattern recognition.

Fig. 23 is a diagram for explaining an input pattern when a neuro calculator performs vertical line image pattern recognition.

Fig. 24 is a diagram for explaining an input pattern when a neuro calculator performs horizontal line image pattern recognition.

Fig. 25 is a diagram for explaining an input pattern when the neuro calculating section recognizes the left inclined line image pattern.

Fig. 26 is a diagram for explaining an input pattern when the neuro calculating section recognizes the right inclined line image pattern.

Fig. 27 is a flowchart for explaining processing by which a finger discriminating apparatus according to the second embodiment performs front image pattern recognition by a neuron operation to discriminate the kind of paper.

Fig. 28 is a flowchart for explaining processing by which the finger discrimination apparatus of the second embodiment performs horizontal line image pattern recognition by a neuron operation to discriminate the types of paper.

Fig. 29 is a flowchart for explaining processing by which the finger discriminating apparatus of the second embodiment performs vertical line image pattern recognition by neuro operation to discriminate the paper species of the paper.

Fig. 30 is a flowchart for explaining processing by which the finger discriminating apparatus of the second embodiment recognizes the inclined line image pattern by the neuron operation to discriminate the paper species of the paper.

The present invention can specify the species (gold species) of the paper sheets from a part of the paper sheets (paper pieces), and can be realized as one unit, and can also be assembled to various paper sheet processing apparatuses.

(First embodiment)

1 is a system configuration diagram of the first embodiment of the present invention.

The paper discrimination apparatus 10 of the first embodiment shown in FIG. 1 includes a central processing unit 11, a paper sheet inrush sensor unit 12, an optical sensor unit 13, an amplifier unit 14, and an A / D converter unit. (15), an image storage unit 16, an image processing unit 17, a librarian comparison unit 18, a reference data unit 19, a librarian data unit 20, and a storage unit 21. In addition, in the following description, the case where a banknote which is a kind of paper sheet is handled in this Example is demonstrated.

The paper sheet inrush sensor unit 12 is a sensor that detects that a paper sheet (paper money) has entered into the apparatus 10 and notifies the central processing unit 11 of this inrush.

The optical sensor unit 13 is a line sensor and, after being detected by the paper sheet inrush sensor unit 11, acquires an image of the paper sheets conveyed in the apparatus 10 in units of lines. When the central processing unit 11 receives the notification from the paper sheet inrush sensor unit 12, the central processing unit 11 starts the optical sensor unit 13, and the sampling frequency and sampling of the optical sensor unit 13 are sampled according to the conveyance speed of the paper sheets. Control the time. The optical sensor unit 13 receives control from the central processing unit 11, acquires an analog image signal of the conveyed paper sheet, and outputs it to the amplifying unit 14.

The amplifier 14 amplifies the analog image signal input from the optical sensor unit 13 and outputs it to the A / D converter 15.

The A / D converter 15 converts and outputs an analog image signal input from the amplifier 14 into a digital image signal. This digital image signal is a gradation value in which the darkest point is O and the brightest point is the maximum. This gradation value is obtained for each pixel.

The digital image signal output from the A / D conversion unit 15 is sequentially stored in the image storage unit 16. As a result, the digital image signal of the paper sheet (or the paper) image obtained by the optical sensor unit 13 is stored in the image storage unit 16. This digital image signal is larger than the size of the paper sheets, as will be described later, and also includes the paper background image.

Under the control of the central processing unit 11, the image processing unit 17 performs coordinate transformation (tilt correction processing) on an image signal (image data) stored in the image storage unit 18, thereby causing paper sheets (or, Normalizes the image data). Then, density correction processing is performed on the normalized image data. The image storage unit 18 buys paper sheets obtained by performing the above coordinate transformation, density correction, and the like, and outputs the image data of the paper sheets to the comparison unit 18.

The librarian comparison unit 18 pattern-matches the paper data input from the image processing unit 17 with the librarian data stored in the librarian data unit 20 to discriminate the authenticity of the paper sheets input into the apparatus 10. Is done. In addition, although this is a feature of the present invention, the librarian comparison unit 18, if the image data input from the image processing unit 17 is a piece of paper image data (image data of a piece of paper (paper piece)), The type (gold paper type) of the paper money image is determined by data generation of an image superimposed (attached) on the reference data of each paper money type, and purchase of the created image data (comparation paper money image data) and pattern matching with the data. .

The reference data unit 19 stores reference data of all kinds of banknotes handled by the apparatus 10. This reference data is data used for identifying (estimating) the kind of paper money, and is image data created by reading the true winding by the optical sensor, and density correction similar to the paper money image data is performed. This density correction is performed in order to prevent spots from occurring when the above-mentioned piece paper image data is affixed to reference data.

The librarian comparison unit 18 creates the comparison bill image data for each of the gold types by a synthesis process of bonding the image data of the paper money to the reference data of each gold type of the bill, and compares the bill image data for comparison of these gold types. By comparing with the librarian data of the same kind, the kind of paper is identified (differentiated).

The librarian data unit 20 stores the librarian data used by the librarian comparison unit 18 to discriminate the authenticity of paper sheets (paper money) and the type of paper species. This librarian data is also prepared for all types of banknotes handled by the apparatus 10 in the same manner as the reference data. This librarian data is created by reading the true winding with an optical sensor (for example, the optical sensor unit 13). In this case, in general, since a deviation occurs in the gradation value (density) of each pixel for each true winding, a predetermined number of true windings are read in advance by an optical sensor, and the standard deviation of the gradation average value and gradation value of each pixel is obtained, and the value thereof. The data associated with the pixel is purchased and stored as data.

In authenticity discrimination, for example, the difference between the gradation values of the corresponding pixels is calculated, and if the value is within the standard deviation, it is determined that both pixels match, and the percentage of the matching pixels is greater than or equal to that. The authenticity of banknotes is discriminated against as a guide.

In addition, in the paper type discrimination of the paper money, in the pattern matching between the librarian data and the comparison bill image data, for example, a distance (for example, Euclidean distance) between both image data is calculated and the paper money is determined based on the distance. Discuss gold grades.

The storage unit 21 stores the result of the paper kind of paper money identified by the librarian comparison unit 18, the authenticity determination result of the banknote, and the like.

FIG. 2: is a figure explaining the outline | summary of the paper-sheet discrimination process by the said paper-sheet discrimination apparatus 10. As shown in FIG.

The bio image data 31 shown in FIG. 2A is image data obtained by detecting the paper conveyed meandering by the optical sensor unit 13, and the image data ( 41) exists. In FIG. 2A, the broken line portion connected to the piece of paper image data 41 and the outline shows the outline of the other piece of paper that is cut off and lost.

The image processing unit 17 performs coordinate transformation processing on the finger image data 41, meanders and normalizes the finger image data 41 (the paper image data on the upper edge of the upper left corner of the raw image data 31). The top side of (41) overlaps). Then, the density correction is performed so that the white portion (the image data of the portion corresponding to the white portion of the outer edge of the sheet in the sheet image data 41) of the edge of the sheet becomes the reference value with respect to the normalized sheet paper image data 41. Finally, the piece of paper image data 42 as shown in FIG.2 (b) is created.

As shown in FIG. 2C, the librarian comparison unit 18 stores the above-described piece image data 42 in the reference data unit 19, A gold species, B gold species,... The reference data 50 (50A (A gold grade reference data), 50B (B gold grade reference data), ..., 50N (N gold grade reference data)) of each gold grade, etc. are superimposed (attached) on the basis of a corner, The banknote image data 60 for comparison of each gold species is created. The bill paper image data 60 for comparison of each of the gold species is librarian data 70 (70A (librarian data of A gold species), 70B (librarian data of B gold species) of each gold species stored in the librarian data unit 20, ... is image data for pattern matching with 70N (N type of librarian data).

By the way, the pasting of the piece of paper image data 42 onto the reference data 50 at this corner reference is determined by the shape of the piece of paper image data 42 (shape of the piece of paper). In the case of the substantially triangular piece of paper image data 42 as shown in Fig. 2 (c), since the angle portion is assumed to correspond to any one of the two corners of the original banknote, the reference data 50 (Fig. The reference data 60 shown in 2 (c) is the reference data 50 of the golden type B), and the two pieces of comparison are made by attaching the piece of paper image data 42 to each corner of the upper right corner and the lower left corner, respectively. Bill paper image data 60 is created.

If the shape of the paper piece is other than the shape shown in Fig. 2 (c), the paper money image data 42 for comparison is created by affixing the paper piece image data 42 to an appropriate position of the reference data 50 according to the shape. Just do it. In addition, the reference data 50 used for the preparation of the comparison bill image data 60 is subjected to the same density correction as that of the paper image data 42 as described above. In this embodiment, since the same density correction is performed on the librarian data 70 stored in the librarian data unit 20, similar to the paper-sized image data 42, the image data of the reference data 50 and the librarian data 70 May be considered to match. That is, in comparison banknote image data 60, parts other than the paper-stack image data 42 (part of the reference data 50) are the same as the librarian data 70. FIG.

For this reason, the pattern matching of the banknote image data 60 for comparison and the librarian data 70 is substantially the same with the part corresponding to the piece of paper image data 42 of the piece of paper image data 42 and the librarian data 70. Pattern matching. In this case, if the paper money image data 42 is the comparison banknote image data 60 affixed to the original place (original position of an original banknote) on the reference data 50, the comparison banknote image data 60 will It is almost coincident with the reference data 50. Therefore, by comparing the banknote image data 60 and the librarian data 70 for comparison with the same gold species and detecting the reference data 50 with the minimum distance, the gold species of the reference data 50 are subject to differentiation. It can be assumed that it is a kind of paper. In this case, a threshold value may be set for the distance, and the gold type of the paper may be specified as described above only when the minimum distance obtained as a result of the pattern matching is equal to or less than the threshold value.

As described above, in the present embodiment, pattern comparison of the banknote image data 60 for comparison is performed with the librarian data 70 of each type of gold, thereby specifying the type of gold having the highest distance (similarity), and the specified type of gold. I discern it as a golden bell of this paper.

3 is a flowchart for explaining a paper kind estimation process of the paper sheet by the paper sheet discriminating device 10.

First, the piece of paper image data 41 is density-corrected (step S11), and then the piece of paper image data 41 is meandering-corrected (step S12). Subsequently, the piece of paper image data 41 is affixed on the corner reference on the reference data 50 of the first selected paper type (step S13). In this corner affixation, it is decided beforehand which corner to affix. In addition, the gold type of the reference data 50 initially selected is also determined beforehand.

Then, first, the comparison bill image data 60 and the librarian data 70 (firstly, the librarian data 70 of a predetermined gold type) are pattern-matched in the front image (step S14), and then the both data are laterally lateral. Pattern matching is performed on the line image (step S15). Subsequently, both data are pattern matched in the vertical line image (step S16), and finally, both data are pattern matched in the oblique line image (step S17). The results of these four types of pattern matching are stored in a memory (not shown).

As described above, after performing four types of pattern matching on the comparison banknote image data 60 in which the piece of paper image data 42 is affixed to any corner, the four kinds of pattern matching are performed by any one gold type. Is determined for all comparison banknote image data 60 in step S18, and if not yet, the reference data 50 to which the paper image data 42 is to be pasted (the reference data 50 of the gold species currently being selected) Is selected next (step S19), and the process returns to step S13. After step S13, the said pattern matching process is performed with respect to the banknote image data 60 for comparison which stuck the piece of paper image data 42 to the corner selected at this time of the said reference data 50. FIG.

In this way, the pattern matching between the librarian data 70 and the comparison banknote image data 60 is performed on any one of the gold species at each corner of all the comparison banknote image data 60 (reference data 50). The individual piece of image data 60 to which the piece of paper image data 42 is affixed is performed. When the pattern matching with all comparison bill image data ends, it is determined whether the pattern matching with all the librarian data 70 has ended (step S20), and if not, the librarian data 70 of the gold species to be pattern matched next. ) Is selected (step S21), and the process returns to step S13. In step S13, the banknote image data 60 for comparison of the gold paper is created using the reference data 50 of the gold paper newly selected.

In this way, the pattern of the comparison bill image data 60 (the number of the comparison bill image data 60 equal to the number of corners to which the piece image data 42 is affixed) and the librarian data 70 with respect to all the gold grades. Matching is performed. When the pattern matching is completed, the gold species having the highest similarity is estimated to be the paper species of the paper, and the estimation result is stored in the storage unit 21 (step S22). In step S22, in addition to the paper type of the sheet, the position information (corner position) on the banknote of the sheet may also be stored. The position information of this finger can be obtained by storing the latest corner information selected in step S19 in a memory (not shown) or the like.

By the way, as the method of pattern matching, well-known methods, such as the method of using the minimum distance between patterns, can be used.

In addition, in the paper type identification process of the piece of FIG. 3,

(1) Pattern matching in full screen image

(2) Pattern Matching in Transverse Line Images

(3) Pattern matching in vertical line image

(4) Pattern Matching on Inclined Line Images

Although four types of pattern matching are performed and the best similarity (or minimum distance) obtained by these pattern matching is used as an index for identifying the species, only one of the above (1) to (4) or (1) to A combination of one to three types of pattern matching may be selected from (4) to identify the paper species of the paper.

FIG. 4 is a flowchart for explaining details of the concentration correction process in step S11 of FIG. 3.

First, the area S of the white edge portion of the paper sheet in the piece of paper image data 41 is calculated (step S31), and then the transmitted light addition value A of the white edge portion is calculated (step S32).

Subsequently, the transmitted light addition value B (= A / S) per unit area of the white edge portion is calculated (step S33), and then the coefficient C (for H to be 80H (H is a symbol representing a hexadecimal value)). = 80 H / B) (step S34). Then, all pixels of the piece of paper image data 41 are multiplied by C to density-correct the piece of paper image data 41 (step S35).

5 shows a data structure of librarian data 70.

As shown in Fig. 5A, the librarian data 70 is composed of L × N pixels 71 in vertical L columns and horizontal N rows. As shown in Fig. 5B, the pixel 71 has a 16-bit configuration, and the gray level average value (concentration average value) is stored in the upper 8 bits, and the standard deviation sigma is stored in the lower 8 bits. This standard deviation σ is the standard deviation of the gradation value of the pixel 71 obtained from the sample data for producing the librarian data 70.

6 is a diagram showing a data structure of the comparison bill image data 60.

As shown in FIG. 6A, the bill image data 60 for comparison is also composed of L × N pixels 61 in vertical L columns and horizontal N rows similarly to the librarian data 70. As shown in Fig. 6B, the pixel 61 has an 8-bit configuration, and the gradation value (concentration value) is stored in this 8-bit.

In this embodiment, as shown schematically in FIG. 7, in the pattern matching between the comparison bill image data 60 and the librarian data 70, the pixels of the same positions are expressed by the following equation (1). The Euclidean distance d _ij between them is calculated.

here,

P _ij : _{Gray level value} of the pixel of the comparison bill image data 60

P ' _ij : Gray level value of pixel of librarian data 70

to be.

Next, the front surface pattern matching process of step S14 of FIG. 3 will be described.

8 is a schematic conceptual diagram of the method of the front image pattern matching.

In Fig. 8, as indicated by the arrow in the right direction, the Euclidean distance d _ij is obtained by the above equation (1) from the first row to the Nth row, and all pixels represented by the following equation (2) are obtained. Find the Euclidean distance D _all .

FIG. 9 is a flowchart for explaining in detail the front image pattern matching process of step S14 of FIG. In Fig. 9, i and j are variables representing column numbers and row numbers of pixels, respectively. Also, D _EA is a variable for obtaining Euclidean distance in front image pattern matching.

First, 1 is assigned to the variables i and j. Further, 0 is substituted into the variable D _EA (step S41).

Next, Euclidean distance d _ij is calculated by Equation 1 (step S42), and this Euclidean distance d _ij is added to the variable D _EA (step S43). Subsequently, it is determined whether i is equal to L (step S44). If it is not the same, the value of i is incremented by one (step S45), and the process returns to step S42.

As described above, the processing in steps S42 to S45 is repeated until i is determined to be equal to L in step S44, and the total sum of Euclidean distances of all pixels in row i is obtained.

If i = L is determined in step S44, it is then determined whether j = N (step S46). If j is not N, the value of j is incremented by one (step S47), and the flow returns to step S42. In this way, the process of steps S42 to S47 is repeated until j becomes equal to N, and the total sum of Euclidean distances of all the pixels in the image is obtained. This total is stored in D _EA . If it is determined in step S46 that j is equal to N, the process ends.

FIG. 10 is a conceptual diagram schematically showing a pattern matching process in the horizontal line image in step S15 of FIG. 3.

As shown in FIG. 10, in this pattern matching process, an arbitrary row x of the comparison bill image data 60 is selected, all the pixels in this row are scanned, and all the pixels in the scanning line 81 are scanned. The sum of the Euclidean distances d _ij with the pixels of the librarian data 70 is obtained. However, the row x needs to be a row in which the pixels of the piece of paper image data 42 are included. Preferably, the row x is a row x having the maximum number of pixels of the piece of paper image data 42.

FIG. 11 is a flowchart for explaining details of the pattern matching processing in the horizontal line image in step S15 of FIG. 3. In Fig. 11, the variables i and j are variables in which the column number and the row number of the pixel are set, respectively. Also, D _EY is a variable for obtaining the sum of Euclidean distances of all pixels in the selected row y.

First, i is initially set to 1, and the selected row number y is set to j. Further, D _EY is initially set to 0 (step S51). Then, the Euclidean distance d _ij is calculated by Equation 1 (step S52), and the Euclidean distance d _ij is added to the variable D _EY (step S53).

Next, it is determined whether i is equal to L (step S54), otherwise, the value of i is incremented by one (step S55), and the process returns to step S52.

In this way, the process of steps S52 to S54 is repeated until the value of i becomes equal to L, and the total sum of Euclidean distances of all the pixels in the selected row y is obtained. This total is stored in the variable D _EY . If it is determined in step S54 that i is equal to L, the process ends.

FIG. 12 is a diagram schematically illustrating the concept of pattern matching processing in the vertical line image in step S16 of FIG. 3.

In the pattern matching processing in this vertical line image, an arbitrary column x in the comparison bill image data 60 is selected, all pixels in the column x are scanned, and all the pixels in the scanning line 91 are scanned. Calculate the sum of Euclidean distances. In this case, however, the column x to be selected is a column including the pixels of the piece of paper image data 42. Preferably, the column x having the maximum number of pixels is selected.

FIG. 13 is a flowchart for explaining the details of the pattern matching processing in the vertical line image in step S16 of FIG. In Fig. 13, i and j are variables in which the column number and the row number of the image are set, respectively.

First, the selected column number x is set to the variable i, and the variable j is initially set to 1. Further, the variable D _ET for initializing the sum of Euclidean distances of all the pixels in the selected column x is initially set to 0 (step S61).

Next, Euclidean distance d _ij is calculated by Equation 1 (step S62), and this Euclidean distance d _ij is added to the variable D _ET (step S63). If j is equal to N (step S64), if not equal, the value of j is incremented by one (step S65), and the process returns to step S62.

In this way, the process of steps S62 to S65 is repeated until the value of j becomes equal to N, and the total sum of Euclidean distances of all the pixels in the selected column x is obtained. This sum is stored in the variable D _ET . If it is determined in step S64 that j is equal to N, the process ends.

14 and 15 are diagrams schematically showing the concept of pattern matching processing in the oblique line image in step S17 of FIG.

Fig. 14 selects a column of an arbitrary number x, scans the pixel in the left oblique direction at a predetermined angle (for example, 45 degrees) from the selected column, and Euclid in all the pixels of the scanned oblique line. The pattern matching process by the left inclination line image which calculates the sum total of distance is shown. In FIG. 14, the scanning lines 101 and 102 in the case where the column of the number x1 of the comparison banknote image data 60 is selected, and the case where the number x2 is selected are shown, respectively. In this technique, it is preferable to select the column number so that the number of pixels of the finger image data 42 on the scanning line is maximum.

15 is a schematic conceptual diagram of the pattern matching processing in the inclined line image in the case where the scanning direction is in the right oblique direction.

FIG. 15 shows the scanning lines 111 and 112 in the case where the column number of the pixel to be scanned initially is set to x3 and x4, respectively. Also in the pattern matching of the oblique line image in the right oblique direction, it is preferable to first select a column of pixels to be scanned so that the line which scans the largest number of pixels of the piece of paper image data 42 becomes the scanning line.

FIG. 16 is a view for explaining the pattern matching processing in the inclination image line in the left oblique direction shown in FIG. 14 in detail. In Fig. 16, the variables i and j are variables in which the column number and the row number of the pixel are set, respectively. In addition, D _ENR is a variable used to calculate the sum of Euclidean distances of all pixels in the scan line in the left oblique direction.

First, the column number of the pixel to be initially scanned is set to the variable i, and 1 is set to the variable j as the row number of the pixel to be initially scanned. Further, the variable D _ENR is initially set to 0 (step S71).

Next, the above equation (1) is calculated to obtain the Euclidean distance d _ij (step S72), and the Euclidean distance d _ij is added to the variable D _ENR (step S73). Then, it is determined whether i = 1 or j = N (step S74). If neither i or j satisfies the equation, the value of i is incremented by one (step S75), and the value of j is 1. It increments (step S75) and returns to step S72.

In this way, the process in steps S72 to S76 is repeated until it is determined that i = 1 or j = N is satisfied in step S74, and the line in the right oblique direction in which scanning is started from the pixels in the first row x The sum of Euclidean distances of all the pixels in the image is obtained, and the sum is stored in the variable D _ENR . If it is determined in step S74 that i = 1 or j = N, the process ends.

FIG. 17 is a flowchart for explaining in detail the pattern matching processing in the line image in the left oblique direction shown in FIG. 15.

First, the row numbers (= x) and column numbers (= 1) of the first scanning pixel are set in the variables i and j, respectively. Further, the variable D _ENL is initially set to O to obtain the sum of the Euclidean distances of all the pixels in the scan line image in the left oblique direction (step S81).

Next, Euclidean distance d _ij is calculated using Equation 1 above (step S82), and this Euclidean distance d _ij is added to the variable D _ENL (step S83). If it is determined whether i = L or j = N (step S84), and neither of the equations holds, i is incremented by one (step S85) and j is incremented by one (step S86). Return to step S82.

In this manner, the process of steps S82 to S86 is repeated until it is determined that either i = L or j = N is satisfied in step S84, and the Euclidean distance of all pixels in the selected left oblique line image Find the sum of and store the value in the variable D _ENL . If it is determined in step S84 that either i = L or j = N holds, the process ends.

Further, in the first embodiment, pattern matching is performed for data of one vertical line, one horizontal line, or one inclined line as data of a part of the comparison bill image data 60, but a plurality of lines may be used instead of one line. . In addition, the partial data of the comparison banknote image data 60 used by the pattern matching of this invention is not limited to a line, If it is partial data containing the piece paper image data 42 of the comparison banknote image data 60 It may be a shape other than a line.

(2nd Example)

FIG. 18 is a block diagram showing the system configuration of the finger discriminating apparatus 200 according to the second embodiment of the present invention. 18, the same code | symbol is attached | subjected to the component same as the component of the finger discrimination apparatus 10 of 1st Example shown in FIG. 1, and description of these components is abbreviate | omitted.

The difference in the structure of the piece discrimination apparatus 200 and the piece discrimination apparatus 10 uses the image processing part 17A for the piece discrimination function by the librarian comparison part 18, the reference data part 19, and the librarian data part 20. ), The reference data unit 19 and the neuro calculating unit 201 are replaced. The central processing unit 11A analyzes the output of the neuro calculating unit 201, discriminates the paper type of the paper, and controls the image processing unit 17A.

As described above, the image processing unit 17A performs normalization and concentration correction on the piece of paper image data 42, and then, similarly to the librarian comparison unit 18, the piece of paper image data 42 and the reference data unit 19. From the reference data 50 of each gold type stored in the above), the banknote image data 60 for comparison for each gold type is created. Then, the comparison banknote image data 60 is subjected to a reduction process, and all pixels or a part of pixels of the reduced image obtained thereby are output to the neuro calculator 201.

The neuro calculating unit 201 is a neural network, which receives the reduced image created by the image processing unit 17A as an input pattern, and outputs the paper type and its direction (position on the original banknote) of the paper sheet as an output pattern.

In addition, in the second embodiment, the neuro arithmetic unit 201 performs the neuro arithmetic operation by the reduced image, but it is not necessarily the reduced image. Whether or not to use the reduced image as the input pattern may be appropriately determined depending on the scale of the neuro calculating unit 201, the calculation processing time thereof, and the like.

19 is a diagram illustrating the configuration of the neuro calculator 201.

The neuro calculator 201 is a feedforward-coupled neural network composed of an input layer 211, an intermediate layer 212, and an output layer 213, and uses error back-propagation learning as a learning algorithm.

The input layer 211 has input signals I ₁ , I ₂ ,... As input patterns. , I _p is input from the image processing unit 17A. These input signals I _i are gradation values (density) of pixels of the reduced image. Input layer 211 is S ₁ , S ₂ ,. , P _p units (neurons), and the input signal I _i is input to the unit S _i (i = 1, 2, ..., p).

The intermediate layer 212 is made of A ₁ , A ₂ ,. , A _m m units. Each unit A _j (J = 1, 2, ..., m) is combined with all units S _i (i = 1, 2, ..., n) of the input layer 211.

The output layer 213 is R ₁ , R ₂ ,... , _Q q units of R _q , and each unit R _k (k = 1, 2, ..., q) is combined with all units A _j of the intermediate layer 212. The unit R _k outputs an output signal O _k . This output signal O _k is a binary signal of 0 or 1, and only one unit among q units R _k outputs an output signal of 1, and the other unit outputs O.

Each of the q units R _k corresponds to a specific gold grade and the position (direction) of the finger, respectively, and outputs according to the input patterns I (I ₁ , I ₂ ,..., I _p ) input to the input layer 211. Output the signal O _k .

20 is a diagram illustrating a model example of both the unit A _j of the intermediate layer 212 and the unit R _k of the output layer 213.

The unit 220 (unit j) shown in FIG. 20 is a quasi-linear element model in which the output from each unit i of the front layer is given as an input x _i , and has a threshold value θ _j . Here, the internal state u _j of the unit 220 (unit j) is

W _ij : weight of the coupling between unit i and unit j

Calculated as

The output of the unit 220 is a function f (u _j ) which takes this value u _j as a variable, for example, a sigmoid function,

It becomes

Here, T is a parameter called temperature.

21 is a diagram illustrating an input pattern output from the image processing unit 17A to the neuro calculating unit 201.

The image data shown on the right side of FIG. 21 is a comparison bill image data 301 created by the image processing unit 17A, and the size thereof is 76 lines x 160 pixels. The image processing unit 17A reduces the comparison bill image data 301 and creates reduced image data 302 of 10 lines x 22 pixels as shown on the left side in FIG.

In the neuro calculator 201, one pixel of the reduced image data 302 is input as one input signal. The neuro calculating unit 201 carries out four types of pattern recognition processing: (1) front image pattern recognition, (2) horizontal line image pattern recognition, (3) vertical line image pattern recognition, and (4) inclined line image pattern recognition. It is performed by operation. The image processing unit 17A outputs an input signal (input pattern) required for pattern recognition in each of the above (1) to (4) to the neuro calculating unit 201.

In the second embodiment, the configuration of the neural network 201 of the neuro calculator 201 changes as follows in accordance with the pattern recognition (pattern matching) described above (1) to (4).

(1) Front image pattern recognition

Input signal: I ₁ ~ I ₂₂₀

  Number of units in the input layer 211: 220

  Number of units in the middle layer 212: 10

  Number of units in the output layer 213: 64

Output signal: 0 _{1-0 64}

(2) horizontal line image pattern recognition

Input signal: I _{1 to} I ₂₂

  Number of units in the input layer 211: 22

  Number of units in the middle layer 212: 2

  Number of units in the output layer 213: 64

Output signal: 0 _{1-0 64}

(3) vertical line image pattern recognition

Input signal: I ₁ ~ I ₁₀

  Number of units in the input layer 211: 10

  Number of units in the middle layer 212: 3

  Number of units in the output layer 213: 64

Output signal: 0 _{1-0 64}

(4) (Right Slope Direction or Left Slope Direction) Slant Line Image Pattern Recognition

Input signal: (I _{1 to} I ₂ ), (I _{1 to} I ₃ ),. Or (I _{1 to} I ₁₀ )

  Number of units in the input layer 211: 2-10

  Number of units in the middle layer 212: 2

  Number of units in the output layer 213: 64

Output signal: 0 _{1-0 64}

As described above, the number of units of the input layer 211 is equal to the number (pixel number) of the input signal I _i from the image processing unit 17A. In addition, in this embodiment, the gradation value (density) of the pixel is 8 bits, so the range of each input signal I ₁ value is in hexadecimal representation and becomes OOH to FFH.

In the present embodiment, since 16 types of paper types of paper sheets and positions of paper sheets (positions on original banknotes) are assumed in four directions (the upper left corner, the upper right corner, the lower left corner, the lower left corner, and the lower corner of the banknote, based on the corner), 46 kinds of recognition are possible by the combination of the paper type and the position (direction) of the paper. These 46 kinds of recognition, respectively, and corresponds with each of the output signal O _k (k = 1~64) of the output layer (213), the output signal corresponding to the direction (position) of the denomination, of the recognized jipyeon jipyeon O _k Becomes 1.

22 to 26 are diagrams for explaining input signals in pattern recognition of (1) to (4).

If the front image pattern recognition, as shown in Figure 22, around the reduced gradation (concentration) of the pixel of the image data 302 is input to the input signal I ₁ ~I ₂₂₀ as a neural network (201). In the case of horizontal line image recognition, as illustrated in FIG. 23, the gray level value (density) of all pixels in an arbitrary column (diagonal portion) in the reduced image data 302 is determined by the input signals I _{1 to} I. ₂₂ as neural network 201. In addition, in the case of vertical line image pattern recognition, as shown in FIG. 24, the gradation value (density) of all pixels in an arbitrary row (diagonal portion) in the reduced image data 302 is determined by the input signals I ₁ to ₁ . It is input to neural network 201 as I ₁₀ .

In the inclination line image pattern recognition, in the case of the left inclination line image pattern recognition, as shown in FIG. 25, two, starting from the pixel of the arbitrary column (net part) in the 1st row of the reduced image data 302, the tone of the left oblique lines of the respective pixels including the pixel to one to 10 (concentration) of the input signal I ₁ ~I _2, ... Input to the neural network 201 as input signals I _{1 to} I ₁₀ . In addition, in the case of right inclination line image pattern recognition, as shown in FIG. 26, the pixel from 2 pixels to 10 pixels starting from the pixel (net part) of arbitrary column in the 1st row of the reduced image data 302 is shown. the gradation of each pixel with (concentration) as an input signal I ₁ is an input signal I ₁ ~I ₁₀ from ~I ₂ is input to the neural network (201).

FIG. 27 is a flowchart for explaining the operation of the paper sheet discriminating apparatus 200 for discriminating the paper type of the paper sheet by pattern recognition by the front image. In addition, in FIG. 27, the same process number is attached | subjected to the process of performing the same process as the process of the flowchart of FIG.

The image processing unit 17A performs a density correction process on the piece of paper image data read from the image storage unit 18 (step S11), and further meanders and normalizes the piece of paper image data 42 (step S12). The image processing unit 17A then attaches the paper-stack image data 42 to the first selected corner on the first selected librarian data 70 and creates the comparison bill image data 301 (step S13). ). The gold type and the corner (position) to be selected first are determined in advance.

Next, the image processing unit 17A creates the reduced image data 302 of the comparison banknote image data 301 (step S104), and outputs the front image data (all pixels) to the neuro calculating unit 201. (Step S105).

The neuro calculating section 201 receives the front image data as an input pattern, performs a neuro calculation, and outputs an output pattern corresponding to the input pattern (step S106).

The central processing unit 11A receives the output pattern of the neuro calculating unit 201, and determines whether the golden seed is estimated by the neuro calculating unit 201 (step S107). Then, if the paper type is not estimated, it is determined whether the pasting of the paper-stack image data 42 in all the corners has ended with respect to the reference data 50 of the gold type currently selected (step S108), and still the pasting in all the corners. If is not finished, the flow returns to step S13.

On the other hand, when the central processing unit 11 determines that the golden bell is estimated in step S107, the central bell 11 stores the golden bell in the storage unit 21 (step S111), and the processing ends.

In this way, the bill image data 301 for comparison is created while affixing the paper-stack image data 42 to each corner on the reference data 50 sequentially, and the neuro calculating part 201 is carried out with respect to the reduced image data 302. Estimation of the gold species is performed by this. Then, in step S108, the paper type estimation by the neuro calculator 201 of the comparison bill image data 301 when the paper money image data 42 is affixed to all corners on the reference data 50 of the gold type currently selected. When it is determined that this has been completed, it is next determined whether the money type estimation by the neuro calculating unit 201 is finished on the reduced image data 302 of the comparison paper money image data 301 of all the paper types (step S109). Then, if it is determined that the paper type estimation is not finished with respect to the reduced image data 302 of the comparison paper money image data 301 of all the paper types, the next paper type is selected (step S110), and the process returns to step S13.

In this way, the banknote image data 60 for comparison of all the corner | corners is produced about each gold type, and each piece of these reduced image data 302 is estimated with the kind of money by the front image pattern recognition by the neuro calculating part 201. Then, the paper type and direction of the paper are discriminated, and the result of the discrimination is stored in the storage unit 21. In addition, when the kind of money by the neuro calculating part 201 is not estimated with respect to the banknote image data 60 for comparison of all the corners of all kinds of money, in step S109, the kind of money estimation by the neuro calculating part 201 is performed with respect to all the kinds of money. It is determined that it has finished and the process ends.

FIG. 28 is a flowchart for explaining the operation of the paper sheet discriminating apparatus 200 for discriminating the paper type of the paper sheet by pattern recognition by the horizontal line image. In FIG. 28, the same process numbers as those in FIG. 27 are assigned the same step numbers, and descriptions of these steps are omitted.

In the pattern recognition by the neuron operation based on the horizontal line image, when the image processing unit 17A creates the reduced image data 302 (step S104), an arbitrary horizontal line of the reduced image data 302 (fragment image data ( The image data of the line (42)) is output to the neuro calculator 201 (step S125).

The neuro calculating unit 201 receives the horizontal line image data as an input pattern, and outputs an output pattern corresponding to the input pattern to the central processing unit 11 (step S126). The central processing unit 11 discriminates the kind and direction of the paper pieces by analyzing the output pattern. Processing other than the above is similar to pattern recognition in the front image of FIG.

FIG. 29 is a flowchart for explaining the operation of the paper sheet discriminating apparatus 200 for discriminating the paper type of the paper sheet by pattern recognition by the vertical line image. In FIG. 29, the same process numbers are given to the same processing steps as those of FIG. 28, and descriptions of these steps are omitted.

When the image processing unit 17A creates the reduced image data 302 (step S104), the neuro calculator calculates the image data of any vertical line (the line including the reduced image data 302) of the reduced image data 302. Output to step 201 (step S135). The neuro calculating section 201 receives this vertical line image data as an input pattern, and outputs an output pattern corresponding to the input pattern to the central processing section 11 (step S136). The central processing unit 11 analyzes the output pattern to determine whether the paper type and direction of the paper pieces are estimated. If they are estimated, the central processing unit 11 stores the gold type and direction of the estimated banknote in the storage unit 21.

FIG. 30 is a flowchart for explaining the operation of the paper sheet discriminating apparatus 200 for discriminating the paper type of paper sheets by pattern recognition of the inclined line image (left inclined line image or right inclined line image). In FIG. 30, the same process number is attached | subjected to the process of performing the same process as the process of FIG. 27, and description of these steps is abbreviate | omitted.

When the image processing unit 17A creates the reduced image data 302 (step S104), the neuro calculator calculates the image data of any inclined line (the line including the deflection image data 42) of the reduced image data 302. Output to step 201 (step S145). The neuro calculator 201 receives the inclined line image data as an input pattern, and outputs the corresponding output pattern to the central processing unit 11 (step S146). The central processing unit 11 analyzes the output pattern to discriminate the paper type and direction of the paper.

In the second embodiment, when data of a part of the reduced image data 302 is used as an input pattern, image data of one vertical line, one horizontal line, or one inclined line is used as the input pattern. The invention is not limited to these, and for example, a plurality of lines in the longitudinal direction, the transverse direction, or the inclined direction may be used as the input pattern. The partial data of the reduced image data 302 used as the input pattern does not have to be a line shape, and may be any shape as long as the data includes the piece image data 42.

In addition, both the estimation of the kind of money by the librarian comparison unit 18 of the first embodiment and the estimation of the kind of money by the neuro calculation unit 201 of the second embodiment may be carried out to build a system in which the accuracy of the money type estimation is increased.

As described above, according to the present invention, the image data of a piece of paper is affixed to the reference data of each gold type (the image data of the true winding), and the banknote image data for comparison is created for each gold type, and the corresponding banknote image data is corresponded. By pattern matching with the librarian data of the denomination, it is possible to identify the denomination of the paper.

In addition, since the present invention does not need to change the hardware of the existing apparatus, and can be coped with only the change of software, the introduction cost is low. In addition, in recent years, when it is mounted in the banknote processing apparatus used for the operation and the 24-hour operation which is increasing in unmanned stores, even if the situation which a banknote breaks inside the apparatus arises, the balance of a banknote can always be determined from a paper. As it becomes possible, the use value is extremely large.

The present invention can be applied to an overall unmanned machine that handles deposits and withdrawals of bills that will increase in the future with the reduction of manpower, such as ATMs, cash dispensers (Cash Dispenser), vending machines, station ticket machines, etc. As it can, use demand is big.

Claims (20)

Reference data storage means for storing reference data which is image data of a true bill, for at least one gold type; Librarian data storage means for storing librarian data for discrimination of true bills by the same number of gold species as the reference data; Sensor means for acquiring image data of a paper sheet, which is an intercept of paper sheets; The paper money image data obtained by the sensor means is synthesized with the reference data of each gold species stored in the reference data storage means to create comparison paper image data for each gold species, and to purchase the comparison paper money image data. Librarian comparison means for discriminating the gold species of the paper by pattern matching with the same kind of librarian data stored in the data storage means Finger paper discrimination apparatus comprising a. The method of claim 1, And said librarian comparison means performs pattern matching on said comparison banknote image data and said librarian data with respect to all images. The method of claim 1, The librarian comparison means performs pattern matching on the comparison bill image data and a part of the librarian data. The method of claim 3, The librarian comparison means performs pattern matching between the comparison bill image data and the librarian data with respect to a horizontal line. The method of claim 3, The librarian comparison means performs pattern matching between the comparison bill image data and the librarian data with respect to a vertical line. The method of claim 3, The librarian comparison means performs pattern matching between the comparison banknote image data and the librarian data with respect to an inclined line. The method of claim 1, The librarian comparing means sequentially performs pattern matching of all images, pattern matching of lateral lines, pattern matching of vertical lines, and pattern matching of inclined lines with respect to the comparison bill image data and the librarian data. The discriminating apparatus of the finger piece, characterized in that it discriminates the gold species of the paper piece from the result. The method of claim 1, And said librarian comparing means creates at least one bill paper image data for comparison by affixing said sheet of paper image data at a corner of said reference data. The method of claim 1, The said librarian comparison means determines the position which affixes the said paper money image data to the said reference data at the time of creating the said banknote image data for comparison, according to the shape of the said paper money image data for comparison. The method of claim 1, Said librarian comparison means distinguishes the position on the banknote of a paper sheet in addition to the paper type of paper sheet. Reference data storage means for storing reference data which is image data of a true bill, for at least one gold type; Sensor means for acquiring image data of a paper sheet, which is an intercept of paper sheets; Image processing means for synthesizing the piece of paper image data obtained by the sensor means with the reference data of each gold species stored in the reference data storage means to create comparison bill image data for each gold species; Neuro arithmetic means which receives the said banknote image data for comparison as an input pattern, performs a neuro calculation with respect to the input pattern, and outputs the output pattern which shows the gold type of the said paper. Finger paper discrimination apparatus comprising a. The method of claim 11, And the input pattern is a part of the comparison bill image data. The method of claim 12, And said input pattern is image data of a vertical line of said comparison bill image data. The method of claim 12, And said input pattern is image data of a horizontal line of said banknote image data for comparison. The method of claim 12, And said input pattern is image data of an inclined line of said comparison bill image data. The method of claim 11, And said neuro calculating means outputs, as an output pattern, information relating to the position of the paper money on the banknote in addition to the paper type of the paper money. delete delete delete delete

Images