JPH0573753A - Sheet paper recognition processing method - Google Patents

Abstract

PURPOSE:To provide the method which is strong against transverse deviation, skew, or the like of sheet paper and quickly recognizes paper with a high reliability. CONSTITUTION:The almost all surface of paper 1 is read with plural light sources 3a and 3b different by wavelength to obtain picture signals different by levels. Features of correlations between these signals are extracted to obtain a specific feature picture corresponding to the pattern on the surface of sheet paper 1. Next, the feature picture distribution is obtained with the distance from teh center or the centroid of this feature picture as a parameter based on this feature picture. Thus, the feature picture distribution obtained from the feature picture even is not changed in the case of skew of sheet paper 1. Consequently, sheet paper is recognized without an influence of not only positional devication but also skew.

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 processing method for recognizing the type and authenticity of paper sheets such as bills and securities.

[0002]

2. Description of the Related Art Generally, in a financial institution or the like, an automatic depositing / dispensing machine is installed in order to automate deposit / saving and transfer processing. When a customer inserts a banknote into such an automatic depositing / dispensing machine, the banknote discriminating section therein identifies the type and authenticity of the banknote. In this identification processing, first, an optical or magnetic pattern is read along one or a plurality of lines parallel to the banknote transport direction. The optical pattern has certain characteristics for each denomination due to the visual pattern of the bill. The magnetic pattern also has the same characteristics due to the magnetic distribution of ink on the bill. The detected pattern, which is an analog signal, is compared with a predetermined threshold value and binarized at many points on the 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 the dictionary data for all the detected lines, the type and authenticity of the bill can be identified.

[0003]

By the way, in the above-mentioned conventional technique, usually, the characteristics of a partial region along a specific line on a banknote are extracted, and the pattern detected from the region is identified. Only doing. Therefore, when a banknote is forged only in a portion having a very small area other than the detection area, the banknote may be identified as a genuine banknote. That is, there is a possibility that such forgery cannot be detected when the banknotes are partially stuck together. Further, in order to increase the reliability of bill identification, there is also a method of detecting various parameters such as external dimensions and thickness of bills and identifying the type and authenticity of bills from various angles. However, in such a method, if even one of these parameters is out of the standard range, the bill is determined to be a counterfeit bill.
Therefore, the rate of determination as a fake bill increases, which is a problem in practical use. Of course, this kind of problem is not limited to bill appraisal,
The same applies to the appraisal of securities and various types of paper sheets.

On the other hand, in order to speed up the processing of the automatic depositing / dispensing machine, it is necessary to run the paper sheets on the conveying path at high speed. However, if the paper sheets are made to travel at a too high speed, it becomes difficult for the paper sheets to follow the predetermined guideline, and the banknotes are likely to be laterally offset or skewed. Therefore, the reading of the paper sheets is often performed deviating from the above-described specific line, or is performed in an oblique direction without following the specific line. As a result, the recognition rate of paper sheets has decreased. To solve these problems,
First, it is preferable that the widest possible portion of the paper sheet is read at a high resolution and used as a reference for identification so that forgery or bonding of a small area can be detected. In addition, it is preferable to identify the type and authenticity by using one criterion having the highest reliability in order to improve the recognition rate.

Further, in the method of reading a part of a paper sheet, if there is a lateral deviation or skew in the conveyance of the paper sheet, the detection data may fluctuate and erroneous recognition may occur. It is desired that the identification process is not affected even when the paper sheet is slightly misaligned or skewed. The present invention has been made in view of the above points, and provides a paper sheet recognition processing method that is strong against lateral misalignment, skewing, and the like, and that can recognize paper sheets quickly with high reliability. The purpose is that.

[0006]

The first method of the present invention comprises:
The paper surface of the paper sheet is alternately irradiated by using a plurality of light sources each having a different wavelength, and for each read pixel, the characteristics of the correlation of the reflected light intensity with respect to each wavelength are extracted and digitized, and For each of the read pixels, the digitized data is detected in the output image area formed by the two-dimensional coordinates to detect the end coordinates on the main scanning direction coordinate and the sub scanning direction coordinate of the paper sheet image. , Detecting the center point coordinates of the paper sheet image from the edge coordinates, calculating the distance between the coordinates of the pixel represented as the feature and the center point coordinate from the characteristic image, and determining the distance magnitude. The output of the feature image that applies to the distance is integrated into the sequentially assigned one-dimensional sequence to extract the feature image distribution, and the shape of the feature image distribution is compared with a previously prepared reference distribution shape, Specializing in recognition It is an.

In the second method, the paper surface of the paper sheet is alternately irradiated with a plurality of light sources having different wavelengths, and the correlation of the reflected light intensity with respect to each wavelength is read for each read pixel. The features are extracted and digitized, and the feature image formed by all the digitized data is projected on the main scanning direction coordinates and the sub scanning direction coordinates to obtain respective pixel number distributions, While obtaining the barycentric coordinates of the pixel number distribution in the scanning direction, the barycentric coordinates of the pixel number distribution in the sub-scanning direction are obtained, and the barycentric coordinates in the main scanning direction,
The total barycentric coordinates of the characteristic image are combined with the barycentric coordinates in the sub-scanning direction, the distance between the coordinates of each pixel on the characteristic image and the total barycentric coordinates is obtained, and the distances are given in order of magnitude 1. Detecting the characteristic image distribution by integrating the output of the characteristic image corresponding to the distance in the dimensional series, and comparing the shape of the characteristic image distribution with a reference distribution shape prepared in advance to recognize the paper sheet. It is characterized by.

[0008]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a configuration diagram of an embodiment of a paper sheet recognition processing method of the present invention. In the figure, this device is
It is composed of a reading unit 10 and a recognition unit 20. The reading unit 10 optically reads the paper surface of the paper sheet 1 set in the apparatus and outputs the image signal thereof. The recognition unit 20 extracts a characteristic image as a recognition target based on the image signal output from the reading unit 10 and compares it with data prepared in advance to recognize the paper sheet.

The reading unit 10 stores bills, securities, etc.
The paper sheet 1 requiring identification is set. An image line sensor 2 is arranged above the paper sheet 1 to read the image signal. The image line sensor 2 has a width sufficiently longer than the width of the paper sheet 1, and its resolution is, for example, 400 DPI (reading density of 400 dots per inch).

Further, above the paper sheet 1, parallel to the image line sensor 2, line-shaped light sources 3a and 3b having wavelengths greatly different from each other are arranged. In the case of this embodiment, these light sources 3a and 3b are arranged substantially symmetrically to the left and right of the image line sensor 2. As the light sources having widely different wavelengths, for example, a white light source to which a red filter and a blue filter are attached, or a light source dedicated to infrared light and a light source dedicated to ultraviolet light are used. In addition, the paper sheet 1 is assumed to be conveyed in the direction shown by an arrow 4 in the figure, and the light sources 3a and 3b irradiate the alternative paper sheet 1 while the paper sheet 1 is conveyed, The image line sensor 2 is configured to read the reflected light having different properties at substantially the same position.

Further, the reading area is much larger than the area of the sheet 1 to be set, and the sheet 1 to be set 1
Even if the paper is laterally offset or skewed, it can be sufficiently accommodated in the reading area. The signal output of the image line sensor 2 is always set to 0 in the parts other than the paper sheets. That is, the portions other than the region indicated by the paper sheet 1 are configured so that the irradiation light from the light sources 3a and 3b is not reflected. The recognition unit 20 includes a feature image extraction processing unit 21.
And a center point coordinate detection processing section 22, a characteristic image distribution extraction processing section 23, and a comparison / discrimination processing section 24. Specifically, the recognition unit 20 includes a memory that receives and stores an image signal, a processor that performs arithmetic processing on the image signal, and the like.

The reading section 10 of FIG. 1 operates as follows. First, only the light source 3a is turned on, the image line sensor 2 receives the reflected light of the paper sheet 1, and the multi-value digital signal Da for one line is obtained. Next, the light source 3a is turned off and the light source 3b is turned on, and the image line sensor 2 receives the reflected light at the same location on the paper sheet 1 and obtains a similar multilevel digital signal Db. After that, the paper sheet 1 is conveyed in the direction of the arrow 4, the light source 3a is turned on again, and the next line is read. This reading line interval is determined by the image line sensor 2
Select the same resolution as. Thus, when the multivalued digital signals Da and Db are input to the recognition unit 20, the recognition unit 20 performs the following operation.

First, the characteristic image extraction processing section 21 determines the signal D
Features of the correlation between a and Db are extracted and digitized. That is, the signal Da has a signal pattern according to the characteristic reflected light intensity with respect to the wavelength of the light source 3a. On the other hand, the signal Db
Is a signal pattern according to the characteristic reflected light intensity with respect to the wavelength of the light source 3b. Therefore, depending on the image of the paper sheet 1, they are classified into a case where they are significantly different and a case where they are not so different. In this embodiment, such a characteristic image is extracted as a binary image.

FIG. 2 is a graph showing the correlation of the reflected light intensity. In the figure, when the signal Da is plotted on the horizontal axis and the signal Db is plotted on the vertical axis, focusing on the reflected light intensity of a specific pixel that has been read, the feature of that pixel is that the region 3
It exists in either 1, 32 or 33. The hatched areas 31 and 32 in the figure are characteristic areas in which the intensity of the reflected light of either one is significantly larger than that of the other. On the other hand, the white area 33 in the figure is an area where there is almost no characteristic of the reflected light intensity at any level. Therefore, the regions 31 and 32 where the features appear from the signals Da and Db
Boundary lines 34 and 35 are drawn in order to distinguish the regions 33 in which the features do not appear. The level defined by the boundary line 34 will be referred to as a threshold level THL1, and the level defined by the boundary line 35 will be referred to as a threshold level THL2. If attention is paid to each pixel, the signal D
The signal Db is determined corresponding to the value of a,
The signal Da can be mapped to the signal Db.

Therefore, the signal Db is the mapping value (function value) of the signal Da, and the threshold levels THL1 and THL1.
HL2 is a functional expression defined by the following expressions (1) and (2). Here, f is a function representing the correlation between the signals Da and Db. Inv f is the inverse function of the function f in which the signal Db and the signal Da are exchanged. THL1: Db = f (Da) (1) THL2: Db = inv f (Da) (2) This function f and the function inv f are as shown in the graph of FIG.
There is a symmetrical positional relationship centered on the diagonal line of the diagram where Da = Db. This f is, for example, the following (3) or
It can be expressed by equation (4). It should be noted that α <1 in the equation (4) is set so that the straight lines on both sides of the straight line Da = Db do not intersect with each other. Db = Da + α (3) where α is an arbitrary integer α <0 Db = αDa (4) where α is an arbitrary real number 1>α> 0 Here, in the following embodiments, for example, the formula (3) is used. The case where the indicated function is set as the threshold level will be described.

Returning to FIG. 1 again, as explained above,
The characteristic image extraction processing unit 21 of FIG. 1 extracts and digitizes the characteristic of the correlation of the reflected light intensity with respect to each wavelength for each read pixel. For example, when there is a large relative difference between the signal Da obtained from the reflected light of the wavelength λa and the signal Db obtained from the reflected light of the wavelength λb, the characteristic point is as shown in FIG.
Belongs to the hatched area 31 or 32. On the other hand, when there is not much relative difference between the two signals Da and Db, it belongs to the region 33. The characteristic image extraction processing unit 21 uses the signals Da,
From the mapping relationship in the equation (3) of Db, the feature of each pixel is
It is judged whether it belongs to the set of the following formula (5) or (6). E1: {Db <Da + α ∪ Da <Db + α} (5) E2: {Db ≧ Da + α ∩ Da ≧ Db + α} (6) Equation (5) is a set E with feature points in regions 31 and 32 in FIG.
1 and the equation (6) is a set E2 having a feature point in the area 33. When it belongs to the set E1, it is digitized as "1", that is, black bit, and when it belongs to the set E2, it is digitized as "0", that is, white bit. The digitized data corresponding to the pixels of the entire surface of the paper sheet thus read is obtained and the characteristic image is extracted. This characteristic image is temporarily stored in a buffer memory (not shown) having the size of the output image area.

This method of the present invention uses the signals Da and Db described above.
By the pattern drawn on the paper surface of the paper, the correlation with
It takes advantage of the big difference. In this case, (1) above
The threshold level defined by the equations (2) and (2) is an important factor that determines the correlation between the signals Da and Db, and is an important point for identifying the authenticity of the paper sheet. Therefore, the threshold levels THL1 and THL2 should be set such that the characteristic image is greatly different between a genuine note and a counterfeit note after performing various tests.

FIG. 3 is a conceptual diagram of the central coordinate detecting process. The central coordinate detection processing unit 22 detects the central point of rectangular paper sheet coordinates existing in the output image area.
FIG. 3 shows the output image area in which the horizontal axis represents the main scanning coordinate X and the vertical axis represents the sub scanning coordinate Y. In FIG. 3, the image of the set rectangular paper sheet is contained in the output image area regardless of the amount of lateral deviation or skew generated by the paper sheet traveling system (not shown) of the reading unit 10. The edge coordinates of the paper sheet image are detected on the two-dimensional coordinates in the output image area, and the center point of the paper sheet image is detected from the edge coordinates. When the paper sheet is set, the output signal D
a and the output signal Db became the first point on the main scanning coordinate X of the output image area to become larger than 0, and the last point on the main scanning coordinate X of the output image area to become larger than 0. The point is detected as the end coordinate of the paper sheet coordinate in the main scanning direction. Further, the first point on the sub-scanning coordinate Y of the output image area that becomes larger than 0 and the last point on the sub-scanning coordinate Y of the output image area that becomes larger than 0 are the paper sheets. The coordinates are detected as the end coordinates in the sub-scanning direction. The signal for detection uses either the output signal Da or the output signal Db. In FIG. 3, the leftmost coordinate of the paper sheet coordinate on the main scanning coordinate X is MINX, the rightmost coordinate of the paper sheet coordinate is MAXX, and the uppermost coordinate of the paper sheet coordinate on the sub-scanning coordinate Y. Is the minimum coordinate, and the lowest coordinate of the paper sheet coordinate is MAXY. Then, based on the values of the end coordinates MINX, MAXX, MINY and MAXY, the center point coordinates MX, M are calculated as follows.
Y is required. MX = (MINX + MAXX) / 2 MY = (MINY + MAXY) / 2 The center point coordinates (MX, MY) of this paper sheet image coincide with the intersection of the diagonal lines of the rectangular paper sheet image.

The characteristic image distribution extraction processing unit 23 is first extracted by the characteristic image extraction processing unit 21, and is temporarily stored in the buffer memory. From the characteristic image F (x, y) in the sub-scanning direction Y in the main scanning direction X. Scan sequentially to the coordinates,
The distance r to the center point coordinates (MX, MY) of the paper sheet image detected by the center point coordinate detection processing unit 22 is obtained. Next, the one-dimensional sequence ΦR (r) of the number of black bits of the feature image, which is attached in the order of the size of the distance r, is output. That is, the characteristic image F (x, which exists on the concentric circle formed by the distance r from the center point coordinates (MX, MY) of the paper sheet image to arbitrary coordinates (x, y)
The value of y) is integrated. The distance r from the center point of the paper sheet image to the characteristic image F (x, y) is obtained by the equation shown. In this case, the integer value rounded off to the nearest whole number is used. Center point coordinates (MX,
MY) is the center of the concentric circle, and the integrated value ΦR (r) of the characteristic image of the black bit existing on the circle formed by the radius r of the concentric circle (distance from the center point coordinates of the paper sheet image) is The following formula
It will be as shown in (7). ΦR (r) = ΣΣF (x, y) (7) where F is a binary polarity value in the feature image,
The black bit is "1" and the white bit is "0". x, y
Is the position coordinate of each pixel, x is the coordinate value in the main scanning direction, and y is the coordinate value in the sub scanning direction. Further, ΣΣ means that the sum of the polarity values F at all points on the concentric circle centered on the coordinate (MX, MY) is obtained. That is, the above equation (7)
Represents the integral value of the characteristic image F (x, y) existing on a circle having a radius r from the center point coordinates (MX, MY) of the paper sheet image.

The comparison / discrimination processing unit 24 has a characteristic image distribution series ΦR (r) extracted by the characteristic image distribution extraction processing unit 23.
The distribution shape of is compared with the distribution shape stored in advance, and the authenticity, type, and front and back of the target paper sheet are collectively recognized.

The characteristic image distribution series ΦR (r) is the characteristic image F
Even if the paper sheet image at (x, y) is displaced or skewed in the output image area, the center point coordinates (MX, MY) detected by the central image extraction processing unit 22 correspond to the displacement. From the center point coordinates (MX, MY), the feature image F (x,
By setting the distance r to y) to be the radius of the concentric circle, skew deviation can be dealt with, so that it can always be obtained stably. That is, since the reading area that is much larger than the width indicated by the paper sheets is set in the reading unit, even if the set paper sheets are misaligned or skewed, the characteristic image distribution series ΦR
It has no effect on the distribution shape of (r). The reference comparison data is a black bit distribution series Φ composed of a one-dimensional array.
Only R (r) is required, and the authenticity, type, and front and back of the target paper sheet can be collectively recognized by performing simple pattern matching. According to this method, it is possible to perform high-speed and high-accuracy recognition processing from an output image with high resolution detected by one detection means without being affected by positional deviation or skew.

The characteristic image extraction processing method according to the present invention described above is based on the equations (5) and (6).
The black bit distribution series ΦR (r) was obtained for the binary feature image by binarizing the features represented by the correlation and the features not represented by the correlation.
The feature image to be extracted can be treated as multivalued instead of binary. The characteristic image distribution extraction processing in this case will be described.
Of the regions classified by THL1 and THL2 in FIG. 2, the region 31 is determined from the output relationship of the signals Da and Db.
And if it belongs to the area 32, it is regarded as a characteristic image. That is, in the output signal correlation diagram of FIG.
The signal Da and the signal Db belonging to only 2 are used as the output signals due to the characteristics. This is the same as the idea in the binary feature image extraction processing described above, but the shape of the output data is different. The characteristic image F (x, y) is obtained only when the signals Da and Db satisfy the equation (5). As a result, the main scanning coordinate at the reading position is x and the sub scanning coordinate is y.
Then, the signal Da and the signal Db at the reading position are
Therefore, the characteristic image F (x, y) is F (x, y) = | Da−Db | where | Da−Db |> | α | That is, the output relationship between the signal Da and the signal Db is expressed by
Only when the condition (5) is satisfied, the feature image F (x,
A value is assigned to y). When the output relationship between the signals Da and Db does not satisfy the equation (5), the characteristic image F (x,
The value of y) is 0. After that, the center point of the paper sheet image is detected, and the feature image distribution series ΦR (r) is obtained by integrating the feature images F (x, y) existing on the circle formed by the distance r from the center point. In the subsequent processing, recognition is performed by the same method as the processing on the binary characteristic image described above.

Next, an embodiment according to another method of the present invention will be described. FIG. 4 is a block diagram of an embodiment of another method of the present invention. The reading unit 10 optically reads the set paper sheets and outputs an image signal, and the recognition unit 2
0'extracts a characteristic image to be recognized from the image signal output from the reading unit 10 and compares the distribution shape thereof with a reference distribution shape stored in advance to determine whether the paper sheet is true, false, kind or front and back. Is to recognize. The reading unit 10 has the same configuration as that of the first embodiment described above, and thus duplicated description will be omitted. The recognition unit 20 'includes the feature image extraction processing unit 2
1, a center-of-gravity image detection processing unit 25, a characteristic image distribution extraction processing unit 26, and a comparison / discrimination processing unit 27. Among them, the characteristic image extraction processing unit 21 has the same configuration as that of the first embodiment described above, and thus the duplicate description will be omitted. The center-of-gravity coordinate detection processing unit 25 projects a black bit in the two-dimensional coordinates of the binary characteristic image output by the characteristic image extraction processing unit 21 in the horizontal direction, that is, the main scanning direction and the vertical direction, that is, the sub-scanning direction. Then, the black bit distribution is obtained, and the barycentric coordinates are detected from the black bit number distribution.

At an arbitrary coordinate y existing in the reading width YW in the vertical direction, that is, the sub-scanning direction in FIG. 4, the number of black bits existing in the reading width XW in the horizontal direction, that is, the main scanning direction is calculated and calculated. The cumulative value of the black bits is the black bit number distribution at the coordinate y. The black bit number distribution ΦY (y) at the coordinate y is as follows. ΦY (y) = ΣF (x, y) where x is the coordinate in the main scanning direction, XW is the reading width in the main scanning direction, and F (x, y) is the coordinate x in the main scanning direction.
The polarity value of the binary characteristic image at the coordinate y in the sub-scanning direction, where the black bit is 1 and the white bit is 0. Similarly, the black bit distribution series ΦX at the coordinate x in the main scanning direction
(x) is as follows. ΦX (x) = ΣF (x, y) where YW is the reading width in the sub-scanning direction.

Next, the barycentric coordinates GX, GY are calculated from the black bit distribution series ΦX (x), ΦY (y) calculated as described above. Figure 5
FIG. 13 is an explanatory diagram of a centroid dividing process. In FIG. 5, first, the black bit distribution series ΦX of the feature image is expressed as shown in the following equation for the range of the coordinate x in the main scanning direction indicated by the feature image.
The barycentric coordinate GX in the black bit distribution series ΦX (x) is obtained by dividing the sum of the first moments of (x) by the sum of the black bits in the range. GX = ΣΦX (x) x / ΣΦX (x) Next, the barycentric coordinate GY in the black bit distribution series ΦY (y) is obtained for the range of the coordinate Y in the sub-scanning direction. GY = ΣΦY (y) y / ΣΦY (y)

The characteristic image distribution extraction processing unit 26 is sequentially scanned in the main scanning direction x from the characteristic image F (x, y) extracted by the characteristic image extraction processing unit 21 and temporarily stored in the buffer memory in the sub-scanning direction y. To do. Then, from the barycentric coordinates (GX, GY) of the characteristic image F (x, y) detected by the barycentric coordinate detection processing unit 25, the coordinates (X) of the pixel represented as the feature in the characteristic image F (x, y) are calculated. , Y), the values of the characteristic image F (x, y) existing on a concentric circle formed by the distance r to (Y, Y) are integrated. The distance r from the barycentric coordinates (GX, GY) of the characteristic image F (x, y) to the characteristic image F (x, y) is calculated by the equation in the figure, and is a value rounded to the nearest whole number.
The integral value ΦR (r) of the black-bit feature image existing on the circle formed by the radius r of the concentric circle with the center of gravity coordinates (GX, GY) of the feature image F (x, y) as the center point of the concentric circle is Desired. ΦR (r) = ΣF (x, y) That is, the integral value of the number of characteristic images F (x, y) on the concentric circle of the distance r from the barycentric coordinates (GX, GY) of the characteristic image F (x, y). It outputs whether ΦR (r) exists. The comparison / discrimination processing unit 27 compares the distribution shape of the characteristic image distribution series ΦR (r) extracted by the characteristic image distribution extraction processing unit 26 with a pre-stored distribution shape to determine whether the target paper sheet is true or false, Recognize the type and front and back together.

The characteristic image distribution sequence ΦR (r) is the characteristic image F
Even if the paper sheet image at (x, y) causes misalignment or skew in the output image area, the barycentric coordinates (GX, GY) detected by the barycentric image extraction processing unit 25 correspond to the misalignment. , And the feature image F (x, y) from the barycentric coordinates (GX, GY)
Since the skew r is dealt with by setting the distance r to the radius of the concentric circle, it is possible to always obtain a stable value. That is, since the reading area that is much larger than the width indicated by the paper sheets is set in the reading unit, even if the set paper sheets are misaligned or skewed, the characteristic image distribution series ΦR (r) It has no effect on the distribution shape. The reference dictionary data is only the black bit distribution series ΦR (r) composed of a one-dimensional series. By performing simple pattern matching, the authenticity of the target paper sheet, the type, and the front and back are highly accurate. It is recognizable. According to this method, it is possible to perform high-speed and high-accuracy recognition processing from an output image having a high resolution without being affected by misalignment or skew.

The characteristic image extraction processing method according to the present invention described above is based on the equation (5) and the equation (5) in the first embodiment described above.
On the basis of (6), the correlation between the signal Da and the signal Db is divided into those which are expressed as features and those which are not expressed, and binarized, and the black bit distribution series ΦR ( r), but the feature image to be extracted is 2
It can be handled as multi-valued instead of value. The characteristic image distribution extraction processing in this case is the same as that described in the first embodiment described above, and thus redundant description will be omitted.

In the method of the present invention, in principle, the entire sheet is read and read to identify the sheet. However, for example, the present invention is carried out only in a specific area of the sheet. Even if it does, it is possible to identify it with high identification power. Further, in the above-mentioned embodiment, the number of light sources is two, but if three or more light sources are provided and the correlation between them is obtained, more accurate identification can be performed.

[0030]

According to the method of the present invention described above, the image of the entire area of the paper sheet requiring identification is read, and the characteristic image distribution is obtained from the distance from the center or the center of gravity. The characteristic image distribution can be set so as not to affect the positional deviation and skew of the paper sheet, and the paper sheet can always be stably recognized. Therefore, the recognition of the paper sheets can be performed quickly with high reliability, and the processing speed can be increased in the automatic depositing / dispensing machine or the like.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of a paper sheet recognition processing method of the present invention.

FIG. 2 is a signal output correlation diagram.

FIG. 3 is a conceptual diagram of central coordinate detection processing.

FIG. 4 is a block diagram of an embodiment of another method of the present invention.

FIG. 5 is an explanatory diagram of a centroid dividing process.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Paper sheet 2 Image line sensor 3a, 3b Light source 4 Conveyance direction 10 Reading unit 20, 20 'Recognition unit 21 Characteristic image extraction processing unit 22 Center point coordinate detection processing unit 23 Characteristic image distribution extraction processing unit 24 Comparison discrimination processing unit 25 Centroid coordinate detection processing unit 26 Feature image distribution extraction processing unit 27 Comparison determination processing unit

 ─────────────────────────────────────────────────── ─── Continued Front Page (72) Inventor Shunji Sakai 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd.

Claims (2)

[Claims] 1. The paper surface of a paper sheet is alternately irradiated by using a plurality of light sources having different wavelengths, and the characteristic of the correlation of the reflected light intensity with respect to each wavelength is extracted for each read pixel and the numerical value is extracted. In addition to converting the digitized data into the output image area formed by the two-dimensional coordinates, the edges of the paper sheet image on the main scanning direction coordinates and the sub scanning direction coordinates of the entire read pixels. The partial coordinates are detected, the center point coordinates of the paper sheet image are detected from the edge coordinates, and the distance between the coordinates of the pixel represented as the feature and the center point coordinate is calculated from the characteristic image, The feature image distribution is extracted by integrating the output of the feature images corresponding to the distance to the one-dimensional sequence assigned in the order of the magnitude of the distance, and the shape of the feature image distribution is compared with a reference distribution shape prepared in advance. Recognition of the paper sheets Paper sheet recognition method and performing. 2. The paper surface of the paper sheet is alternately irradiated by using a plurality of light sources having different wavelengths, and the characteristic of the correlation of the reflected light intensity with respect to each wavelength is extracted for each read pixel and the numerical value is extracted. In addition, the characteristic image formed by all the digitized data is projected on the main scanning direction coordinates and the sub scanning direction coordinates to obtain each pixel number distribution, and the pixel number distribution in the main scanning direction is obtained. And the barycentric coordinates of the pixel number distribution in the sub-scanning direction, and the barycentric coordinates in the main scanning direction and the barycentric coordinates in the sub-scanning direction are combined to obtain the barycentric coordinates of the feature image. The distance between the coordinates of each pixel on the characteristic image and the coordinates of the entire center of gravity is obtained, and the output of the characteristic image that fits the distance is integrated into a one-dimensional sequence in the order of the distance size to obtain the characteristic image distribution. A paper sheet recognition processing method, which comprises extracting and comparing the shape of the characteristic image distribution with a reference distribution shape prepared in advance to recognize the paper sheet.