WO2011114447A1 - Paper discriminating device and method of discriminating paper - Google Patents

Abstract

A paper discriminating device is configured such that a directional dictionary image for each edge is generated based on an edge domain detected from each learning image and an edge direction in the edge domain using the learning image which is an image taken from an authentic paper; a remaining edge domain which lies only in an input image is extracted by comparing the input image which is an image taken from the paper to be discriminated with the directional dictionary image; the density of the remaining edge domain in a given area domain is detected; and the usable discrimination and kind discrimination are implemented.

Description

Paper sheet discriminating apparatus and paper sheet discriminating method

The present invention relates to a paper sheet discriminating apparatus and a paper sheet discriminating method for discriminating whether a paper sheet such as a banknote is correct or not, and in particular, even if there is diversity in the design and quality of the paper sheet. The present invention relates to a paper sheet discriminating apparatus and a paper sheet discriminating method capable of performing optimum damage discrimination and type discrimination with accuracy.

2. Description of the Related Art Conventionally, a banknote discriminating apparatus that discriminates banknotes that are contaminated such as dirt, wrinkles, and folds from banknotes that are not generated using an image sensor or the like is known. In the following description, a banknote in which such fouling has occurred is referred to as a “damaged ticket”, and a banknote in which such a fouling has not occurred is referred to as a “correct ticket”.

Such a bill discriminating device describes a captured image of a genuine note (hereinafter referred to as “reference image”) acquired using an image sensor or the like and a captured image of a bill to be discriminated (hereinafter referred to as “discrimination target image”). Is generally discriminated by comparing and analyzing.

And, in performing such comparison and analysis, various methods such as using an image edge have been proposed. Here, the edge refers to a portion where the density of the image has changed remarkably, and the density gradient direction of the edge is referred to as the edge direction.

For example, in Patent Document 1, an area in which stains are conspicuous in design is specified in advance as a stain detection target region based on a reference image, and an edge difference in the stain detection target region is analyzed to analyze the damage. A technique for performing the discrimination is disclosed.

In addition, the area where stains are conspicuous in the design refers to an area where edges are not concentrated, that is, a simple design area where there is little change in shading.

Patent Document 2 discloses a technique for comparing the reference image and the discrimination target image in pixel units for the edge direction, and analyzing the difference in the edge direction detected by the comparison, thereby determining the fitness. ing.

JP 2006-302109 A Japanese Patent No. 3207931

However, when the technique of Patent Document 1 described above is used, the comparison is limited only to the area where the contamination is conspicuous. Therefore, there is a problem that it is not possible to determine the correctness when the contamination is generated in other areas. It was.

In addition, when the technique of Patent Document 2 described above is used, the reference image and the discrimination target image are compared on a pixel-by-pixel basis. Therefore, when there is a printing deviation due to multi-plate printing, the deviation is minute. However, there was a high possibility of being identified as a non-payment.

In this respect, it can be said that discriminating such a case as a damaged ticket can be performed with high accuracy, but on the other hand, a banknote that does not hinder the distribution is handled as a damaged ticket, which is not appropriate.

From these facts, how to realize a banknote discriminating apparatus or a banknote discriminating method capable of discriminating optimal damages with high accuracy even when the design and quality of banknotes are diverse. It has become a challenge. The diversity here refers to, for example, the variety of designs such as whether or not the pattern has many edges and the stains are not noticeable, and the variation in print quality due to multi-plate printing.

Note that this problem is not limited to banknotes, and similarly occurs when there are variations in the design and quality of paper sheets such as gift certificates and passbooks. In addition, it is a problem that occurs in the same way in the type discrimination of paper sheets based on such diversity.

The present invention has been made in order to solve the above-described problems caused by the prior art, and even when there is diversity in the design and quality of paper sheets, it is possible to determine the optimum damage with high accuracy and An object of the present invention is to provide a paper sheet discriminating apparatus and a paper sheet discriminating method capable of discriminating types.

In order to solve the above-described problems and achieve the object, the present invention is a paper sheet discriminating apparatus that discriminates a paper sheet based on a captured image of a paper sheet, and is a captured image of a legitimate paper sheet. When the edge position and the edge direction at the edge position are detected from the learning image, the direction dictionary corresponding to the edge direction among a plurality of direction dictionary images prepared for each predetermined edge direction range. A direction-specific dictionary image generation unit that generates the direction-specific dictionary image by allocating the edge position to a corresponding position of the image, and an input image that is a captured image of a paper sheet to be determined is the direction-specific dictionary image And a discriminating means for discriminating the damage and type of the paper sheet.

Further, according to the present invention, in the above invention, when an edge position and an edge direction at the edge position are detected from the input image, the edge position is separated for each edge direction range based on the edge direction. The direction-specific input image generation means for generating the direction-specific input image, and all the edge positions that exist only in the direction-specific input image among the direction-specific input image and the direction-specific dictionary image related to the same edge direction range. And a remaining edge extracting means for extracting a remaining edge region by superimposing the edge direction range of the edge direction range, and the determining means is configured to determine the determination target based on the remaining edge region extracted by the remaining edge extracting means. It is characterized by discriminating whether the paper sheet is normal or not.

Further, the present invention is the above invention, wherein the direction-specific dictionary image generation means is such that the edge direction at the detected edge position is within a predetermined range from a boundary line between two adjacent edge direction ranges. In addition, the edge position is assigned to the corresponding position in the direction dictionary image corresponding to each of the two edge direction ranges.

The present invention further includes a density detection unit that detects a density of the remaining edge region extracted by the remaining edge extraction unit in the above-described invention, and the determination unit detects the density detected by the density detection unit. It is characterized by discriminating whether or not the paper sheet to be discriminated is based on the density.

Further, according to the present invention, in the above invention, the direction-specific dictionary image generation means uses, as a valid edge region, a pixel to which a predetermined number or more of the edge positions are allocated using a predetermined number of learning images. It is characterized by.

Further, the present invention is characterized in that, in the above-mentioned invention, the direction-specific dictionary image generating means performs expansion processing for expanding the edge position or the effective edge region to surrounding pixels for each direction-specific dictionary image. .

Further, the present invention is characterized in that, in the above-mentioned invention, the direction-specific dictionary image generation means performs the expansion process giving priority to the edge direction range corresponding to the direction-specific dictionary image.

Further, according to the present invention, in the above invention, a peripheral edge counting unit that counts the number of pixels to which the edge positions included in a predetermined vicinity range of each pixel in the directional dictionary image are assigned, and the peripheral edge count Statistic calculation means for calculating a statistic obtained by statistically calculating the number counted for each pixel by the means for a plurality of learning images, and the direction-dependent dictionary image generation means is configured to perform the statistic calculation means by the statistic calculation means. When the statistic is calculated, the directional dictionary image is generated by setting the pixel value of the pixel in the directional dictionary image based on the statistic.

The present invention also relates to a paper sheet discrimination method for discriminating a paper sheet based on a captured image of a paper sheet, the edge position and the edge position from a learning image that is a legitimate captured image of the paper sheet When the edge direction is detected, the edge position is distributed to the corresponding position in the direction-specific dictionary image corresponding to the edge direction among a plurality of direction-specific dictionary images prepared for each predetermined edge direction range. The direction-specific dictionary image generation step for generating the direction-specific dictionary image in FIG. 6 and the input image, which is a captured image of the paper sheet to be discriminated, are compared with the direction-specific dictionary image to determine whether the paper sheet is normal or not. And a discrimination step for discriminating the type.

According to the present invention, when an edge position and an edge direction at the edge position are detected from a learning image that is a captured image of a valid paper sheet, a plurality of directions prepared for each predetermined edge direction range are detected. Of the dictionary images, direction-specific dictionary images are generated by allocating the detected edge positions to the corresponding positions in the direction-specific dictionary image corresponding to the detected edge direction, and the captured image of the paper sheet to be determined The input image is compared with the direction-specific dictionary image to determine the correctness and type of the paper, so even if there is a variety in the design and quality of the paper, There is an effect that it is possible to perform the optimum damage discrimination and type discrimination with accuracy.

Further, according to the present invention, when an edge position and an edge direction at the edge position are detected from the input image, the edge position is separated for each direction by separating the edge position for each edge direction range based on the detected edge direction. Generates an input image and extracts the remaining edge region by superimposing the edge positions that exist only in the direction-specific input image for all edge direction ranges from the direction-specific input image and direction-specific dictionary image for the same edge direction range. In addition, since it is decided to determine whether or not the paper sheet to be discriminated is based on the extracted remaining edge area, the remaining edge area can be extracted more clearly from a plurality of edge directions, and the pattern is complicated. Highly accurate and optimal damage discrimination and type discrimination can be performed even for paper sheets that are not easily noticeable due to their design. The effect say.

Further, according to the present invention, when the edge direction at the detected edge position is within a predetermined range from the boundary line between two adjacent edge direction ranges, the direction dictionary corresponding to each of the two edge direction ranges. Since the detected edge position is assigned to the corresponding position of the image, there is an effect that it is possible to generate a dictionary image that is resistant to rotational deviation in consideration of a predetermined rotational deviation absorption amount in the detected edge direction.

In addition, according to the present invention, the density of the extracted remaining edge region is detected, and whether or not the paper sheet to be determined is determined based on the detected density is determined. Therefore, it is possible to perform the damage discrimination that is narrowed down to the fouled portion having the thickness, and to prevent erroneous discrimination such as wrinkles extending linearly on the sheet surface.

In addition, according to the present invention, since a predetermined number of learning images are used and pixels to which a predetermined number or more of edge positions are assigned are used as effective edge regions, a dictionary that takes into account printing misalignment between regular tickets There is an effect that an image can be created.

In addition, according to the present invention, since the edge processing or the effective edge region is expanded to the peripheral pixels for each of the directional dictionary images, the positional deviation is prevented without reducing the extraction capability of the remaining edge region. There is an effect that a strong dictionary image can be created. In addition, there is an effect that the expansion amount can be increased as compared with the case where there is no dictionary image for each direction.

Further, according to the present invention, since the expansion process is performed with priority given to the edge direction range corresponding to the direction-specific dictionary image, it is possible to create a dictionary image expanded in the direction in which the edge detection deviation occurs. There is an effect.

In addition, according to the present invention, the number of pixels to which the edge positions included in a predetermined neighborhood range of each pixel in the direction-specific dictionary image are distributed is counted, and the number counted for each pixel is obtained for a plurality of learning images. Since the statistic is calculated, and when the statistic is calculated, the directional dictionary image is generated by setting the pixel value of the pixel in the directional dictionary image based on the statistic. There is an effect that it is possible to create a dictionary image corresponding to the peripheral characteristics (design diversity, etc.) of each.

FIG. 1 is a diagram showing an outline of a paper sheet discrimination method according to the present invention. FIG. 2 is a block diagram illustrating the configuration of the bill discriminating apparatus according to the first embodiment. FIG. 3 is a diagram for explaining edge direction detection processing performed by the edge direction detection unit. FIG. 4 is a diagram for explaining the valid edge determination process performed by the valid edge determination unit. FIG. 5 is a diagram for explaining expansion processing performed by the expansion processing unit. FIG. 6 is a diagram illustrating a modification of the expansion process. FIG. 7 is a diagram for explaining the remaining edge extraction processing performed by the remaining edge extraction unit. FIG. 8 is a diagram for explaining the remaining edge analysis process performed by the remaining edge analysis unit and the determination process performed by the determination unit. FIG. 9 is a flowchart illustrating a processing procedure executed by the bill discriminating apparatus according to the first embodiment. FIG. 10 is a flowchart illustrating a processing procedure of dictionary image creation processing executed by the banknote determination apparatus according to the first embodiment. FIG. 11 is a flowchart illustrating the procedure of the damage determination process executed by the banknote determination apparatus according to the first embodiment. FIG. 12 is a diagram illustrating an outline of dictionary image creation of the bill discriminating apparatus according to the second embodiment. FIG. 13 is a block diagram illustrating the configuration of the bill discriminating apparatus according to the second embodiment. FIG. 14 is a flowchart illustrating a processing procedure of dictionary image creation processing executed by the banknote discriminating apparatus according to the second embodiment. FIG. 15 is a flowchart illustrating the procedure of the damage determination process executed by the banknote determination apparatus according to the second embodiment.

Hereinafter, preferred embodiments of the paper sheet discrimination method according to the present invention will be described in detail with reference to the accompanying drawings. In the following, the outline of the paper sheet discriminating method according to the present invention will be described with reference to FIG. 1, and then an embodiment of a bill discriminating apparatus to which the paper sheet discriminating method according to the present invention is applied will be described with reference to FIGS. 15 will be used for explanation. In the following, a description will be given of a case where the damage determination is mainly performed in the paper sheet determination.

In the following description, the banknote discriminating apparatus describes a direction-specific dictionary image created for each direction by processing a learning image, which is a captured image of a valid paper sheet, as a “dictionary image”.

First, the outline of the paper sheet discrimination method according to the present invention will be described with reference to FIG. FIG. 1 is a diagram showing an outline of a paper sheet discrimination method according to the present invention.

As shown in the figure, the paper sheet discriminating method according to the present invention is characterized in that it detects the edge point and the edge direction of a genuine ticket image and creates a dictionary image for each edge direction. is there. Here, the edge point is a pixel corresponding to an edge in the image.

Further, when comparing and analyzing the discrimination target image and the dictionary image, a difference between the discrimination target image and the dictionary image (hereinafter referred to as “residual edge point”) is taken, and the density of the residual edge point in a predetermined size region is calculated. There is also a feature in that it is used to determine whether the damage is correct or not.

Hereinafter, the paper sheet discrimination method according to the present invention will be described in detail. The paper sheet discriminating method according to the present invention can be broadly divided into a stage for creating a dictionary image using the correct bill image 1 as input data and a stage for discriminating damage using the discrimination target image 5 as input data.

In the paper sheet discriminating method according to the present invention, first, as a stage of creating a dictionary image, all edge points and their edge directions are detected for a plurality (= N) of genuine ticket images 1 (in the figure). (See (1)). Then, a dictionary image for each edge direction is created based on the edge points and the edge directions of the N sheets of genuine ticket images (see (2) in the figure).

For example, as shown in (2) of the figure, the edge direction is divided into six directions in increments of 60 degrees of 360 degrees. Then, a unique designation of directions A to F is assigned to each of the six divided directions. In the following, the division of the direction is described as “quantization”, and the number of divisions is described as “quantization number”.

In the paper sheet discrimination method according to the present invention, a plane corresponding to the quantization (hereinafter referred to as “direction plane”) is prepared, and it corresponds to the edge direction of each edge point of the detected correct ticket image 1. The fact that an edge point has been detected is stored at the corresponding coordinates on each direction plane. Then, a dictionary image is created with the coordinates having the number of times of storage equal to or greater than a predetermined number as effective edge points.

For example, as shown in (2) of the figure, the direction plane A corresponding to the A direction, the direction plane B corresponding to the B direction, the direction plane C corresponding to the C direction, the direction plane D corresponding to the D direction, and the direction plane D corresponding to the E direction. A direction plane F corresponding to the direction planes E and F is prepared, each edge point is stored in each direction plane A to F for each edge direction, and a dictionary image for each direction is created. Details of the dictionary image creation will be described later with reference to FIGS.

Here, the facing direction may be stored in the same direction plane. This is because edge points in the edge direction that are 180 degrees symmetrical are often detected in the vicinity of each edge point. For example, edge points in the A and D directions can be stored in the direction plane A, edge points in the B and E directions can be stored in the direction plane B, and edge points in the C and F directions can be stored in the direction plane C, respectively. . In this case, the number of directional planes is 3 for 6 quantizations.

Also, at the stage of creating the dictionary image, the paper sheet discrimination method according to the present invention expands each pixel that is an edge point by a predetermined amount. For example, (2) of FIG. 1 shows an example in which the pixel Px of the portion surrounded by the circle o is expanded to the expansion range D0 and registered in the dictionary image. Details of the predetermined amount of expansion will be described later with reference to FIG.

By the way, the conventional technique using the edge direction (for example, Patent Document 2) does not use a method for creating a dictionary image for each edge direction, unlike the paper sheet discrimination method according to the present invention. Therefore, the dictionary image is one plane data for one sheet of paper, and it is necessary to determine one edge point and one edge direction.

On the other hand, in the paper sheet discrimination method according to the present invention, a dictionary image is created for each edge direction. For example, an edge point having an edge direction near the quantization boundary spans two edge directions. Can be handled. That is, one image of the dictionary image has two edge directions. Therefore, it is possible to perform a damage determination that absorbs a rotational deviation or the like that occurs during printing or scanning of a paper sheet. In the following, the degree of deviation from the quantization boundary to be handled across two edge directions will be referred to as “rotational deviation absorption amount”.

The same applies to the above-mentioned expansion of the predetermined amount. The pixel value and edge direction of one edge point are expanded by expansion to absorb the design diversity and various positions and rotation loss. Discrimination can be performed. Further, when a certain pixel that was not originally an edge point is included in an expansion range of a different direction plane (that is, within a “predetermined amount”) due to such a predetermined amount of expansion, the pixel corresponds to each direction plane. Can have an edge direction. This will be described later with reference to FIG. Further, since the coordinates having a predetermined number or more of edge points from the plurality of regular image images are determined as the effective edge points by the effective edge point determination, a plurality of edge directions can be given to one coordinate of the dictionary image. . Therefore, it is possible to perform a damage determination that absorbs a variety of paper sheet designs and a plurality of rotations and misalignments.

Next, at the stage of discriminating damage, in the paper sheet discriminating method according to the present invention, the discrimination target image 5 and each dictionary image in which the edge point and the edge direction are detected are detected as in the case of the correct ticket image 1. Compare (see (3) in FIG. 1). Then, damage determination is performed using the density of the remaining edge points extracted by the comparison in a predetermined size region (see (4) in FIG. 1). Details of this point will be described later with reference to FIG.

By the way, in the technique (for example, patent document 1) using the conventional remaining edge point, the method of performing a damage determination according to whether the number of pixels detected as a remaining edge point exceeds a predetermined threshold value was used. . Therefore, wrinkles that extend in a straight line on the sheet surface are often determined as fouling.

On the other hand, in the paper sheet discrimination method according to the present invention, since the density of the remaining edge points in the predetermined size region is used, only the stain having the width and the height can be discriminated. It is possible to prevent erroneous discrimination such as.

As described above, in the paper sheet discrimination method according to the present invention, a dictionary image is created for each edge direction in which the edge points are expanded by a predetermined amount, and the damage determination is performed using the density of the remaining edge points in the predetermined size region. It was. Therefore, even if there is diversity in the design and quality of the paper sheets, it is possible to perform the optimum damage determination with high accuracy.

Hereinafter, an embodiment of the bill discriminating apparatus to which the paper sheet discriminating method according to the present invention described with reference to FIG. 1 is applied will be described in detail. In the following, an embodiment of a banknote discriminating apparatus that creates a dictionary image in which each edge point is expanded by a predetermined amount will be referred to as Example 1, and a banknote discriminating apparatus that creates a dictionary image based on the number of peripheral edges of each edge point will be described. An embodiment will be described as a second embodiment. In each of the following embodiments, a case where the banknote discriminating apparatus captures an image of a banknote using an image line sensor (hereinafter referred to as “line sensor”) will be described.

FIG. 2 is a block diagram illustrating the configuration of the bill discriminating apparatus 10 according to the first embodiment. In addition, in the same figure, only the component required in order to demonstrate the characteristic of the banknote discrimination | determination apparatus 10 is shown, and the description about a general component is abbreviate | omitted.

As shown in the figure, the banknote discriminating apparatus 10 includes a line sensor unit 11, a control unit 12, and a storage unit 13. The control unit 12 includes a setting unit 12a, an image data acquisition unit 12b, an edge direction detection unit 12c, an effective edge determination unit 12d, an expansion processing unit 12e, a remaining edge extraction unit 12f, and a remaining edge analysis unit. 12g and the determination part 12h are further provided. The storage unit 13 also stores setting information 13a, a filter group 13b, direction plane edge point amounts 13c, direction dictionary images 13d, and discrimination reference information 13e.

The line sensor unit 11 is a sensor that receives transmitted light or reflected light from a bill conveyed by a conveyance mechanism (not shown), and is configured by arranging a plurality of light receiving sensors in a straight line. The line sensor unit 11 also performs a process of outputting the acquired data to the image data acquisition unit 12b of the control unit 12.

The control unit 12 generates banknote image data based on the output from the line sensor unit 11, detects edge points and directions of the generated image data, and corrects a plurality of (= N) image data. If it is a ticket image, a dictionary image is created based on the detected edge point and its direction, and if the image data is a discrimination target image, the discrimination target image and the dictionary image are compared and analyzed to determine the banknote. It is a processing part which performs the damage determination of.

The setting unit 12a is a processing unit that extracts various parameters such as the number of quantization and the rotational deviation absorption amount included in the setting information 13a of the storage unit 13 and performs initial settings related to dictionary creation and damage determination. The setting unit 12a also performs processing for outputting the extracted various parameters to the edge direction detection unit 12c.

The image data acquisition unit 12b is a processing unit that performs a process of synthesizing the output from the line sensor unit 11 for each banknote and generating image data for the entire banknote. In addition, the image data acquisition part 12b produces | generates N sheets of genuine ticket image data collectively, when creating a dictionary image.

The image data acquisition unit 12b also performs a process of outputting the generated image data to the edge direction detection unit 12c.

The edge direction detection unit 12c is a processing unit that performs processing for smoothing the image data output from the image data acquisition unit 12b and detecting all edge points and directions in the smoothed image data.

Note that, in such smoothing and detection of edge points and their directions, the edge direction detection unit 12c uses various filters stored in advance in the filter group 13b of the storage unit 13. This will be described later with reference to FIG.

Further, the edge direction detection unit 12c indicates the detected edge point and its direction to the valid edge determination unit 12d when creating a dictionary image, and the remaining edge extraction unit 12f when performing banknote damage determination. In addition, the respective output processes are performed together.

Here, the edge direction detection processing performed by the edge direction detection unit 12c will be described in more detail with reference to FIG. FIG. 3 is a diagram for explaining edge direction detection processing performed by the edge direction detection unit 12c.

In the following, edge direction detection processing using a filter will be shown, but when applying such a filter to an image, a pixel on the image where the center pixel of the filter overlaps will be referred to as a “target pixel” hereinafter. I will do it.

As shown in the figure, the edge direction detection unit 12c smoothes the genuine note image 1 or the discrimination target image 5 output by the image data acquisition unit 12b (see (1) in the figure). For such smoothing, for example, a smoothing filter f1 having a coefficient of 1 in the vicinity of 3 × 3 centering on the target pixel can be used.

Then, by performing such smoothing, it is possible to remove the noise component in the correct ticket image 1 or the discrimination target image 5. Therefore, the application of the smoothing filter f1 may be repeated a plurality of times (for example, three times) until an optimum noise removal result is obtained. Alternatively, the size of the smoothing filter f1 may be increased (for example, in the vicinity of 7 × 7) and applied only once. Even when the smoothing filter f1 is not applied, it is possible to detect edge points and edge directions.

Subsequently, the edge direction detection unit 12c detects an edge point of the correct ticket image 1 or the discrimination target image 5 (see (2) in the figure). Here, for the detection of the edge point, for example, a Laplacian filter f2 can be used. That is, the zero cross point detected by applying the Laplacian filter f2 becomes the edge point of the correct ticket image 1 or the discrimination target image 5.

Next, the edge direction detection unit 12c detects the edge direction of the correct ticket image 1 or the discrimination target image 5 by using, for example, the Prewitt filters f3 and f4 (see (3) in the figure). Here, the white arrows in the five black pixels included in the rectangle r shown in the figure indicate the detected edge directions. Then, the edge direction at each edge point is detected (see (4) in the figure).

In the description using the same figure, an example in which the edge point and its direction are detected using the Laplacian filter f2, the Prewitt filters f3 and f4 is shown, but other filters such as a Sobel filter may be used. Various filters may be appropriately combined. Further, the size and coefficient values of various filters are not particularly limited.

Returning to the description of FIG. 2, the valid edge determination unit 12d will be described. The effective edge determination unit 12d is a processing unit that determines whether or not each pixel in the direction plane is an effective edge point, and performs such determination using the edge point amount 13c for each direction plane stored for each pixel in the direction plane. I do.

The effective edge determination unit 12d also performs a process of outputting each direction plane storing the effective edge points to the expansion processing unit 12e.

Here, the effective edge determination process performed by the effective edge determination unit 12d will be described in more detail with reference to FIG. FIG. 4 is a diagram for explaining the effective edge determination process performed by the effective edge determination unit 12d. Note that (A) in the figure shows an edge point amount counting process at an arbitrary edge point P1 of the genuine ticket image, and (B) in the same figure shows an edge point quantity counting process at an arbitrary edge point P2. (C) in the figure shows the valid edge determination processing at the edge points P1 and P2.

As shown in FIG. 5A, when the edge direction detection unit 12c detects that the edge direction of the edge point P1 is the A direction, the valid edge determination unit 12d The edge point amount of the pixel P1 is counted up.

Further, as shown in FIG. 5B, when it is detected that the edge direction of the edge point P2 is within the rotational deviation absorption amount from the boundary between the A direction and the F direction, the effective edge determination unit 12d The edge point amounts of the pixel P2 at the same position in the direction plane A and the pixel P2 at the same position in the direction plane F are counted up.

That is, as shown in (B) of the figure, when the edge direction of the edge point is within the rotational deviation absorption amount from the boundary, the effective edge determination unit 12d detects the pixel at the same position in the two direction planes. Count up the amount of edge points.

Then, the valid edge determination unit 12d performs an edge point amount counting process for each edge point for all the regular image acquired in creating the dictionary image, and obtains an edge point amount equal to or greater than a predetermined determination threshold for each direction plane. The pixel is stored for each direction plane as an effective edge point.

Specifically, as shown in (C) of the figure, when the number of read correct bills is “200” and the determination threshold is “5” for any pixel, the pixel P1 in the direction plane A is Since an edge point amount of “200” that is equal to or greater than the determination threshold is obtained, it is stored as an effective edge point in the direction plane A.

Similarly, the pixel P2 in the direction plane A is also stored as an effective edge point in the direction plane A because the edge point amount of “180” that is equal to or greater than the determination threshold is obtained.

However, since the pixel P2 in the direction plane F has only obtained an edge point amount of “3” that does not satisfy the determination threshold, it is not stored as an effective edge point in the direction plane F.

Returning to the description of FIG. 2, the expansion processing unit 12e will be described. The expansion processing unit 12e is a processing unit that performs a process of expanding all effective edge points on each direction plane by a predetermined amount. The expansion processing unit 12e also performs a process of storing each direction plane after performing the expansion processing in the storage unit 13 as a dictionary image for each direction.

Here, the expansion processing performed by the expansion processing unit 12e will be described in more detail with reference to FIG. FIG. 5 is a diagram for explaining expansion processing performed by the expansion processing unit 12e. (A) in the figure shows a basic example of the dilation process, (B) in the figure shows a procedure from the dilation process to creation of a dictionary image, and (C) in the figure shows a plurality of dilations. The explanatory diagrams about the pixels at the same position included in the range are respectively shown.

As shown in FIG. 5A, in a coordinate system composed of an m-axis and an n-axis, a pixel in which “m” is “i” and “n” is “j” is assumed as a target pixel. In the following, when indicating the position of a pixel, the pixel is referred to as “P” and described as “P (i, j)”.

Here, the expansion processing refers to processing for converting the pixel value of a neighboring pixel adjacent to the target pixel into the pixel value of the target pixel. A filter can be used for the conversion of the pixel value.

For example, when the target pixel is expanded with respect to adjacent eight neighboring pixels (hereinafter referred to as “1-pixel expansion”), P (i−1, j−1), P (i−1, j), P Each pixel value of (i−1, j + 1), P (i, j−1), P (i, j + 1), P (i + 1, j−1), P (i + 1, j), P (i + 1, j + 1) Is converted into the pixel value of the target pixel P (i, j) (see the black pixel in FIG. 5A).

This expansion process is mainly intended to absorb misalignment and rotational displacement due to expansion of the pixel range (see FIG. 1).

Note that, here, one-pixel expansion has been described as an example, but such one-pixel expansion can be repeated by specifying a predetermined expansion amount. For example, when 3 pixels are designated as a predetermined expansion amount, the 1-pixel expansion may be repeated three times. Further, when such a three-pixel expansion is performed, the pixel value of the target pixel finally expands to each pixel in the vicinity of 7 × 7 with the target pixel as the center.

Further, as shown in FIG. 5B, the expansion processing unit 12e performs the above-described expansion processing on all effective edge points on each direction plane (see (B-1) in FIG. 5). Then, after all the effective edge points are expanded, each direction plane is stored in the storage unit 13 as a dictionary image for each direction (see (B-2) in FIG. 5).

Here, the pixels at the same position included in the expansion range of a plurality of direction planes will be described. As shown in FIG. 5C, for example, the pixel P3 at a predetermined position is included in both the arbitrary expansion range D1 of the directional plane A and the arbitrary expansion range D2 of the directional plane F that have undergone the expansion process. Shall.

In such a case, the pixel P3 when viewed as one regular image rather than by direction has two edge directions (range of the angle θ shown in the figure) in the A direction and the F direction. Therefore, regardless of whether the pixel P3 of the discrimination target image is in the A direction or the F direction, it can be determined that such an edge direction exists in the genuine ticket image.

By the way, in the description using FIG. 5, the case where the expansion processing is performed on all adjacent neighboring pixels centering on the target pixel has been described. However, dilation processing may be performed on adjacent neighboring pixels in the direction corresponding to the edge direction.

Therefore, in the following, a modification of the expansion process in such a case will be described with reference to FIG. FIG. 6 is a diagram illustrating a modification of the expansion process. In the following description, a pixel (see FIG. 5) in which “m” is “i” and “n” is “j” will be described as a target pixel. Further, the case of indicating the position of the pixel is the same as the case of the description of FIG. 5 described above.

As shown in FIG. 6A, when the pixel of interest has an edge direction in the A or D direction, the expansion process can be performed only on adjacent neighboring pixels corresponding to the A or D direction. it can. Specifically, P (i−2, j), P (i−2, j + 1), P (i−2, j + 2), P (i−1, j), P (i−1, j + 1), P (i-1, j + 2), P (i, j-1), P (i, j + 1), P (i + 1, j-2), P (i + 1, j-1), P (i + 1, j), Expansion processing may be performed on each pixel such as P (i + 2, j-2), P (i + 2, j-1), and P (i + 2, j).

Further, as shown in FIG. 5B, when the target pixel has an edge direction in the B or E direction, P (i−1, j−2) corresponding to the B or E direction. ), P (i−1, j−1), P (i−1, j), P (i−1, j + 1), P (i−1, j + 2), P (i, j−2), P (I, j-1), P (i, j + 1), P (i, j + 2), P (i + 1, j-2), P (i + 1, j-1), P (i + 1, j), P (i + 1 , J + 1), P (i + 1, j + 2), etc., the expansion process may be performed.

Further, as shown in FIG. 6C, when the target pixel has an edge direction in the C or F direction, P (i−2, j−2) corresponding to the C or F direction. ), P (i-2, j-1), P (i-2, j), P (i-1, j-2), P (i-1, j-1), P (i-1, j), P (i, j-1), P (i, j + 1), P (i + 1, j), P (i + 1, j + 1), P (i + 1, j + 2), P (i + 2, j), P (i + 2) , J + 1), P (i + 2, j + 2), etc., the expansion process may be performed.

Returning to the description of FIG. 2, the remaining edge extraction unit 12f will be described. The remaining edge extraction unit 12f performs a process of generating a difference image obtained by extracting a difference between the determination target image and the dictionary image and extracting a remaining edge point that exists only in the determination target image when determining whether the bill is correct or not. It is a processing part to perform.

The remaining edge extracting unit 12f also performs a process of outputting the generated difference image to the remaining edge analyzing unit 12g.

Here, the remaining edge extraction processing performed by the remaining edge extraction unit 12f will be described in more detail with reference to FIG. FIG. 7 is a diagram for explaining the remaining edge extraction processing performed by the remaining edge extraction unit 12f.

As shown in the figure, the remaining edge extraction unit 12f stores each edge point of the discrimination target image 5 detected by the edge direction detection unit 12c in each direction plane for each edge direction ((1) in the figure). reference). Note that each edge point of the discrimination target image 5 includes an edge point other than the effective edge points constituting the dictionary image.

Then, the remaining edge extraction unit 12f extracts the direction-specific dictionary image of the banknote as the discrimination target from the direction-specific dictionary image 13d stored in the storage unit 13 (see (2) in the figure).

Then, the remaining edge extracting unit 12f takes each difference obtained by subtracting each dictionary image corresponding to each direction plane from each direction plane in which each edge point of the discrimination target image 5 is stored ((3) in the figure). reference).

Then, the remaining edge extraction unit 12f generates a difference image 5 ′ by combining the differences (see (4) in the figure). As a result, residual edge points (including edge points other than the effective edge points) that existed only in the discrimination target image 5 are extracted from the difference image 5 ′ (see (4) in FIG. 5).

In the figure, a portion surrounded by a broken line R1 or R2 is shown as a remaining edge point.

Returning to the description of FIG. 2, the remaining edge analysis unit 12g will be described. The remaining edge analysis unit 12g removes the isolated points from the remaining edge points extracted by the remaining edge extraction unit 12f, and generates a density image in which the density of the remaining remaining edge points in a predetermined size is detected. It is a processing unit.

In the removal of the isolated points and the generation of the density image, the remaining edge analysis unit 12g uses various filters stored in advance in the filter group 13b of the storage unit 13. This will be described later with reference to FIG.

The remaining edge analysis unit 12g also performs a process of outputting the generated density image to the determination unit 12h.

The discrimination unit 12h is a processing unit that performs a process of finally discriminating whether or not the banknote is the discrimination target based on the density image binarized with a predetermined threshold. For the predetermined threshold, the determination unit 12h refers to the determination criterion information 13e in the storage unit 13.

Here, the remaining edge analysis process performed by the remaining edge analysis 12g and the determination process performed by the determination unit 12h will be described in more detail with reference to FIG. FIG. 8 is a diagram for explaining the remaining edge analysis process performed by the remaining edge analysis unit 12g and the determination process performed by the determination unit 12h.

As shown in the figure, the remaining edge analyzing unit 12g removes the remaining edge points isolated from the surroundings as isolated points from the remaining edge points of the difference image 5 ′ output by the remaining edge extracting unit 12f (see FIG. (See (1)). For example, in the figure, among the remaining edge points of the difference image 5 ′, the remaining edge point surrounded by the broken line R2 is removed as an isolated point, and the remaining edge point surrounded by the broken line R1 is the final remaining edge point. Shows the remaining case.

Here, an isolated point removal filter f5 can be used to remove such isolated points (see (1) in the figure). The isolated point removal filter f5 converts the target pixel into white when all the eight neighboring pixels of the target pixel are white.

Then, the remaining edge analysis unit 12g applies density detection filters f6 and f7 to the difference image 5 ′ (see (2) in the figure) to detect the density of the remaining edge portion as a pixel value. The density detection filters f6 and f7 are filters of a predetermined size composed of L1 × L2 filter elements.

Here, it is preferable that the predetermined size is the minimum size of the contamination that is desired to be detected in determining whether the damage is correct or not. In addition, all the filter coefficients can be set to 1, but when considering the density of the remaining edge points, the coefficients near the center of the filter may be increased.

In addition, a horizontally long density detection filter f6 is used to detect the density of the remaining edge portion that is horizontally long, and a vertically long density detection filter f7 is used to detect the density of the remaining edge portion that is vertically long. Respectively.

For example, (2-a) shown in the figure is a case where the horizontally long density detection filter f6 is applied to the difference image 5 ′. In this case, the density of the remaining edge portion that is long in the horizontal direction is expressed as each pixel value of the portion surrounded by the broken line Ry.

Further, as shown in (2-b) of the figure, when the vertically long density detection filter f7 is applied to the difference image 5 ′, the density of the remaining edge portion which is vertically long is surrounded by the broken line Rt. It is expressed as each pixel value of the part.

The remaining edge analysis unit 12g outputs a density image 5 ″ obtained by synthesizing the application results of the density detection filters f6 and f7 to the determination unit 12h.

Then, the determination unit 12h binarizes the density image 5 ″ with a predetermined threshold (see (3) in the figure). The predetermined threshold is a pixel value based on the predetermined size of the density detection filters f6 and f7 and the predetermined reference density. The predetermined reference density is a ratio of the number of remaining edge points that are predicted to be detected in a predetermined size in the case of a slip to the total value of filter coefficients.

Here, an example of the predetermined threshold will be specifically shown. For example, “L1” of the density detection filters f6 and f7 shown in (2) of the figure is “15”, “L2” is “51”, and all the filter coefficients are “1”. And Further, it is assumed that the predetermined reference density is “0.05”. In this case, the predetermined threshold value is “38 (≈765 × 0.05), which is a product of the total value“ 765 (= 15 × 51 × 1) ”of the filter coefficients and the predetermined reference density“ 0.05 ”. ) ”.

Therefore, when the predetermined threshold value is “38”, the black portions surrounded by the broken line R shown in (3) of FIG. 5 are all portions where the pixel values exceed “38”. And when the black part enclosed by this broken line R exists, the discrimination | determination part 12h will discriminate | determine the banknote which is a discrimination | determination object as a non-use ticket.

In the description with reference to the same figure, the density detection filter f6 or f7 is a rectangular filter. However, in the case of performing damage determination with emphasis on the density rather than the shape and size of the stain. May be a circular Gaussian filter.

Returning to the description of FIG. 2, the storage unit 13 will be described. The storage unit 13 is a storage unit configured by a storage device such as a hard disk drive or a non-volatile memory. The setting information 13a, the filter group 13b, the edge point amount by direction plane 13c, and the dictionary image by direction 13d are discriminated. The reference information 13e is stored.

The setting information 13a is information relating to the initial setting such as the quantization number and the rotational deviation absorption amount. The filter group 13b defines various filters used in the edge direction detection process (see FIG. 3) and the remaining edge analysis process (see FIG. 8) in advance, and is referred to by the edge direction detection unit 12c and the remaining edge analysis unit 12g. Is done.

The edge point amount 13c by direction plane is an edge point amount in each direction plane corresponding to each edge point of the N sheets of the regular image, and is registered and updated by the valid edge determination unit 12d. The direction-specific dictionary image 13d is a collection of direction-specific dictionary images, and is registered by the expansion processing unit 12e for each quantized direction. Further, the comparison with the discrimination target image is referred to by the remaining edge extraction unit 12f. The discrimination reference information 13e is information relating to a predetermined threshold value in the final damage discrimination, and is referred to by the discrimination unit 12h.

Note that the setting information 13a, the filter group 13b, and the discrimination reference information 13e are definition information defined in advance, but can be changed as appropriate even during operation of the bill discriminating apparatus 10.

Next, a processing procedure of processing executed by the banknote discriminating apparatus 10 according to the first embodiment will be described with reference to FIG. FIG. 9 is a flowchart illustrating a processing procedure executed by the banknote discriminating apparatus 10 according to the first embodiment.

As shown in FIG. 9, the setting unit 12a performs initial settings such as the number of quantization and the amount of rotational deviation absorption (step S101). Then, in creating the dictionary image, the image data acquisition unit 12b acquires N banknotes of genuine bill images (step S102).

Then, the banknote discriminating apparatus 10 performs a dictionary image creation process for creating a dictionary image for each edge direction using N sheets of genuine bill images as input data (step S103). Note that the processing procedure of the dictionary image creation processing will be described later with reference to FIG.

Then, when determining whether the banknote is correct or not, the image data acquisition unit 12b acquires an image of the banknote that is the target of determination (step S104).

And the banknote discrimination device 10 performs a damage discrimination process using the discrimination target image as input data (step S105). Note that the processing procedure of the damage determination processing will be described later with reference to FIG.

Then, the banknote discriminating apparatus 10 determines whether or not there is a next banknote to be discriminated (step S106), and when there is no next banknote to be discriminated (step S106, Yes), the process is terminated. On the other hand, when the determination condition of step S106 is not satisfied (step S106, No), the banknote determination apparatus 10 repeats the processes after step S104.

In actual operation, the dictionary image creation stage (see the part surrounded by the broken line F1 in the figure) and the damage determination stage (see the part enclosed by the broken line F2 in the figure) are separated. It may be done.

Next, a detailed processing procedure of the dictionary image creation processing (see step S103 in FIG. 9) shown in FIG. 9 will be described with reference to FIG. FIG. 10 is a flowchart illustrating a processing procedure of dictionary image creation processing executed by the banknote determination apparatus 10 according to the first embodiment. Note that the dictionary image creation processing referred to here corresponds to processing from creation of dictionary images for each edge direction from N sheets of regular ticket images.

As shown in the figure, the edge direction detection unit 12c smoothes N sheets of genuine images (step S201), and detects each edge point and its direction for N sheets of genuine images (step S202). .

Subsequently, the valid edge determination unit 12d counts up the edge point amount in each corresponding direction plane (the edge point amount 13c for each direction plane in the storage unit 13) for each edge point of the N sheets of the regular ticket image. (Step S203). Then, the effective edge determination unit 12d stores, on each direction plane, pixels that have obtained edge point amounts that are equal to or greater than a predetermined determination threshold in each direction plane (step S204).

Subsequently, the expansion processing unit 12e expands each effective edge point stored in each direction plane by a predetermined expansion amount (step S205). And the expansion | swelling process part 12e memorize | stores each direction plane as the dictionary image 13d according to direction with respect to the memory | storage part 13 (step S206), and complete | finishes a process.

Next, a detailed processing procedure of the damage determination processing (see step S105 in FIG. 9) shown in FIG. 9 will be described with reference to FIG. FIG. 11 is a flowchart illustrating the processing procedure of the damage determination process executed by the banknote determination apparatus 10 according to the first embodiment. Here, the damage determination process corresponds to a process from comparing and analyzing the determination target image and the dictionary image to determining the correctness of the banknote as the determination target.

As shown in the figure, the edge direction detection unit 12c smoothes the discrimination target image (step S301), and detects each edge point of the discrimination target image and its direction (step S302).

Subsequently, the remaining edge extraction unit 12f stores each edge point of the discrimination target image in each corresponding direction plane (step S303). Then, the remaining edge extraction unit 12f takes each difference between each direction plane and each corresponding dictionary image of the direction-specific dictionary image 13d (step S304), and synthesizes each difference to extract a remaining edge point (step S305). ).

Subsequently, the remaining edge analysis unit 12g removes isolated points from the remaining edge points (step S306) and then analyzes the density of the remaining edge points (step S307).

Then, the determination unit 12h determines whether the density of the remaining edge points analyzed by the remaining edge analysis unit 12g is less than a predetermined threshold (step S308). For the predetermined threshold, the discrimination criterion information 13e is referred to.

If the density is less than the predetermined threshold (step S308, Yes), the determination unit 12h determines that the banknote to be determined is a correct note (step S309), and ends the process.

On the other hand, when the determination condition of step S308 is not satisfied (step S308, No), the determination unit 12h determines that the banknote to be determined is a non-performing bill (step S310), and ends the process.

As described above, in the first embodiment, the setting unit performs an initial setting such as quantization, and the image data acquisition unit generates a plurality of genuine ticket image data when creating the dictionary image. In determining whether the image is correct or not, the image data to be determined is generated, the edge direction detection unit detects the edge point and the direction of the generated image data, and the effective edge determination unit is an effective component that is a component of the dictionary image. The edge point is determined, the expansion processing unit expands the effective edge point by a predetermined amount, the remaining edge extraction unit extracts the remaining edge point from the discrimination target image, and the remaining edge analysis unit detects the density of the remaining edge points. The banknote discriminating apparatus is configured such that the discriminating unit discriminates whether the banknotes are correct based on the detected density of the remaining edge points. Therefore, even if there is diversity in the design and quality of the paper sheets, it is possible to perform the optimum damage determination with high accuracy.

By the way, in Example 1 mentioned above, although the banknote discriminating device demonstrated the case where the dictionary image which expanded each edge point by the predetermined amount was demonstrated, here, a banknote discriminating device is set to the peripheral edge number of each pixel. It is good also as creating a dictionary image based on it. Therefore, in the following, an embodiment of a bill discriminating apparatus that creates a dictionary image based on the number of peripheral edges of each pixel will be described as a second embodiment.

FIG. 12 is a diagram illustrating an outline of dictionary image creation of the bill discriminating apparatus according to the second embodiment. As shown in the figure, the banknote discriminating apparatus according to the second embodiment counts the number of peripheral edges for each pixel in each direction plane, and creates a dictionary image for each direction based on the statistics of the number of peripheral edges. Has the main characteristics.

Note that the peripheral edge here refers to an edge point included in a predetermined vicinity range of a certain pixel.

Specifically, the banknote discriminating apparatus according to the second embodiment counts the number of peripheral edges for each pixel in each directional plane for each of N regular ticket images that are input data of the dictionary image (see FIG. (See (1)). Here, for counting the number of peripheral edges, for example, a counting filter (not shown) having 9 × 9 pixels and all coefficients being 1 can be used.

Note that (1) in the figure shows the number of peripheral edges for each pixel on each direction plane of the direction plane groups 1 to N for counting the peripheral edges corresponding to the 1st to Nth genuine bill images. An example is shown in FIG. Although this example does not limit the contents of the dictionary image creation process of the banknote discriminating apparatus according to the second embodiment, the following description is based on this assumption.

The banknote discriminating apparatus according to the second embodiment calculates N statistics (direction plane groups 1 to N) for the counted number of peripheral edges (see (2) in the figure).

Then, the banknote discriminating apparatus according to the second embodiment determines a reference pixel value for each pixel in each direction plane based on the calculated statistic, and stores the reference pixel value in each direction plane for dictionary images (same as above). (Refer to (3) in the figure). The determination of the reference pixel value will be described later with reference to FIG.

Next, the configuration of the banknote discriminating apparatus 10a according to the second embodiment will be described with reference to FIG. FIG. 13 is a block diagram illustrating the configuration of the banknote discriminating apparatus 10a according to the second embodiment. In addition, in the same figure, the same code | symbol is attached | subjected to the component same as the banknote discrimination | determination apparatus 10 which concerns on Example 1 shown in FIG. 2, and about the component which overlaps with Example 1 mentioned above below. The description of will be omitted.

As shown in FIG. 13, the banknote discriminating apparatus 10 a according to the second embodiment is further provided with a peripheral edge counting unit 12 i in the control unit 12, and an effective edge determining unit 12 d and an expansion processing unit 12 e (see FIG. 2). ) Is different from the banknote discriminating apparatus 10 according to the first embodiment described above in that a statistical processing unit 12j is provided. Moreover, it differs from the banknote discrimination device 10 which concerns on Example 1 mentioned above by the point which equips the memory | storage part 13 with the peripheral edge counting plane 13f instead of the edge point amount 13c according to direction plane.

The peripheral edge counting unit 12i is a processing unit that performs processing for counting the number of edge points detected by the edge direction detection unit 12c within a predetermined vicinity range for each pixel.

Further, the peripheral edge counting unit 12i stores the counted peripheral hedge number for each pixel in each direction plane in the peripheral edge counting plane 13f in the case of creating a dictionary image, and remains in the case of determining whether the damage is correct or not. It is also a processing unit that outputs to the edge extraction unit 12f.

The statistical processing unit 12j refers to the peripheral edge counting plane 13f with respect to the number of peripheral edges for each pixel in each direction plane counted by the peripheral edge counting unit 12i, and calculates a statistical amount for N sheets of the regular image. It is a processing part to perform. Such statistics include maximum values, average values, variance values, standard deviations, and the like.

The statistical processing unit 12j is also a processing unit that performs a process of determining a reference pixel value for each pixel in each direction plane based on the calculated statistics. In addition, regarding the determination of the reference pixel value, various values and parameters included in the statistic can be combined.

Here, the maximum value included in the statistic is V, the average value is μ, the standard deviation is σ, and the parameter is α. For example, the reference pixel value can be set to “V + α” based on the maximum value of the statistic. The reference pixel value may be set to “μ + ασ” based on the average value and standard deviation of the statistics. The calculation formula for the reference pixel value can be appropriately changed according to the design of the banknote, the required discrimination accuracy, and the like.

Further, the statistical processing unit 12j stores the determined reference pixel value for each pixel of each direction plane in each direction plane for the dictionary image, and stores each direction plane after storage as a direction-specific dictionary image 13d to the storage unit 13. Remember.

Next, a processing procedure of dictionary image creation processing executed by the banknote discriminating apparatus 10a according to the second embodiment will be described with reference to FIG. FIG. 14 is a flowchart illustrating a processing procedure of dictionary image creation processing executed by the banknote discriminating apparatus 10a according to the second embodiment.

As shown in the figure, the edge direction detection unit 12c smoothes N sheets of genuine images (step S401), and detects each edge point and its direction for N sheets of genuine images (step S402). .

Next, the peripheral edge counting unit 12i counts the number of edge points (the number of peripheral edges) within a predetermined neighborhood range for each pixel in each direction plane of the direction plane groups 1 to N for N sheets of genuine images. (Step S403).

The statistical processing unit 12j then calculates N (direction plane group 1 to N) statistics for the number of peripheral edges for each pixel counted by the peripheral edge counting unit 12i (step S404). Then, the statistical processing unit 12j determines a reference pixel value for each pixel in each direction plane based on the calculated statistical amount (step S405).

The statistical processing unit 12j stores the determined reference pixel value for each pixel in the storage unit 13 as the direction-specific dictionary image 13d (step S406), and ends the process.

Next, the processing procedure of the damage determination process executed by the banknote determination apparatus 10a according to the second embodiment will be described with reference to FIG. FIG. 15 is a flowchart illustrating the processing procedure of the damage determination process executed by the banknote determination apparatus 10a according to the second embodiment.

As shown in the figure, the edge direction detection unit 12c smoothes the discrimination target image (step S501), and detects each edge point and the direction of the discrimination target image (step S502).

Subsequently, the peripheral edge counting unit 12i stores each edge point in the corresponding direction plane for the discrimination target image (step S503). The peripheral edge counting unit 12i then counts the number of peripheral edges for each pixel in each direction plane, and sets the count value as a pixel value for each pixel in each direction plane (step S504).

Then, the remaining edge extraction unit 12f compares each direction plane with each corresponding dictionary image in the direction-specific dictionary image 13d for each pixel (step S505), and adds pixels having pixel values exceeding the reference pixel value in each direction plane. The remaining edge point is set (step S506). Then, the remaining edge extraction unit 12f synthesizes the remaining edge points on each direction plane (step S507).

Subsequently, the remaining edge analysis unit 12g removes isolated points from the remaining edge points (step S508) and then analyzes the density of the remaining edge points (step S509).

Then, the determination unit 12h determines whether the density of the remaining edge points analyzed by the remaining edge analysis unit 12g is less than a predetermined threshold (Step S510). For the predetermined threshold, the discrimination criterion information 13e is referred to.

If the density is less than the predetermined threshold (step S510, Yes), the determination unit 12h determines that the banknote to be determined is a correct note (step S511), and ends the process.

On the other hand, when the determination condition of step S510 is not satisfied (step S510, No), the determination unit 12h determines that the banknote to be determined is a damaged bill (step S512), and ends the process.

As described above, in the second embodiment, the statistical processing unit determines the reference pixel value for each pixel based on the statistic of the number of peripheral edges for each pixel in each direction plane, which is counted by the peripheral edge counting unit. Therefore, it is possible to add a macro element called a statistic based on a micro element such as the number of peripheral edges for each pixel, and even if there is diversity in the design and quality of paper sheets, Optimal damage determination can be performed with accuracy.

Further, in each of the above-described embodiments, description has been made mainly on the case of determining whether a banknote is correct or not, but there is no particular limitation on paper sheets that are targets of correctness determination. For example, it may be a gift certificate or a passbook.

Further, in each of the above-described embodiments, description has been made mainly on the case of determining whether the paper sheet is correct or not, but the method according to the present invention may be applied to the paper sheet type determination. In such a case, for example, for a banknote, a dictionary image is generated for each denomination of the banknote, and the discrimination target image and the dictionary image are compared for all denominations (that is, for the number of denominations). Then, it is possible to determine that the denomination of the dictionary image when the density of the remaining edge points is a predetermined value or less is the denomination of the determination target image.

As described above, the paper sheet discriminating apparatus and the paper sheet discriminating method according to the present invention are highly accurate and optimal for determining the correct damage and types even when the design and quality of the paper sheets are diverse. This is useful when discrimination is desired, and is particularly suitable for application to a device that discriminates highly circulated paper sheets such as banknotes.

DESCRIPTION OF SYMBOLS 1 Correct image 5 Discrimination object image 10 Banknote discrimination apparatus 10a Banknote discrimination apparatus 11 Line sensor part 12 Control part 12a Setting part 12b Image data acquisition part 12c Edge direction detection part 12d Effective edge determination part 12e Expansion processing part 12f Remaining edge extraction part 12g Remaining edge analysis unit 12h Discriminating unit 12i Peripheral edge counting unit 12j Statistical processing unit 13 Storage unit 13a Setting information 13b Filter group 13c Directional plane edge amount 13d Directional dictionary image group 13e Discrimination reference information 13f Peripheral edge counting plane

Claims (9)

1. A paper sheet discriminating apparatus that discriminates paper sheets based on a captured image of paper sheets,
   When an edge position and an edge direction at the edge position are detected from a learning image that is a captured image of a valid paper sheet, the edge among a plurality of directional dictionary images prepared for each predetermined edge direction range Direction-specific dictionary image generating means for generating the direction-specific dictionary image by allocating the edge position to the corresponding position of the direction-specific dictionary image corresponding to the direction;
   And a discriminating means for discriminating the damage and type of the paper sheet by comparing an input image, which is a captured image of the paper sheet to be discriminated, with the directional dictionary image. Classification device.
2. When the edge position and the edge direction at the edge position are detected from the input image, the direction-specific input generates a direction-specific input image by separating the edge position into the edge direction ranges based on the edge direction. Image generating means;
   A residual edge region is extracted by superimposing the edge positions that exist only in the input image by direction among the input image by direction and the dictionary image by direction related to the same edge direction range for all the edge direction ranges. And a remaining edge extracting means,
   The discrimination means includes
   2. The paper sheet discriminating apparatus according to claim 1, wherein the paper sheet discriminating apparatus determines whether or not the paper sheet to be discriminated is correct based on the remaining edge area extracted by the remaining edge extracting unit.
3. The direction-specific dictionary image generation means includes:
   When the edge direction at the detected edge position is within a predetermined range from the boundary line between two adjacent edge direction ranges, the corresponding dictionary images corresponding to the two edge direction ranges respectively correspond 2. The paper sheet discriminating apparatus according to claim 1, wherein the edge position is assigned to a position.
4. Density detecting means for detecting the density of the remaining edge region extracted by the remaining edge extracting means;
   The discrimination means includes
   The paper sheet discriminating apparatus according to claim 2, wherein the paper sheet discriminating apparatus determines whether or not the paper sheet to be discriminated is correct based on the density detected by the density detecting unit.
5. The direction-specific dictionary image generation means includes:
   The paper sheet according to any one of claims 1 to 4, wherein a pixel to which a predetermined number or more of the edge positions are allocated is set as an effective edge region by using a predetermined number of the learning images. Discriminator.
6. The direction-specific dictionary image generation means includes:
   6. The paper sheet discriminating apparatus according to claim 5, wherein expansion processing for expanding the edge position or the effective edge region to surrounding pixels is performed for each of the direction-specific dictionary images.
7. The direction-specific dictionary image generation means includes:
   The paper sheet discriminating apparatus according to claim 6, wherein the expansion processing is performed by giving priority to the edge direction range corresponding to the dictionary image according to direction.
8. Peripheral edge counting means for counting the number of pixels to which the edge position included in a predetermined neighborhood range of each pixel in the dictionary image according to direction is distributed;
   Statistic calculation means for calculating a statistic obtained by statistically calculating the number counted for each pixel by the peripheral edge counting means for the plurality of learning images; and
   The direction-specific dictionary image generation means includes:
   When the statistic is calculated by the statistic calculator, the directional dictionary image is generated by setting a pixel value of a pixel in the directional dictionary image based on the statistic. The paper sheet discrimination apparatus according to claim 1.
9. A paper sheet discriminating method for discriminating a paper sheet based on a captured image of the paper sheet,
   When an edge position and an edge direction at the edge position are detected from a learning image that is a captured image of a valid paper sheet, the edge among a plurality of directional dictionary images prepared for each predetermined edge direction range A direction-specific dictionary image generation step of generating the direction-specific dictionary image by allocating the edge position to a corresponding position of the direction-specific dictionary image corresponding to a direction;
   And a discrimination step of discriminating the damage and type of the paper sheet by comparing an input image, which is a captured image of the paper sheet to be discriminated, with the directional dictionary image. Classification method.

Images