JP3716873B2 - Paper sheet identification device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a paper sheet identification apparatus that compares the input image obtained by reading a paper sheet with an optical sensor with a plurality of pre-registered reference images to identify the authenticity or type of the paper sheet, and in particular, The present invention relates to a paper sheet identification device that can quickly and accurately identify paper sheets having similar optical patterns between denominations such as dollar bills.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, in a banknote recognition apparatus for identifying the authenticity and denomination of a banknote being conveyed, a reference image obtained by reading the banknote with an optical sensor for each denomination is prepared in advance, and a banknote to be identified is optically detected. In many cases, the input image read in step 1 is compared with each reference image.
[0003]
For example, in Japanese Patent Laid-Open No. 5-143829, if a print pattern of a banknote to be discriminated is read by a sensor, data in the longitudinal direction of the banknote to be discriminated is calculated, and the calculated data is compared with standard data. A bill discriminating device configured to detect continuity of a pattern in a longitudinal direction is disclosed.
[0004]
In Japanese Patent Publication No. 7-89386, when the authenticity of a paper sheet is determined by comparing a detection pattern read by a sensor with a reference pattern stored in advance, the distribution frequency of a large number of paper sheet patterns. A plurality of basic patterns provided separately are provided in advance, and an evaluation score is given depending on which basic pattern is included in the detection pattern. When the matching rate calculated based on the evaluation score is equal to or higher than a predetermined value, the paper sheet Is disclosed as a true / false determination apparatus for paper sheets.
[0005]
[Problems to be solved by the invention]
However, for banknotes having very similar optical patterns between denominations, it is difficult to determine the denomination of banknotes using only the optical patterns, even when the above-described conventional technique is used.
[0006]
For example, US dollar bills include 1 $ bill, 2 $ bill, 5 $ bill, 10 $ bill, 50 $ bill, 100 $ bill, 500 $ bill, 1000 $ bill, 5000 $ bill and 10000 $ bill. However, since the bills of 500 dollars or more are also the same size as 6.8 cm × 16.4 cm, and are the same except for the portrait and the numbers at the four corners, the optical patterns are very similar.
[0007]
For this reason, in order to optically read and identify such banknotes, it is necessary to compare the input image with the reference image in detail, but in reality there are variations in sensitivity characteristics of the image sensor, dirt on the banknotes, etc. Therefore, it is difficult to identify such bills accurately and at high speed.
[0008]
Specifically, the sensitivity characteristics of each pixel of the image sensor vary, and even if the same banknote is transported, it can be detected by different pixels due to left and right shaking during transport. is there.
[0009]
Furthermore, when there are wrinkles, smudges, uneven printing, misprinting or cutting misalignment of the banknote, or when the interval between the image sensor and the banknote varies due to high-speed conveyance, the detection data varies.
[0010]
For this reason, how to quickly and accurately identify paper sheets having similar optical patterns between denominations such as US dollar bills is an important issue.
[0011]
Accordingly, in the present invention, a paper sheet identification device that can accurately and quickly identify paper sheets having similar optical patterns between denominations, such as US dollar bills, by solving the above problems. The purpose is to provide.
[0012]
[Means for Solving the Problems]
To achieve the above object, according to a first aspect of the present invention, a paper for identifying the authenticity or type of a paper sheet by comparing an input image obtained by reading the paper sheet with an optical sensor with a plurality of reference images registered in advance. In the leaf identification device, weight calculation means for dividing the reference data corresponding to each reference image into blocks and calculating a weight for each block based on a distance between corresponding blocks of the plurality of reference data and a variance in each block; Selection means for selecting a plurality of candidate reference data based on the absolute distance between the identification target data corresponding to the input image and each reference data; the distance for each block of each candidate reference data and the identification target data; and A weighted relative distance calculating means for calculating a weighted relative distance between the identification target data and each candidate reference data based on the weight calculated by the weight calculating means; Calculating means, characterized by comprising the identifying means for identifying the authenticity or type of the paper sheet based on a plurality of weighting the relative distance calculated is.
[0013]
In addition, the second invention generates a Laplacian image by quadratic differentiation of the input image or the reference image, divides the Laplacian image into predetermined blocks, and has identification target data or reference data having a histogram for each block. And a weighted relative distance calculating unit that determines a distance between the identification target data and the plurality of candidate reference data for each block, and calculates the distance between the calculated blocks and the block. A weighted relative distance is calculated by adding a product with a corresponding weight.
[0014]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. In this embodiment, the case where the present invention is applied to a bill identifying device for identifying US bills is shown.
[0015]
FIG. 1 is a diagram illustrating a configuration of a banknote recognition apparatus 10 used in the present embodiment.
[0016]
The bill identifying device 10 shown in FIG. 1 is a device that quickly identifies the authenticity and type of a bill read by the sensor unit 11 having an image sensor. Specifically, the concept of weighted relative distance is introduced, Identification is performed by emphasizing a portion where the printing pattern between the banknotes is different.
[0017]
As shown in FIG. 1, the banknote recognition apparatus 10 includes a sensor unit 11, a data creation unit 12, an identification processing unit 13, a weight calculation unit 14, a reference data storage unit 15, and a candidate reference data selection unit 16. And a weighted relative distance calculation unit 17.
[0018]
The identification processing unit 13, the weight calculation unit 14, the candidate reference data selection unit 16, and the weighted relative distance calculation unit 17 are the identification unit, the weight calculation unit, the selection unit, and the weighted relative distance calculation unit described in claim 1, respectively. The data creation unit 12 corresponds to the data creation means described in claim 2.
[0019]
The sensor unit 11 reads banknotes conveyed from a conveyance path (not shown) with an image sensor, and outputs the read input image to the data creation unit 12. In this embodiment, a case where only one side of a banknote is to be identified will be described. However, when both sides of a banknote are to be identified, both sides of the banknote are read by an image sensor.
[0020]
The data creation unit 12 is a processing unit that creates data corresponding to the input image (hereinafter referred to as “identification target data”) from the input image of the bill to be identified read by the sensor unit 11.
[0021]
Specifically, when the data creation unit 12 receives an input image from the sensor unit 11, the data creation unit 12 performs preprocessing such as skew correction, alignment, and smoothing on the input image, and then performs a Laplacian filter. Is applied to create a Laplacian image, the Laplacian image is divided into 12 blocks, and a histogram is taken for each block to be used as identification target data.
[0022]
In order to determine the type of identification target data, various banknotes such as 1 $ banknotes, 5 $ banknotes, 10 $ banknotes are read in advance, and data to be compared for identification (hereinafter referred to as “reference data”). Is required.
[0023]
Therefore, the data creation unit 12 is configured to create reference data corresponding to each reference image when the reference image is read by the sensor unit 11.
[0024]
However, unlike the case where the identification target data is created, this reference data is not created by only one trial, but is created by reading the same denomination banknotes a plurality of times and using the average value.
[0025]
That is, the reference data to be compared with the identification target data is premised on such a plurality of trials because it is not appropriate to be affected by banknote stains or optical variations.
[0026]
For this reason, when creating the reference data for the nth banknote, the nth banknote is read P times to create the corresponding P histogram data, and each histogram data is averaged for each block. As reference data.
[0027]
Specifically, assuming that the histogram of the block m of the p-th histogram data (where 0 ≦ p <P) used to create the reference data for the n-th banknote is snmp (l), n types A histogram tnm (l) (where 0 ≦ l <L, where L is the size of the histogram) of the block m of the reference data corresponding to the banknote of the eye,
tnm (l) = (Σp snmp (l)) / P
It can be obtained from the following formula. However, the variable p is 0 or more and P-1 or less.
[0028]
When the identification processing unit 13 receives the identification target data, the identification processing unit 13 uses the candidate reference data selection unit 16 and the weighted relative distance calculation unit 17 to obtain the weighted relative distances of the two reference data close to the identification target data. It is a processing unit for identifying the authenticity and type of the input banknote based on the weighted relative distance.
[0029]
When reference data is received from the data creation unit 12, weight data between the reference data is calculated using the weight calculation unit 14, and the weight data and the reference data are stored in the reference data storage unit 15. .
[0030]
Specifically, when the identification target data is received, the identification target data and the reference data stored in the reference data storage unit 15 are output to the candidate reference data selection unit 16 to request selection of candidate reference data. When the two candidate reference data are obtained, the reference data, the weight data corresponding to the candidate reference data, and the identification target data are output to the weighted relative distance calculation unit 17.
[0031]
And if this weighted relative distance calculation part 17 calculated the weighted relative distance with each reference | standard pattern, the kind of the said banknote will be identified on the condition that the difference of this weighted relative distance is more than a predetermined threshold value. To do.
[0032]
The weight calculation unit 14 is a processing unit that receives a plurality of reference data from the identification processing unit 13 and calculates weight data between corresponding blocks of the two reference data. For example, when 1 $ banknote, 5 $ banknote and 10 $ banknote are reference banknotes, there are three cases: 1 $ banknote and 5 $ banknote, 1 $ banknote and 10 $ banknote, 5 $ banknote and 10 $ banknote. The weight between each block is calculated.
[0033]
Specifically, the Chebyshev distance (hereinafter referred to as “distance”) dijm between the blocks m of the reference data i and the reference data j is:
dijm = Σl | tim (l) -tjm (l) |
The weight wijm between the blocks m of the reference data i and j is obtained from the following formula:
wijm = dijm / (vim x vjm)
It is obtained from the following formula. However, l is 0 or more and L-1 or less.
[0034]
Note that vnm is the variance of the block m of the reference data corresponding to the nth banknote,
vnm = Σp (Σl | snmp (l) −tnm (l) |) ^ 2 / P
It is obtained from the following formula. However, "^" indicates a power, p is 0 or more and P-1 or less, and l is 0 or more and L-1 or less.
[0035]
Thus, the weight data wijm between the blocks increases as the distance dijm increases, and decreases as the variance of the reference data blocks increases.
[0036]
The reason why the weight data wijm is used in the present embodiment is that the printing pattern of both bills in the block m is considered to be different as the distance dijm increases.
[0037]
That is, in order to identify bills accurately and at high speed, it is important not to focus on similar parts but to focus on parts with different characteristics. In this embodiment, such weight data is introduced. It is said.
[0038]
In addition, the variance here refers to the variance for each block of the plurality of histograms created for the same denomination when obtaining the standard data of the banknote to be referred to. The larger the is, the smaller the weight data in the block.
[0039]
However, for example, in the case of 1 $ banknote and 5 $ banknote, and in the case of 1 $ banknote and 10 $ banknote, the parts with different features are not necessarily the same, so weight data is calculated between two types of banknotes. To do.
[0040]
The reference data storage unit 15 is a storage unit that stores reference data corresponding to each reference image created by the data creation unit 12 and weight data calculated by the weight calculation unit 14.
[0041]
The candidate reference data selection unit 16 is a processing unit that receives a plurality of reference data and identification target data stored in the reference data storage unit 15 and selects two reference data that are close in distance to the identification target data.
[0042]
Specifically, the distance between the blocks corresponding to the identification target data and the reference data is calculated, and the distance between all the blocks is added as the distance between the identification target data and the reference data.
[0043]
Here, when the histogram of the block m of the identification target data is hm (l) (where 0 ≦ l <L), the distance Dim between the blocks and the sum of the distances between the blocks (hereinafter referred to as “distance between blocks”). "Absolute distance")
Dim = Σl | hm (l) -tnm (l) |
Di = Σm Dim
It becomes. However, l is 0 or more and L-1 or less, and m is 0 or more and M-1 or less.
[0044]
As described above, the candidate reference data selection unit 16 calculates the sum of the distances between the blocks of the identification target data and each reference data (hereinafter referred to as “absolute distance”), and calculates the two that are close to the absolute distance. Select as candidate reference data.
[0045]
The weighted relative distance calculation unit 17 receives the two reference data selected by the candidate reference data selection unit 16, the weight data between the two reference data, and the identification target data, and the identification target data and the two reference data Is a processing unit for calculating a weighted relative distance.
[0046]
Specifically, the weighted relative distance si between the identification target data and the reference data i is:
si = Σm (wijm × Dim)
The weighted relative distance sj between the identification target data and the reference data j is
sj = Σm (wijm × Dim)
It becomes. However, m shall be 0 or more and M-1 or less.
[0047]
In this way, the weighted relative distance calculation unit 17 obtains the distance between the corresponding blocks of the identification target data and the reference data, and adds the data obtained by multiplying the weight data of the corresponding block by this distance for all blocks. Thus, the weighted relative distance is obtained.
[0048]
That is, if there is a difference between all the blocks, the banknote to be identified and the banknote to be referred to can be dealt with by an absolute distance obtained by adding all the blocks to the block distance. The US banknote has a characteristic that the optical pattern is similar between the denominations, and therefore, a weighted relative distance is introduced to emphasize a block having a large difference.
[0049]
As described above, the banknote recognition device 10 is configured to select two reference data whose distances are close to the identification target data as candidate reference data, and compare the weighted relative distances with each candidate reference data to identify the banknote. Therefore, even bills having similar optical patterns between denominations can be identified accurately and at high speed.
[0050]
Next, the processing procedure of the banknote identification device 10 will be described.
[0051]
FIG. 2 is a flowchart showing a processing procedure of the banknote recognition apparatus 10 shown in FIG.
[0052]
As shown in FIG. 2, in this banknote identification device 10, the banknote used as a reference object is first read with the image sensor of the sensor part 11, and the reference data corresponding to each read reference image are produced (step 201).
[0053]
Next, the weight calculation unit 14 calculates weight data for each block corresponding to each reference data (step 202), and registers the weight data and each reference data in the reference data storage unit 15 (step 203). The registration of the reference data and the weight data is executed in advance prior to the bill identification.
[0054]
When the banknote to be identified is read by the image sensor of the sensor unit 11 (step 204), the data creation unit 12 performs preprocessing such as skew correction, alignment, and smoothing on the input image. Then, a Laplacian image is created by applying a Laplacian operator to the input image, the Laplacian image is divided into 12 blocks, and identification target data including a histogram of each block is created (step 205).
[0055]
Next, the candidate reference data selection unit 16 selects two candidate reference data close to the identification target data from a plurality of reference data registered in advance (step 206), and calculates a weighted relative distance for each of the two candidate reference data. (Step 207).
[0056]
Then, based on the weighted relative distance calculated by the weighted relative distance calculation unit 17, the identification processing unit 13 identifies the authenticity or type of the banknote (step 208).
[0057]
If the identification process is to be continued, the process proceeds to step 204 and the processes of steps 204 to 208 are repeated (step 209). If the process is not continued, the process is terminated. In a normal process, a plurality of banknotes are set in the apparatus in a lump, and are fed out one by one by a feeding mechanism (not shown) for high-speed conveyance, identification is performed, and the process ends when all banknotes are exhausted. That is, only step 204 to step 209 are executed.
[0058]
Here, for convenience of explanation, the creation of the reference data shown in step 202 has been described in a simplified manner. However, this reference data is created in the same manner as the identification target data creation procedure shown in step 205.
[0059]
Next, the process of each part of the banknote identification device 10 shown in FIG. 1 will be specifically described with reference to FIGS.
[0060]
FIG. 3 is a diagram showing a processing concept of the data creation unit 12 shown in FIG.
[0061]
FIG. 3A is a diagram showing an example of block division performed by the data creation unit 12. As shown in FIG. 3, the data creation unit 12 generates a Laplacian image 30 to which a Laplacian filter is applied from block 0 to block. 11 blocks are divided into 12 blocks.
[0062]
Each block thus divided includes processing units for calculating weight data by the weight calculating unit 14, selecting reference data by the candidate reference data selecting unit 16, and calculating a weighted relative distance by the weighted relative distance calculating unit 17. Become.
[0063]
FIG. 3B is a diagram showing the concept of creating a histogram in each block. As shown in FIG. 3, pixel values are totaled in the X-axis direction for each Y coordinate of block 0, and the sum of the pixel values is calculated. Is stored in the histogram of block 0.
[0064]
Then, the histogram of each block created in this way is normalized as shown in FIG. 3C, and is composed of histograms respectively corresponding to block 0 to block 11 as shown in FIG. 3D. Identification target data is created.
[0065]
Here, the case where the identification target data is created has been illustrated, but when the reference data is created, a plurality of histogram data corresponding to each banknote to be referred to is created, and each histogram is averaged. Processing is added.
[0066]
Next, the processing concept of the weight calculation unit 13 shown in FIG. 1 will be described.
[0067]
FIG. 4 is a diagram illustrating an example of weight data calculated by the weight calculation unit 13 illustrated in FIG. Here, a case is shown in which weight data between the reference data for 1 dollar bills and the reference data for 10 dollar bills is calculated.
[0068]
FIG. 4A is a diagram showing the distance (Dh) between the blocks of the 1-dollar bill reference data and the 10-dollar bill reference data. As shown in FIG. The distance is large and the distance between blocks 2, 4 and 7 is small.
[0069]
Therefore, when comparing the 1 $ banknote and the 10 $ banknote for each block, it can be seen that the patterns printed on the banknotes are similar for the blocks 2, 4 and 7, and the patterns are different for the blocks 3, 5 and 8. Therefore, the weights are reduced for the blocks 2, 4 and 7, and the weights are increased for the blocks 3, 5 and 8.
[0070]
FIG. 4B is a diagram showing the dispersion (Ph) of the reference data for 1 dollar bills. As shown in FIG. 4, in the case of this 1 dollar bill, the dispersion of the block 4 is relatively large. The variance of the blocks is relatively small. For this reason, the weight of the block 4 is relatively small.
[0071]
FIG. 4 (c) is a diagram showing the dispersion (Qh) of the standard data of 10 $ banknotes. As shown in the figure, in the case of 10 $ banknotes, the dispersion of blocks 1 to 4 is relatively large. The variance of other blocks is relatively small. For this reason, the weights of blocks 1 to 4 are relatively small.
[0072]
FIG. 4D is a diagram showing weight data (Wh) of each block between the reference data for 1 $ banknotes and the reference data for 10 $ banknotes. Note that the weight data Wh is as described above.
Wh = Dh / (Ph × Qh)
Calculated by
[0073]
As shown in the figure, the block 3 has a large distance Dh but the variances Ph and Qh are large, so the weight is suppressed to about 10, and the blocks 8 and 9 have a large distance Dh and a small variance. A large weight of about 20 is given.
[0074]
That is, this weight data Wh means that the blocks 8 and 9 should be emphasized for 1 $ bills and 10 $ bills.
[0075]
Next, the processing procedure of the candidate reference data selection unit 16 shown in FIG. 1 will be described.
[0076]
FIG. 5 is a flowchart showing a candidate reference data selection procedure performed by the candidate reference data selection unit 16 shown in FIG.
[0077]
As shown in FIG. 5, the candidate reference data selection unit 16 first selects one reference data from among a plurality of reference data stored in advance in the reference data storage unit 15 (step 501). And the distance between each block of the identification target data (step 502).
[0078]
Then, the absolute distance is calculated by adding the obtained distances of the respective blocks (step 503), and it is confirmed whether there is other reference data that has not been processed (step 504).
[0079]
If unprocessed reference data exists, the process proceeds to step 501 and repeats the processes of steps 501 to 503, and the absolute distance between the identification target data and all the reference data is calculated. Two pieces of reference data are selected from those close to each other (step 505).
[0080]
Next, the processing concept of the weighted relative distance calculation unit 17 shown in FIG. 1 will be described.
[0081]
FIG. 6 is a diagram showing a processing concept of the weighted relative distance calculation unit 17 shown in FIG. Here, it is assumed that the reference data of 1 $ banknote and 10 $ banknote is selected as candidate reference data.
[0082]
As shown in FIG. 6, the weighted relative distance calculation unit 17 obtains distances A0 to A11 between the 1-dollar bill reference data and the identification target data blocks, and the product of the distance between the blocks and the weight data W0 to W11. Is added to calculate the weighted relative distance SA between the reference data of 1 dollar bill and the identification target data.
[0083]
Next, distances B0 to B11 between the 10-dollar bill standard data and the identification target data block are obtained, and the distance between the blocks and the product of the weight data W0-W11 are added to identify the 10-dollar bill reference data. The weighted relative distance SB of the target data is calculated.
[0084]
The weighted relative distances SA and SB are respectively output to the identification processing unit 13, and the identification processing unit 13 identifies the banknote based on the weighted relative distances SA and SB.
[0085]
Specifically, when the difference in weighted relative distance | SA−SB | is equal to or smaller than a predetermined value, it is determined that the banknote to be identified does not correspond to any banknote, and | SA−SB | If the value is larger than the value, a bill corresponding to the reference data having a smaller weighted relative distance is output as the identification result.
[0086]
As described above, in the present embodiment, when the sensor unit 11 reads the identification target banknote, the data creation unit 12 creates identification target data corresponding to the input image, and the candidate reference data selection unit 16 Two candidate reference data close to the identification target data are selected. The weighted relative distance calculation unit 17 calculates the weighted relative distance between the two candidate reference data and the identification target data, and the identification processing unit 13 identifies the authenticity and type of the banknote based on the weighted relative distance. Therefore, the following effects can be obtained.
[0087]
1) A bill having an optical pattern similar to each denomination such as a US dollar bill can be accurately and quickly identified.
[0088]
2) The banknote can be identified based on the difference in characteristics between the banknotes to be referred to.
[0089]
In the present embodiment, the Laplacian image obtained by applying the Laplacian operator to the input image is divided into 12 blocks, and the histogram of each block is used as identification target data. Therefore, the Laplacian image or the differential image itself is identified. Compared to the case of target data, the amount of calculation is reduced and high-speed identification processing can be performed.
[0090]
However, the present invention is not limited to the case where the histogram created for each block of the Laplacian image is used as the identification target data, and an input image, a Laplacian image, a differential image, or the like can be used as the identification target data. In this case, the reference data needs to have a corresponding structure.
[0091]
In this embodiment, the candidate reference data selection unit 13 selects two candidate reference data. However, the present invention is not limited to this, and three or more candidate reference data are selected. It can also be applied to cases.
[0092]
For example, when the candidate reference data selection unit 13 selects three candidate reference data, first, two of the three candidate data are selected, and the above-described weighted relative distance is calculated for the two candidates. Then, the smaller weighted relative distance is left as a candidate. Next, a weighted relative distance is further calculated between the candidate and the remaining one of the three candidates, and the smaller weighted relative distance is used as the final result.
[0093]
Furthermore, although the case where the Chebyshev distance is used has been described in the present embodiment, the present invention is not limited to this, and a correlation coefficient or the like can be used instead of the distance concept.
[0094]
In addition, the weight calculation method is proportional to Dh and has been inversely proportional to the product of Ph and Qh. However, the present invention is not limited to this, and instead of this calculation method, Dh is a constant value. As described above, it is possible to use a calculation method such as setting the weight of a block with Ph and Qh equal to or less than a certain value to 1 and setting the others to 0.
[0095]
Moreover, although this embodiment demonstrated the case where this invention was applied to the banknote identification device which identifies a US banknote, this invention is not limited to this, It applies to various paper sheets, such as securities. be able to.
[0096]
【The invention's effect】
As described above, in the first invention, the reference data corresponding to each reference image is divided into blocks, and the weight for each block is determined in advance based on the distance between the corresponding blocks of the plurality of reference data and the variance within each block. When the input image is read, first, a plurality of candidate reference data is selected based on the absolute distance between the identification target data corresponding to the input image and each reference data. Based on the distance for each block of the candidate reference data and the identification target data and the weight calculated by the weight calculation means, the weighted relative distance between the identification target data and each candidate reference data is calculated, and then the plurality of calculated weights are calculated. It identifies the authenticity or type of the paper sheet corresponding to the identification target data based on the relative distance,
1) Using the above-mentioned weighted relative distance, which is useful for emphasizing a portion where the printed pattern between each banknote is different, a banknote having a similar optical pattern between each denomination such as a US dollar banknote can be accurately and quickly. It becomes possible to identify.
2) By introducing the concept of the above weighted relative distance, it becomes possible to identify banknotes based on the difference in characteristics between the banknotes to be referred to.
There is an effect.
[0099]
The second invention generates a Laplacian image by second-order differentiation of the input image or the reference image, divides the Laplacian image into predetermined blocks, and obtains identification target data or reference data having a histogram for each block. Since it is configured so as to be created, it is possible to perform high-speed identification processing with a smaller amount of computation than when the input image, Laplacian image, or differential image itself is used as identification target data.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a banknote identification device used in the present embodiment.
FIG. 2 is a flowchart showing a processing procedure of the banknote recognition apparatus shown in FIG.
FIG. 3 is a diagram showing a processing concept of a data creation unit shown in FIG. 1;
4 is a diagram showing an example of weight data calculated by a weight calculation unit shown in FIG.
FIG. 5 is a flowchart showing a candidate reference data selection procedure performed by the candidate reference data selection unit shown in FIG. 1;
6 is a diagram showing a processing concept of a weighted relative distance calculation unit shown in FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 ... Banknote identification apparatus, 11 ... Sensor part, 12 ... Data preparation part,
13 ... Identification processing unit, 14 ... Weight calculation unit, 15 ... Reference data storage unit,
16 ... Candidate reference data selection unit, 17 ... Weighted relative distance calculation unit,

Claims (2)

In a paper sheet identification device for identifying the authenticity or type of the paper sheet by comparing an input image obtained by reading the paper sheet with an optical sensor with a plurality of pre-registered reference images,
Weight calculation means for dividing the reference data corresponding to each reference image into blocks and calculating a weight for each block based on a distance between corresponding blocks of the plurality of reference data and a variance in each block;
Selection means for selecting a plurality of candidate reference data based on an absolute distance between identification target data corresponding to the input image and each reference data;
Weighted relative distance calculating means for calculating a weighted relative distance between the identification target data and each candidate reference data based on each candidate reference data and the distance of each block of the identification target data and the weight calculated by the weight calculating means. When,
An identifying means for identifying the authenticity or type of the sheet based on a plurality of weighted relative distances calculated by the weighted relative distance calculating means. A data creation means for creating a Laplacian image by quadratic differentiation of the input image or the reference image, dividing the Laplacian image into predetermined blocks, and creating identification target data or reference data having a histogram for each block; Equipped,
The weighted relative distance calculation means obtains a distance between the identification target data and the plurality of candidate reference data for each block, and adds a product of the obtained distance between the blocks and a weight corresponding to the block to perform weighting. The paper sheet identification apparatus according to claim 1, wherein a relative distance is calculated.

Images