JPH0327955B2 - - Google Patents

Description

[Detailed Description of the Invention] [Table of Contents] Overview Industrial Application Fields Conventional Technology Problems to be Solved by the Invention Means for Solving the Problems (Fig. 1) Working Examples (a) Overall Explanation (Fig. 2, Fig. 3) (b) Explanation of group division (Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8) (c) Explanation of other embodiments Effects of the invention [Summary] In a standard pattern creation method for a paper sheet discrimination machine that identifies paper sheets by comparing the reading output of the paper sheet with a reference pattern, By extracting items with a high degree of approximation as a group and creating reference patterns from the extracted group, it is possible to accurately discriminate with a small number of reference patterns.

[Industrial application field]

The present invention relates to a method for creating a reference pattern for a machine for validating paper sheets such as banknotes, and in particular, a method for creating a standard pattern for a paper sheet validating machine that can reduce the number of standard patterns to be created by taking into account the traveling conditions of paper sheets. This invention relates to a reference pattern creation method.

Banknote validators are widely used in automatic service machines such as automatic teller machines and automatic ticket vending machines.

As shown in Fig. 9A, such a banknote validating machine has reading heads HD1 to HD3 arranged so as to read a specific trace line of a banknote CC to be inputted, and a banknote CC to be inputted between guides GD as shown in the figure. Bill CC is detected by sensors S1 and S2, and head HD1~
The HD 3 reads position-corresponding patterns T1 to T3 of banknotes CC sent by a feed roller (not shown).
Then, the authenticity determination unit GJ compares the read output with the reference pattern (recognition unit) in the memory PM and recognizes the banknote.
This is to determine the authenticity of CC.

If a standard pattern is prepared for each trace line T1 to T3 for this banknote CC, identification is possible in principle. Even if the same banknote is inserted, the reading signal will change.

That is, by shifting the banknote CC in the X direction or obliquely moving it in the R direction, the heads HD1 to HD3 may read patterns outside the predetermined trace lines T1 to T3 of the banknote CC, or read new and old banknotes. The read signal changes as shown in FIG. 9B depending on the condition of the banknote.

Therefore, it is desired to create a reference pattern that can be used to identify these banknotes.

[Conventional technology]

To create this reference pattern, for example, as seen in Patent Application Publication No. 134699/1983,
Conventionally, a plurality of banknotes are read in advance in various running conditions, the reading outputs are collected, the reading outputs are classified into groups using the physical position of the running condition (adjacent, skewed, etc.) as a parameter, and each group is divided into groups. A standard pattern was created.

For example, when considering the stacking of banknotes,
For CC, as shown in Fig. 10A, we obtained a read output for 15 trace lines within the maximum range of offset, and physically divided this trace line into recognition bands A to E according to the possible offset state. It was defined as a group. That is, the banknotes are read by giving the banknote CC a forced amount of aggregation, and the readout of 15 trace lines is obtained, and this readout is divided into physical groups A to E as shown in FIG. 10B. Was.

Furthermore, in consideration of skew, banknotes CC have been read with a forced skew amount, and the banknotes have been similarly divided into groups.

These are performed one after another on a plurality of different banknotes, and the reading outputs are collected, and a reference pattern, which is a recognition unit, is created from the reading outputs classified into groups based on such physical position parameters.

For example, considering 15 trace lines for shift and 8 trace lines for skew, it is necessary to create 120 reference patterns.

Furthermore, if the input direction is not regulated, it is necessary to consider four input directions, and in addition, if multiple denominations can be discriminated, this is necessary for each denomination.

[Problem that the invention seeks to solve]

According to such conventional reference pattern creation methods, the read output for creating the reference pattern is divided into groups based on physical parameters such as offset and skew, and a reference pattern is created for each group, resulting in a huge amount of work. There is a problem in that a reference pattern must be created, resulting in an increase in the capacity of the reference pattern memory.

SUMMARY OF THE INVENTION An object of the present invention is to provide a method for creating a reference pattern for a paper sheet discrimination machine, which can reduce the number of required reference patterns and reduce the capacity of the reference pattern memory without reducing the discrimination accuracy. shall be.

[Means for solving problems]

FIG. 1 is a diagram explaining the principle of the present invention.

In the present invention, as shown in FIG. 1A, in a method for creating a reference pattern for a paper sheet discrimination machine that reads paper sheets to be discriminated and compares the read output with a reference pattern to perform discrimination, A step (S1) of reading paper sheets while changing the running state and collecting a plurality of reading outputs in a memory, and a pattern creation section extracting a group of items with a high degree of similarity from the output distribution of the plurality of reading outputs. Step (S2) and
The pattern creation section has a step (S3) of creating a reference pattern from the read outputs belonging to the extracted group, and the extraction step (S2) includes a step of calculating a distance distribution from the center of the distribution, removing the farthest distance and returning to the calculation as a new target read output group a predetermined number of times, and calculating the variance of the distance distribution each time; , a step of extracting the target read output group with the lowest variance as one with a high degree of approximation; and a step of calculating the read output group excluding the extracted read output group as the target read output group. and a return step.

That is, whereas conventional group division was based on physical position elements, the present invention attempts to divide groups based on the obtained output distribution.

[Effect]

The starting point of the present invention is to focus on the output distribution of read output. In other words, if we plot the strength or density output of the collected reading output (hereinafter referred to as sample) obtained from each paper sheet, excluding the physical position parameter, and obtain the distribution, as shown in Figure 1B, each sample is equal to the total number of samples. The distribution is not scattered over the entire surface within a certain circle (range) from the distribution center, but is concentrated at a plurality of specific positions within the circle.

Therefore, from the output distribution of the entire sample, individual concentrated groups (recognition units) a, b, c, d...
can be divided into groups by extracting them according to the degree of approximation of the output.

Such a distribution is not a recognition standard for each position based on conventional position parameters, but represents data obtained from the entire sample as one recognition standard. For example, the recognition unit bands B and D in Figure 10A are In the conventional sample distribution based on position parameters, items with a high degree of approximation belong to different recognition standards, but in the present invention, they belong to the same recognition standard. Therefore, the number of divided groups becomes small. Also, the broken image like trace line 13 is similar to trace line 1.
It belongs to an independent recognition standard different from that of 4, and can be divided into groups with significant differences.

〔Example〕

(a) Overall description of one embodiment FIG. 2 is a block diagram of one embodiment of the present invention, and FIG. 3 is an explanatory diagram of read output.

In the figure, 1 is a banknote data collection unit, which reads patterns on a plurality of trace lines of the inserted banknote CC and outputs the read output as in FIG. 9A, and 2 is a memory, which collects banknote data. 3 which stores the read output from part 1;
is a group division/reference pattern creation unit, which is composed of a computer and divides the read outputs (hereinafter referred to as samples) collected in the memory 2 into groups based on the output distribution, and creates reference patterns for the divided groups. It is.

In this embodiment, the banknote data collecting section 1 is provided with a head for reading two trace lines on each of the front and back sides of the input banknote, and obtains reading output for four trace lines in one run. Each readout is the third
As shown in the figure, it is indicated by the voltage value from the head,
Divided into nine zones, D ₀ to D ₈ , each zone
The voltages from D ₀ to D ₈ become data for one trace line.

Then, the banknotes are put in at a plurality of offset amounts and skew amounts, and a large number of samples in each of the nine zones are obtained.

Also, a sample is obtained in the same manner by changing the feeding direction.

For example, 4000 specimens for a total of 1000 specimens are obtained, stored in the memory 2, and data collected.

Each sample in the memory 2 is divided into groups by a group division/reference pattern creation section 3 based on the output distribution through calculations, and a reference pattern for the divided group is further created.

(b) Explanation of group division FIG. 4 is a flow diagram of group division processing according to an embodiment of the present invention, FIG. 5 is an explanatory diagram of distance calculation from the center of distribution in FIG. 4, and FIG. A histogram explanatory diagram of the calculated distance, Figure 7 is the 4th
FIG. 8 is an explanatory diagram of optimal group selection, and FIG. 8 is an explanatory diagram of group division.

In the present invention, optimal groups (recognition criteria) a, b,
In order to obtain c and d, the degree of approximation of the sample is determined.
For this purpose, the following statistical calculations are performed by group division and
This is performed by the reference pattern creation section 3.

First, the distribution center and distance distribution of the target sample/
Find the variance.

For this reason, the average values ₀ to ₈ of each zone D ₀ to D ₈ of the target sample and their variances V ₀ to V ₈ are determined as the center of the sample.

That is, if the number of samples is N, each zone D ₀ to _{D 8}
The sum a ₀ to _{a 8} of N samples of is obtained by.

Next, the sum of squares b ₀ of N samples of each zone D ₀ to D ₈
Find ~ _b8 .

From formula (1), the average value of each zone D ₀ to D _{8 is 0} to _8.
teeth, From equation (2), the variance V ₀ of N samples in each zone D ₀ to D ₈
~ _V8 becomes.

In this way, each zone as the center of the specimen
Once the average value _K (k=0 to 8) and variance V _K for each of D ₀ to D ₈ are determined, the variance V _D from the center O (FIG. 8A) of the distribution of the entire sample is determined.

This can be done by calculating the statistical distance l from the center and finding the variance V _D of the distance l.

Calculation of this distance is multidimensional with k zones and cannot be displayed in the drawing, but for simplicity, it is displayed as two dimensions, X and Y, with two zones, as shown in FIG. 5.

For example, if zone D ₀ is taken as X and zone D ₁ as Y, then
The coordinates M _X and M _Y of the distribution center _AC for N samples are ₀ to ₁ in equation (3), and the respective variances V _X and V _Y are V ₀ and V ₁ in equation (4). .

If the variances V _X and V _Y are normally distributed, based on well-known probability theory, an effective sample exists within the dotted line in FIG. 5 when the distribution center is O.

Therefore, in order to check whether each sample is within the dotted line in FIG. 5, the distance l between each sample and the distribution center is determined.

This distance l is not the usual Euclidean distance, but the Mahalanobis distance shown below, which takes probability into consideration.

l _i = (D _0i − ₀ ) ² /V ₀ + (D _1i − ₁ ) ² /
V ₁ +...(D _8i - ₈ ) ² /V ₈ ...(5) That is, the center of the distribution of the total number of samples N ( ₀ ,...
D ₈ ) and calculate the Mahalanobis distance from this center to each sample.

Using the Mahalanobis distances thus obtained, a histogram (distance distribution) for the number of samples N is created as shown in FIG.

This histogram shows a chi-square distribution when n variables are normally distributed independently of each other in statistics.

Next, the variance V _D of this histogram is determined by the following formula.

V _D = _N 〓 ⁱ⁼¹ 1/N (l _i ) ² - (1/N _N 〓 ⁱ⁼¹ (l _i )) ² ……(6) Next, check whether the step has been performed a predetermined number of times. If the process is not repeated a predetermined number of times, a predetermined number of samples at the farthest distance from the distribution center in FIG. 6 are removed and the steps are repeated.

The number of samples N _K to be removed at this time is determined by the condition that the limit distance is K, and the condition that _i > K is satisfied for the distance l _i , and the removal ratio is
The limit distance K is set so that the value of T _H =N _K /N (N is the number of samples before removal) becomes a predetermined value, and the number of samples N _K is removed.

In short, remove a fixed proportion of the farthest samples.

Removing the farthest sample from the distribution center in this way limits the sample object to the area inside the dotted line in the figure, as shown in FIG. 8A.

Therefore, when the sample center is found again in the step, the sample center moves to O ₁ as shown in Fig. 8B, and in the same way, the farthest sample from this distribution center O is removed, and the dotted line in Fig. 8B limited to the specimens within.

These steps are repeated, and when the number of removals reaches the specified j times, a dispersion diagram of the distance distribution for j times is obtained as shown in Figure 7, and the dispersion V _D
Find the lowest value of and extract the target samples as a group.

This target sample group has the lowest variance V _D , so
This is the optimal group with the least variation.

For example, in the example of FIG.
By repeating , the distribution center O becomes as shown in Fig. 8B,
The sample moves as shown in C and D, and the sample in circle b whose variance is the lowest value is extracted as one of the optimal groups.

Next, step,
, for example, extract the sample in circle c as one of the optimal groups, and then repeat step , for the samples that were similarly removed (the samples in a and d in Figure 8D). For example, the samples in circle a are extracted as the optimal group, and the samples in circle d are similarly extracted.

In this way, optimal groups with a high degree of approximation, that is, recognition criteria, are extracted from the samples as the output distribution, and all samples are divided into groups based on the output distribution.

As is well known, the reference pattern is created by calculating the average value of each zone of each sample group in each divided sample group to create a reference pattern for each divided group.

(c) Description of other embodiments In the embodiments described above, the identification standard pattern of banknotes was explained as an example, but the present invention can also be applied to paper sheets that are distributed to an unspecified number of people, such as checks and securities.

Further, the reading head may be either a magnetic head or an optical head, and can be applied with or without regulation of the input direction.

Although the present invention has been described above using one embodiment, the present invention can be modified in various ways according to the gist of the present invention, and these are not excluded from the present invention.

〔Effect of the invention〕

As explained above, according to the present invention, items with a high degree of approximation are extracted as a group from the distribution of reading outputs of paper sheets, and the groups are divided. Therefore, the number of reference patterns to be created can be significantly reduced. This has the effect of significantly reducing the capacity of the reference pattern memory.

Also, since the output distribution of read output is used,
It also has the effect that a highly accurate reference pattern can be created with a small number of patterns, and is used as a reference pattern for a paper sheet discrimination machine, contributing to enabling highly accurate discrimination.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of the principle of the present invention, Fig. 2 is a configuration diagram of an embodiment of the present invention, Fig. 3 is an explanatory diagram of the readout output of the embodiment of Fig. Embodiment processing flow diagram, FIG. 5 is an explanatory diagram of distance calculation in the embodiment of FIG. 4, FIG. 6 is an explanatory diagram of the distance histogram of the embodiment of FIG. 4, and FIG. 7 is an illustration of the optimal group of the embodiment of FIG. 4. FIG. 8 is an explanatory diagram of group division according to the present invention, FIG. 9 is an explanatory diagram of banknote discrimination, and FIG. 10 is an explanatory diagram of a conventional discrimination reference pattern creation method. In the figure, CC...Banknotes, HD1-HD3...Head, 1...Data collection unit, 2...Memory, 3...
Group division/reference pattern creation section.

Claims (1)

[Scope of Claims] 1. A method for creating a standard pattern for a paper sheet classification machine that reads paper sheets to be classified and performs the classification by comparing the read output with a standard pattern, comprising: running a plurality of sheets of paper; A step (S1) in which reading is performed while changing the state and a plurality of read outputs are collected in memory; and a step (S2) in which the pattern creation section extracts a group of highly similar items from the output distribution of the plurality of read outputs. , a step (S3) in which the pattern creation section creates a reference pattern from the read outputs belonging to the extracted group, and the extraction step (S2) includes: a center of the distribution of the target read output group; The process of calculating the distance distribution from the center of the distribution, removing the one with the farthest distance, and returning to the calculation as a new target read output group is repeated a predetermined number of times, and the variance of the distance distribution each time is calculated. a step of extracting a target read output group with the lowest variance as one with a high degree of approximation; and a step of calculating a target read output group excluding the extracted read output group from the read output group. 1. A method for creating a reference pattern for a paper sheet discriminator, comprising the step of returning to step 1.