JPH11154253A - Method for judging authenticity of paper sheet - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a paper sheet authenticity judging method capable of systematically judging the authenticity of each paper sheet by using plural featured values. SOLUTION: This authenticity judging method consists of a step for receiving an image of a paper sheet to be inspected 1, a step for calculating plural featured values from the image 2 and a step for judging the authenticity of the paper sheet to be inspected based on the obtained featured values. This authenticity judging system is provided with a preprocessing for previously finding out weights of respective featured values by the use of a genetic algorithm, and an authenticity judging processing part 4 wherein a specially weighted Euclid distance is calculated 3 considering the weights of respective featured values based on the featured values obtained from the paper sheet to be inspected and previously found weights for respective featured values, and the authenticity of the paper sheet to be inspected is judged based on the calculated specially weighted Euclid distance.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for judging the authenticity of paper sheets such as bills and securities.

[0002]

2. Description of the Related Art As a conventional method of determining the authenticity of a paper sheet, an image of a banknote to be inspected is captured by an image sensor, the characteristic amount of the banknote to be inspected is calculated from the image, and inspection is performed based on the obtained characteristic amount. There is a method of determining the authenticity of a target bill.

At this time, it is conceivable to judge the authenticity of the banknote to be inspected comprehensively using a plurality of feature amounts. However, some features may not be able to handle all features uniformly, for example, those that easily reflect disturbances such as wrinkles, and those that are not. It was difficult to determine the authenticity of the target bill.

[0004]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method for judging the authenticity of a paper sheet by using a plurality of feature values to comprehensively determine the authenticity of the paper sheet. I do.

Further, according to the present invention, when the authenticity of a paper sheet is comprehensively determined using a plurality of characteristic amounts, a suitable weight can be easily set for each characteristic amount. It is an object of the present invention to provide a method of judging the authenticity.

[0006]

According to the present invention, there is provided a method for judging the authenticity of a paper sheet, wherein an image of a paper sheet to be inspected is taken, a plurality of feature amounts are calculated from the image, and a plurality of obtained features are obtained. In the method of determining the authenticity of the sheet to be inspected based on the amount, in the pre-processing to determine the weight for each feature amount in advance by a genetic algorithm, and the feature amount obtained from the sheet to be inspected, A specially weighted Euclidean distance in consideration of the weight for each feature amount is calculated based on the weight for each feature amount obtained in advance, and the paper sheet to be inspected is calculated based on the calculated specially weighted Euclidean distance. A true / false judgment process for judging the kind of true / false.

A plurality of feature amounts obtained from the paper sheet to be inspected are defined as x _i (i = 1, 2,... N), an average of a genuine bill sample for the feature amount x _i is m _xi , and a feature amount x _i , The variance of a genuine note sample with respect to σ _xi ² , the number of features
_{2. The} method according to claim 1, wherein the special weighted Euclidean distance is calculated based on, for example, Equation 4 below, where w _i is the weight corresponding to _i. .

[0008]

(Equation 4)

The preprocessing includes, for example, a feature value x _{i for} each of a plurality of genuine bill samples and a plurality of fake bill samples prepared in advance.
A first step of obtaining a based on the feature amount obtained from a sample of authentic bill, and the average m _xi genuine note for the feature amount x _i,
The calculated and variance sigma _xi ² genuine note for the feature amount x _i 2
A third step of generating a predetermined number of individuals each consisting of a gene representing a weight corresponding to each of the feature amounts x _{1 to} x _n to generate an initial individual population;

[0010] For each individual, a special weighted Euclidean distance L is calculated for each genuine note sample and each counterfeit note sample based on the above equation 1, and based on the calculation result, the following equation 5 is used. A fourth step of obtaining a fitness F to be represented, a fifth step of determining whether or not an end condition is satisfied, and a fifth step of creating a new individual population in consideration of the fitness F if the end condition is not satisfied. The one having the seventh step of repeating the fourth to sixth steps until it is determined that the termination condition is satisfied in the sixth step and the fifth step is used.

[0011]

(Equation 5)

For example, each feature quantity x _i is set to i (i = 1, 2,... N) for each area of the paper sheet, and b _{i, j} is set to the projection result of the image obtained from each area i. And the reference data corresponding to the area i obtained from the genuine bill sample is a
_{When i and j} are used, the one represented by the following Expression 6 is used.

[0013]

(Equation 6)

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in which the present invention is applied to a bill authenticity judging method will be described below.

[1] Description of true / false judgment processing

FIG. 1 shows the procedure of the authenticity judgment processing.
First, data is taken in from a banknote to be inspected (step 1). As the data, for example, as shown in FIG. 2, the image data of each region e ₁ to e _n inspected banknotes 100 are used. In this example, n = 8.

Next, a feature value is calculated (step 2). Feature amount, for example, is determined for each of the regions e ₁ to e _n. In this example, the projection result of the image data obtained from the area i (i = 1, 2,..., N) is referred to as acquisition data bi _{, j,} and the reference data corresponding to the area i is _{ai, j}
Then, the feature quantity x _i for this area is expressed by the following equation (7). The reference data a _{i, j} is data obtained from a plurality of genuine bills.

[0018]

(Equation 7)

In this example, there are eight types of feature values x ₁ to x ₈ .

Next, a specially weighted Euclidean distance L taking into account the weight for each feature value is calculated (step 3). First, a special weighted Euclidean distance will be described with reference to FIG. A feature amount space based on _i feature amounts xi is assumed as shown in FIG. In FIG.
The filled circles represent the distribution of genuine bill samples, and the crosses represent the distribution of fake bill samples. The specially weighted Euclidean distance for the test object having the distribution indicated by the symbol “△” refers to the distance from the center O of the distribution of the genuine sample to the distribution indicated by the symbol “△”. The special weighted Euclidean distance for each genuine note sample refers to a distance from the center O of the genuine note sample distribution to the distribution of each genuine note sample. The special weighted Euclidean distance for each counterfeit sample refers to a distance from the center O of the distribution of the counterfeit note to the distribution of each counterfeit sample.

In this example, a plurality of feature amounts obtained from the paper sheet to be inspected are defined as x _i (i = 1, 2,... N), and the average of the genuine bill sample for the feature amount x _i is m _xi , Assuming that the variance of a genuine note sample with respect to the feature quantity x _i is σ _xi ² , the number of feature quantities is n, and the weight corresponding to the feature quantity x _i is w _i , a special weight with a weight for each feature quantity is taken into account. The Euclidean distance L is calculated by the following Expression 8.

[0022]

(Equation 8)

The method of obtaining the feature weight w _i in Equation 8 will be described later. In multivariate analysis, “weighted Euclidean distance” which is generally used
Is defined in a form in which the weight w _i of the feature amount is omitted in Expression 8. In this specification, the generally used "weighted Euclidean distance" is called "general weighted Euclidean distance", and the distance expressed by Expression 8 is called "special weighted Euclidean distance". To distinguish them.

When the special weighted Euclidean distance L is calculated, it is determined whether or not the calculated value L is equal to or less than a predetermined threshold (step 4). If the specially weighted Euclidean distance L is equal to or less than a predetermined threshold, the inspection target banknote is determined to be genuine (step 5). If the specially weighted Euclidean distance L is larger than the predetermined threshold, the inspection target banknote is false. The ticket is determined (step 6).

[2] Description of Feature Value Weight Setting Process FIG. 4 shows a feature value weight setting process. The process of setting the feature amount weight is a process that is performed in advance before performing the authenticity determination process.

A plurality of genuine bill samples and a plurality of fake bill samples are prepared in advance. Data is acquired from each of these prepared specimens (step 11).
As the data, for example, as described above, the image data of each region e ₁ to e _n of each sample is used.

Next, for each sample, obtained feature amount x _i (step 12).

[0028] Based on the feature amount obtained from a sample of authentic bill, and the average m _xi genuine note for the feature amount x _i, characteristic amounts x _i
And the variance σ _xi ² of the genuine bill with respect to
3).

Next, weight optimization processing is performed by a genetic algorithm (GA).

First, an initial population is created (step 14). That is, a preset number, for example, 100
Individuals are created by random numbers.

FIG. 5 shows an individual. Each individual, when the feature value are n, and a gene w ₁ to w _n representing the weight for each feature quantity x ₁ ~x _n.

Each gene is represented by, for example, 4 bits. An example of the relationship between the 4-bit gene pattern and the weight value is shown in Table 1 below.

[0033]

[Table 1]

Next, the fitness F is calculated for each individual (step 15). The fitness F is calculated based on the following equation 9.

[0035]

(Equation 9)

That is, first, a specially weighted Euclidean distance L is calculated for each genuine note sample and each counterfeit note sample, based on Equation 8 above, for each individual. FIG. 6 shows the specially weighted Euclidean distance L of each genuine note sample and each counterfeit note sample obtained for a certain individual.

As shown in FIG. 6, L _f (min) in Equation 9 represents the smallest specially weighted Euclidean distance in a counterfeit sample. L _t (max) represents the largest specially weighted Euclidean distance in a genuine note sample.

Further, L _t (i) indicates each of the Euclidean distances of a genuine note sample that is larger than the smallest specially weighted Euclidean distance L _f (min) in a counterfeit sample.

L _f (j) indicates the special weighted Euclidean distance of the counterfeit sample which is smaller than the largest special weighted Euclidean distance L _t (max) in the true sample. ing.

C in Equation 9 is a constant. P ₁ in the formula 9, a SPECIAL weighted Euclidean distance group for legitimate bill represents the overlap of the magnitude of the SPECIAL weighted Euclidean distance group against counterfeit, the larger the P _1, the overlapping portion It is large and the fitness F is small. Finally, it is preferable that P ₁ becomes zero.

P in Equation 9_TwoIs a special weight
Euclidean distance group and special weight for counterfeit
Represents the size of the separation from the Euclidean distance group with
, P _TwoIs larger, the overlap is larger, and the fitness F is
Become smaller. And finally, P_TwoIs negative
Is preferred.

That is, it can be said that an individual having a higher fitness F is an optimal individual.

When the fitness F has been calculated for all individuals in this way, elite saving is performed (step 16). That is, of all the individuals, the individual having the highest fitness F is selected as an elite and is separately copied and stored.

Next, it is determined whether the termination condition is satisfied (step 17). As the end condition, for example, whether or not a predetermined number of generations have been changed is used.

If the termination condition is not satisfied, a reproduction process is performed (step 18). That is, 100 individuals are selected with a probability proportional to the fitness.

Next, a crossover process is performed (step 1).
9). That is, any two individuals are selected from the individuals selected in step 18, and crossover processing is performed between the selected individuals.

Next, a mutation process is performed (step 20). That is, an individual is selected with a certain probability, and the value of an arbitrary gene of the individual is inverted with another probability.

When the mutation process is performed, step 15
Return to Then, the processing of step 15 to step 20 is repeated for the individual generated in step 18 to step 20 and the individual stored in the elite, and
If it is determined that the termination condition is satisfied in, the weight for each feature value is determined (step 21). That is, the weight corresponding to each gene of the individual having the highest fitness F among the remaining individuals is set as the weight for each feature amount.

When the weight is set for each feature value as described above, the following effects can be obtained as compared with the case where no weight is set.

That is, when a general weighted Euclidean distance is measured for a plurality of genuine bill samples and a plurality of fake bill samples without setting a weight for each feature amount, a general weighted Euclidean distance for each sample is obtained. Figure 7 shows the Euclidean distance
The distribution was as shown in FIG. In FIG. 7A, an undeterminable area exists between the genuine note area and the fake note area.

On the other hand, when the weights are set for the respective feature values as described above and the special weighted Euclidean distances for a plurality of genuine bill samples and a plurality of fake bill samples are measured, The special weighted Euclidean distance for the sample was distributed as shown in FIG. 7 (b). FIG.
In (b), a margin area where no data exists exists between the genuine note area and the fake note area, and the authenticity determination accuracy is increased.

It should be noted that the condition that the size of the margin area shown in FIG. 7B becomes equal to or larger than a predetermined value may be used as the end condition in step 17 in FIG.

[0053]

According to the present invention, the authenticity of a sheet can be comprehensively determined using a plurality of feature values. Further, according to the present invention, when the authenticity of a paper sheet is comprehensively determined using a plurality of feature amounts, a suitable weight can be easily set for each feature amount.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a true / false judgment processing procedure.

FIG. 2 is a schematic diagram for explaining a feature amount.

FIG. 3 is a schematic diagram for explaining a specially weighted Euclidean distance.

FIG. 4 is a flowchart illustrating a setting procedure of a feature amount weight;

FIG. 5 is a schematic diagram showing an individual.

FIG. 6 is a graph showing a specially weighted Euclidean distance L of each genuine note sample and each counterfeit note sample obtained for a certain individual.

FIG. 7 shows a distribution of a general weighted Euclidean distance for each sample when a general weighted Euclidean distance is measured for a plurality of true banknote samples and a plurality of counterfeit banknotes without setting a weight for each feature amount. And a distribution of a special weighted Euclidean distance for each sample when a special weighted Euclidean distance is measured for a plurality of true banknote samples and a plurality of counterfeit banknotes by setting a weight for each feature amount. It is a graph.

Claims (4)

[Claims] 1. A method of capturing an image of a sheet to be inspected, calculating a plurality of feature amounts from the image, and determining whether the sheet to be inspected is true or false based on the plurality of obtained feature amounts. In the preprocessing in which the weight for each feature amount is obtained in advance by a genetic algorithm, and based on the feature amount obtained from the sheet to be inspected and the weight for each previously obtained feature amount, Calculating a special weighted Euclidean distance in consideration of the weight for, and based on the calculated special weighted Euclidean distance, determining whether the paper sheet to be inspected is true or false. A method for judging the authenticity of paper sheets characterized by the following. 2. A plurality of feature amounts obtained from a sheet to be inspected are defined as x _i (i = 1, 2,..., N), an average of a genuine note sample with respect to the feature amount x _i is m _xi , and a feature amount Assuming that the variance of a genuine note sample with respect to x _i is σ _xi ² , the number of feature quantities is n, and the weight corresponding to feature quantity x _i is w _i , the specially weighted Euclidean distance is based on the following equation 1. 2. The method according to claim 1, wherein the authenticity of the paper sheet is calculated. (Equation 1) 3. The pre-processing includes: a first step of obtaining a feature value x _i for each of a plurality of genuine bill samples prepared in advance and a plurality of fake bill samples; Feature x
the average m _xi genuine note for _i, second step of calculating the variance sigma _xi ² genuine note for the feature amount x _i, individuals consisting of the genes representing the weights corresponding to the characteristic amounts x ₁ ~x _n predetermined Generate number and create initial population
Step: For each individual, a specially weighted Euclidean distance L is calculated for each genuine note sample and each counterfeit note sample based on the above-described equation 1, and based on the calculation result, the following equation 2 is calculated. A fourth step for determining the fitness F to be performed; a fifth step for determining whether or not the termination condition is satisfied; and a sixth step for creating a new individual population in consideration of the fitness F if the termination condition is not satisfied. The method according to claim 2, further comprising a seventh step of repeating the fourth to sixth steps until it is determined in the fifth step that the termination condition is satisfied. . (Equation 2) 4. Each area of a paper sheet is represented by i (i = 1, 2,...).
and n), and the projection results of an image obtained from each region i b _i, and _j, the reference data corresponding to the region i obtained from genuine note sample a _i, a _j, each feature amount x _i is ,
The authenticity determination method for paper sheets according to any one of claims 1, 2 and 3, represented by the following equation (3). (Equation 3)