JPH0973573A - Medium counterfeit discriminating system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately discriminate the counterfeit of a medium by removing the influence of a damaged part and calculating (normalizing) a correction quantity. SOLUTION: A normalizing part 2 compares the integrated value obtained by summing up the input picture element values of the medium 8 with reference data in a correction quantity calculation reference data storing part 5 to execute normalizing processing which removes an error such as dirt. A counterfeit discrimination part 3 compares an input picture element value after normalization and reference data in a counterfeit discrimination reference data storing part 6 to discriminate the paper money to be true when the number of picture elements meeting reference is not less than a fixed number. A damaged part discrimination part 4 specifies a candidate area based on the picture element number out of reference and discriminates the damaged part by referring to reference data in a damaged part discrimination reference data storing part 7. Then the normalizing part 2 executes normalization processing again by excluding a damaged part area.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a medium authenticity discriminating system for discriminating the authenticity of paper sheets such as banknotes and securities by reading their images.

[0002]

2. Description of the Related Art Conventionally, currency exchange machines for recognizing securities, gift certificates, banknotes, etc., automatic teller machines at financial institutions, automatic ticket vending machines at railways, etc., recognize banknotes and securities, etc. to make a genuine / counterfeit judgment. It has a built-in device. In such a recognition device, a pattern of a ticket, an optical reflection pattern or a transmission pattern of a figure is detected and converted into an electric signal. Note that such an optical pattern is read by an image reader or the like. Further, there is also one in which a magnetic pattern is read by a sensor or the like, converted into an electric signal, and the same processing is performed.

Noise components such as unevenness of print density and dirt and scratches attached during the circulation process are added to a medium such as a bill. Further, the input pattern converted into an electric signal by the detector is also subject to level fluctuations due to changes in the sensitivity of the detector over time. In order to eliminate the influence based on the reading condition of the detector as much as possible and to perform a stable authenticity determination, conventionally, normalization is performed to correct the sensitivity change of the detector with respect to the input pattern. . In this normalization processing, the input pixel values obtained by reading the medium in pixel units along a predetermined line on the medium are summed for all pixels, and the pixel values are set to respective pixel values so that the integrated value thus obtained becomes a predetermined value. There is a method of multiplying by a coefficient for correction. As a result, a dark image or a light image as a whole can be corrected to a standard image level and authenticity discrimination can be performed. Authenticity discrimination using such a normalization method is described in, for example, Japanese Patent Laid-Open No. 2-186493.

[0004]

The conventional medium authenticity discrimination system as described above has the following problems to be solved. When the correction is performed based on the integrated value obtained by summing the pixel values of the regular input pixels as described above, it is suitable for the correction when the medium is dirty as a whole or the density is too high. Some are damaged,
When oil or the like adheres and there is a heavy stain in a concentrated manner, there is a problem that the error contained in the correction amount increases and correct correction cannot be performed. Therefore, if an erroneous determination that a genuine bill is a false bill occurs and the number of media rejected from the device increases, the convenience of the user who manages the device decreases. On the other hand, if the discrimination criterion is relaxed, such an unfit note can be accommodated, but there is a risk of erroneous acceptance of a counterfeit note.

[0005]

The present invention adopts the following constitution in order to solve the above points. <Structure 1> The medium authenticity discrimination system of the present invention sums up the input pixel values obtained by reading the medium pixel by pixel for all pixels along a predetermined line on the medium, and obtains the integrated value thus obtained. A normalization unit that normalizes the input pixel value by comparing with the correction amount calculation reference data, compares the normalized input pixel value with the true / false determination reference data, and determines the number of pixels that satisfy this criterion and the reference. The number of pixels that do not satisfy is stored, the number of pixels that satisfies the criterion is compared with the genuineness / counterfeit determination reference data to determine whether the medium is true or false, and a pixel number that does not satisfy the criterion is referred to The pixels are connected to each other, these are regarded as a series of areas, and a loss area determination unit for determining whether or not the series area is a loss area with reference to loss area determination reference data is provided.
The normalization unit sums up the input pixel values excluding the areas determined to be loss areas, compares the integrated value thus obtained with the correction amount calculation reference data again, and normalizes the input pixel values.

<Explanation> The medium is read in pixel units by optical or magnetic reading means along the scanning lines and converted into electric signals. Here, the word pixel is also used when reading magnetically. The input pixel value is a value converted into an electric signal and converted into a numerical value. The normalization refers to removing an error component such as a change with time of the reading device and an overall stain of the medium from the input pixel value. If the number of pixels satisfying the criterion is a certain number or more, it is determined as a genuine note. The pixel number is identification information for distinguishing each pixel from another pixel, and need not be a number. The damaged portion of the medium has a significantly different pixel value from the other portions and is concentrated in one area, so that it is detected. If the loss part is excluded, the normalized result will be closer to the reference data, so that the damaged medium can be accurately discriminated. The term "damage" refers to the case where it is not broken or the case where dirt such as oil is concentrated on a part of the surface.

<Structure 2> The true / false determination unit compares the normalized input pixel value with the true / false determination reference data, and stores the number of pixels satisfying this criterion and the pixel number not satisfying the criterion. Then, the number of pixels satisfying the criterion is compared with the true / false discrimination reference data to discriminate the authenticity of the medium, and the loss portion discriminating unit refers to the pixel number which does not satisfy the criterion, and connects the neighboring pixels ,
These are regarded as a series of areas, the loss area determination reference data is referenced to determine whether the series of areas is a loss area, and the normalization unit sums the input pixel values excluding the areas determined to be loss areas. , The integrated value thus obtained is again compared with the correction amount calculation reference data to normalize the input pixel value, and the true / false determining unit, the loss determining unit, and the normalizing unit show the results of normalizing the respective operations. Repeat until convergence.

<Explanation> Only a part of the data of the area determined to be the loss portion is initially detected. However, as this operation is repeated, the difference in pixel value between the normal portion and the loss portion becomes clear, and the loss portion is lost. The whole is detected. That is, if there are many pixel values that do not satisfy the standard even after normalization by excluding the detected region, for pixels that do not satisfy the standard again after normalization,
Determine if it does not belong to the loss department. Then, the pixels in the area determined to be the loss portion are excluded again and normalization is performed. By repeating this process, it is possible to perform normalization with the influence of the loss portion being sufficiently excluded. Therefore, the normalized result gradually converges.

<Structure 3> The authenticity discriminating unit, the loss discriminating unit, and the normalizing unit repeat the respective operations until the normalized result converges, and the authenticity discriminating unit performs the operation a preset number of times. If the medium cannot be discriminated as a genuine bill even after repeating the above steps, the medium is discriminated as a counterfeit bill. <Explanation> In the case of counterfeit bills, it does not converge even if this process is repeated. Therefore, the threshold value of the number of repetitions is set to prevent infinite repetition and the processing is ended.

<Structure 4> When the rate of change in the number of pixels in the area determined to be a loss portion falls below a certain level, the true / false determination unit, loss portion determination unit, and normalization unit converge the normalized results. To judge. <Explanation> The region determined to be a loss part gradually appears as a whole by repeating the normalization. Therefore, the number of pixels in the area determined to be a loss portion approaches a constant value when the normalization is repeated, and thus the rate of change converges. This is detected at an appropriate timing.

<Structure 5> The loss portion determination unit refers to pixel numbers that do not satisfy the criterion, connects neighboring pixels, regards these as a series of regions, and the number of pixels forming this region is constant. When the value is equal to or more than the reference value of, the loss portion determination reference data is set so as to determine the area as a loss portion. <Explanation> If a series of regions that do not satisfy the authenticity determination criterion is wider than a certain reference value, it can be regarded as a loss portion.

<Structure 6> The loss area determination unit sets a fixed upper limit reference value and a fixed lower limit reference value, and when the number of pixels forming a series of regions is between the upper limit reference value and the lower limit reference value, The area is discriminated as a loss portion, and if it exceeds the upper limit reference value, it is discriminated as a fake bill. <Explanation> When the area is wide enough to greatly exceed a certain reference value, most of the pixel values are considered to be significantly different from genuine bills. At this time, false bills are discriminated to prevent erroneous discrimination.
If the set of areas that do not meet the criteria for authenticity is extremely wide, the bank will not be able to exchange the unfit note. Therefore, it is possible to prevent such a medium from being discriminated as a genuine note and prevent erroneous recognition.

<Structure 7> The loss area determination unit sets a fixed upper limit reference value and a fixed lower limit reference value, and averages when the number of pixels forming a series of regions is between the upper limit reference value and the lower limit reference value. If the normalized pixel value is within a certain range, that portion is determined as a normal portion instead of a loss portion. <Description> Based on the correlation between the number of pixels forming a series of areas and the average normalized pixel value, a normal portion and a loss portion are separated to enable more accurate discrimination.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to specific examples. <Specific Example 1> FIG. 1 is a functional block diagram showing a specific example of the system of the present invention. Before explaining this system,
First, the cause of the problem in the conventional medium authenticity discrimination method is analyzed. It is now assumed that the pixel value of each pixel (i) in the average pattern is z _i , the additional noise such as stains, scratches, and print misalignment is n _i , and the temporal change factor of the detector sensitivity is a (t). However, the additive noise (n _i ) takes a different value depending on the medium, and the detector sensitivity change factor a (t) takes a value that changes with time. The input pattern (x _i ) is a value obtained by adding additional noise (n _i ) to the average pattern (z _i ) and converting it to an electrical level, and therefore can be expressed by the following equation. x _i = a (t) · (z _i + n _i ) ... (1)

When the number of pixels is N and the integral value of all pixels, which is the correction amount for normalization, is calculated using the above equation (1), the following equation is obtained. Σ _{i = 1} ^N x _i = a (t) · (Σ _{i = 1} ^N z _i + Σ _{i = 1} ^N n _i ) = a (t) · (S _Z + S _n ) ... (2) Σ z _i is all pixels The result of adding the pixel values of is expressed as an integrated value Sz. Further, Σn _i is a sum of additional noise such as stains for all pixels and is represented as an integral value Sn. Σx _i is the integrated value of the input pattern for all pixels. The normalization is to correct the input pattern (xi) of the equation (1) using the correction amount of the equation (2).
When the normalized input pattern of each pixel is yi, the following equation is obtained. yi = a (t) · (Z _i + n _i ) / a (t) · (S _z + S _n ) = (z _i + n _i ) / (S _z + S _n ) ... (3) This expression (3) is detected. Factor of the instrument sensitivity over time (a (t))
As can be seen from the expression that does not include, the normalization can eliminate the influence of the variation of the light and shade over the entire surface of the medium such as the temporal change of the detector sensitivity.

When n _{i in the} equation (3) is 0, that is, when there is no additional noise, the normalized input pattern of each pixel is expressed by the following equation. Y _i = z _i / S _z (4) However, the reason why the medium that has been slightly soiled or damaged over a relatively wide range can be satisfactorily corrected is that the noise component (n _i ) has both positive and negative gradations. This is because the noise is a very small noise, and the integrated value (S _n ) thereof cancels each other out and becomes very small compared to the integrated value (S _z ) of the average pattern included in the correction amount and can be ignored. . In the case of locally dark or light shading, if there is a high-level noise biased to one side, the integrated value (S _n ) of the noise component becomes so large that it cannot be ignored. Therefore, when the normalization as shown in the equation (3) is performed, the pixel value of the normal portion is deviated due to the correction.

FIG. 2 is a graph showing the relationship between the pattern of input pixel values and the pixel value of a medium having such a loss portion. The horizontal axis of this figure shows the pixel number for identifying each pixel, and the vertical axis shows the pixel value indicating the shade of each pixel. As shown in this figure, if there is a loss part that has a significant difference in light and shade, if this is normalized, it will actually be different from the average pattern as shown by the dashed line in the ideal state. The input pattern of is shifted downward. That is, when there is a high density loss part in a part, the input pattern of the other part, which shows a value that is originally close to the average pattern when normalized, is pushed below the position of the average pattern. If such a normalization is performed and then a true / false determination is performed, the following result is obtained.

FIG. 3 shows a pattern of input pixel values including an upper limit value and a lower limit value in the case of authenticity determination. The vertical axis and the horizontal axis have the same configuration as that in FIG. In this way, the upper limit value and the lower limit value are set in a fixed range centered on the average pattern, and compared with the input pattern, and the number of pixels not included in the range of the upper limit value and the lower limit value is counted. If this number is equal to or greater than a certain reference value, it is determined as a fake bill. In this example,
Since the input pattern is close to the lower limit value, a signal of a level that is originally included between the upper limit value and the lower limit value is also lower than the lower limit value and is extracted as a pixel value that is equal to or lower than the reference value. As a result, the result of the genuine / counterfeit determination is determined to be a fake bill. For example, it is assumed that the upper limit value and the lower limit value are set so that the number of pixels accommodated between the upper limit value and the lower limit value is 95% of the total number of pixels as the authenticity determination criterion. In this case, if the number of pixels that exceeds the upper limit value or the lower limit value is less than 5% of the total number of pixels, it is determined as a genuine note. It is assumed that the size of the loss portion of the medium showing such an input pattern is 4% of the whole. In this case, since the size of the loss portion is within the allowable range, the medium has to be judged as a genuine note. However, when the inspection is actually performed, the input pattern deviates from the average pattern by the above-described normalization, so that the number of pixels having a pixel value below the lower limit increases. Since a part of the loss part is less than or equal to the upper limit value, the pixels of the loss part exceeding this upper limit value are 3
If the percentage of pixels that are less than the percentage and the loss portion is 3% of the whole, then 6% of the pixels vertically combined are detected as outside the reference value. Therefore, the medium is determined to be a counterfeit note even though it is originally a genuine note.

In the present invention, in such a case, focusing on the fact that high level signals are concentrated in the loss portion, normalization is performed after excluding the concentrated abnormally high level signals. , To prevent misjudgment.

<Block Configuration> Returning to FIG. 1, the functional blocks of the system of the present invention will be introduced and the specific configuration thereof will be described. In FIG. 1, this system includes an image reader 1, a normalization unit 2, a true / false determination unit 3, a loss unit determination unit 4, a correction amount calculation reference data storage unit 5, a true / false determination reference data storage unit 6, and a loss unit determination. The reference data storage unit 7 and the like are included. In addition, these systems are actually incorporated in an automatic transaction device. The image reader 1 is a device for reading an optical reflection pattern or a transmission pattern of a pattern or a figure of a medium which is a target of inspection authenticity determination, a magnetic pattern or the like, and converting it into an electric signal. Therefore, it is composed of an image line sensor, a magnetic sensor and the like.

As shown in FIG. 2B, when reading the medium 8 such as bills, for example, the medium 8 is scanned in parallel at appropriate intervals and the data for each pixel is read. It depends on how you get it. For example, as shown in this figure, the medium 8 is provided with a loss portion 10 due to a high level of dirt having a relatively large area in a part thereof.

The normalization unit 2 calculates and normalizes the correction amount using the input pixel value and the reference data stored in the correction amount calculation reference data storage unit 5. In the above description, the integral value (S _z + S _x ) of the pixel values is described as the “correction amount”, but the “correction amount” used in the following description is a value obtained by using the integral value of the pixel values, The coefficient c (c = S / (S _z + S _x ): S is an arbitrary constant) by which the pixel value is multiplied in the normalization process. Further, this correction amount may be calculated from the pixel value obtained by reading the entire surface of the medium, or may be referred to from the pixel value excluding the area of the loss portion as described later.

The true / false determination unit 3 is a unit for determining the level of the pixel value of each pixel, determining the true / false of the medium, and determining the convergence end of renormalization. The authenticity discrimination unit 3 discriminates the authenticity of the medium according to the reference data stored in the authenticity discrimination reference data storage unit 6. The loss determination unit 4 determines the level of the pixel value by the true / false determination unit 3, and when a region considered to be a loss is detected, the loss determination unit 4 determines whether the pixel satisfies a series of regions. This is a part for performing the labeling for extracting the pixel number of, and the extraction of the loss area by the fitness judgment.
The fitness judgment is a process of determining the range of pixel values to be excluded for the correction amount calculation, and the authenticity of the medium is not determined by the fitness judgment itself.

The correction amount calculation reference data storage unit 5 stores, as reference data, an average pattern serving as a reference without additional noise. The true / false determination reference data storage unit 6 stores an upper limit reference value or a lower limit reference value which is a level determination reference for each pixel, a threshold value of the number of pixels for true / false determination, and renormalization used for determination of convergence end in normalization. This is a part for storing the limit value of the number of times. The loss portion determination reference data storage unit 7 is a portion that stores pixel number upper limit data, pixel number lower limit data, normalized density threshold value data, and the like used for correctness determination.

<Operation> FIG. 4 is a flow chart for explaining the general operation of the system of the present invention. In addition, in order to clarify the processing subject and data in each step of this figure, a simple explanatory diagram is shown in the following FIGS. 5 and 6. FIG.
FIG. 5 is an explanatory diagram of an operation subject of the process of FIG. 4. That is, the normalization unit 2 executes, for example, the processes of steps S1, S2, and S8, and the authenticity determination unit 3 executes the processes of steps S3, S4, and S5. In addition, the loss portion determination unit 4 performs steps S6 and S7.
Execute the processing of

FIG. 6 is an explanatory diagram of data used in the processing of FIG. In step S1, the correction amount calculation reference value is used as reference data, and in step S3, the upper limit value,
The lower limit value, the pixel number reference value and the like are used as reference data in step S4. First, in step S1 of FIG. 4, a correction amount is calculated. The correction amount is calculated using each pixel value of the input pattern read by the image reader 1 shown in FIG. 1 and the reference value stored in the correction amount calculation reference data storage unit 5 in advance. For example, the correction amount is obtained as follows using the integral value of the input pattern and a constant stored in advance. Correction amount = constant / integrated value of input pattern (5)

Next, in step S2, normalization processing is performed. That is, step S1 and step S8 described later
Each pixel value of the input pattern is normalized using the correction amount calculated in. In this case, for example, all pixels are multiplied by the correction value shown in the equation (5) or the equation (6) obtained in step S8 described later. In step S3, the upper and lower limit values are compared. Each pixel value of the input pattern normalized in step S2 is compared with the prestored reference data to determine whether it is true or false. For example, each pixel value of the input data is compared with the previously stored upper limit value and lower limit value data, and it is determined whether the pixel value is within the allowable range as described above with reference to FIG. The number of pixels whose pixel value is within the permissible range (lower limit value ≤ pixel value ≤ upper limit value) is first calculated, and at that time, the pixel number having the pixel value outside the permissible range, that is, the pixel value whose pixel value is smaller than the lower limit value The number and the pixel number whose pixel value is larger than the upper limit value are stored separately.

In the next step S4, authenticity discrimination is performed. The number of pixels within the allowable range counted in step S3 is compared with the threshold value of the true / false discrimination reference data storage unit 6 stored in advance, and if the number of pixels is equal to or larger than the required value, it is determined to be a genuine note, and step S9
Proceed to. If not, the possibility of fake bills or defective normalization operation is suspected, and the process proceeds to step S5 for renormalization processing. In step S5, a convergence determination is made.
The renormalization is repeated on the assumption that the medium is genuine and the pixel value of the loss portion is excluded, and the normalized value falls within a certain range. When the loss portion is excluded, the normalized value approaches the average pattern, and it can be finally determined that it is a genuine note. However, in the case of a counterfeit note, if such a determination is made, the processing is repeated indefinitely. To avoid this, the number of repetitions is limited. That is, when the number of repetitions exceeds the limit, the repeating process is stopped, the process proceeds to step S10, and the target medium is determined to be a fake note. If the limit is not exceeded, the process proceeds to step S6 to continue renormalization.

In step S6, labeling processing is performed.
This labeling is performed in order to determine whether or not these pixels are concentrated and fixed in a certain portion when the pixel value exceeds the upper limit value or falls below the lower limit value in step S3. That is, the pixel number of the pixel that is out of the reference is obtained, and when the pixel adjacent to it is out of the reference in the same way, the same label is given, it is regarded as included in the same area, and the lump of pixels is extracted. To do. Pixels that exceed the upper limit and pixels that fall below the lower limit stored in step S3 are individually subjected to labeling processing, and adjacent pixels are connected to each other to form a candidate region for a loss part.

In step S7, lossy part discrimination processing is performed.
The loss portion determination of each candidate area connected in step S6 is performed here. The loss determination is performed by comparing the area of each region, that is, the number of pixels, the average pixel value, and the like with a reference value. When the pixel values of the pixels included in a certain area as a whole exceed a certain range or are at a certain level or less, this is determined as a loss portion. The area determined to be a damaged portion is stored for the next step S8. In step S8,
The correction amount is recalculated. The loss area determined in step S7 is excluded from the pattern normalized in step S2, and the integral value of pixel values is obtained. That is, here, the normalization processing that eliminates the influence of the loss portion is performed. Similarly, the integrated value of the pixel values excluding the corresponding region is obtained from the average pattern stored in advance. When comparing with the average pattern, since it is a comparison of integral values, the same comparison cannot be performed if a loss portion is included. Therefore, the average pattern is also adjusted. The correction amount is calculated by the following formula. Correction amount = integral value of average pattern / integral value of input pattern (6) After calculating the correction amount, the process proceeds to step S2, and the above-described normalization and true / false determination are repeated again. In this way, it is determined that the bill is a genuine bill in step S9, or the bill is a counterfeit bill in step S10, and the determination process ends.

In the present invention, as described above, the pixels adjacent to each other are connected by the labeling in step S6 of FIG. 4 to form a loss candidate area, and the loss determination is performed in step S7.
Since such a loss portion has a relatively large area, and the pixel value tends to deviate significantly from the upper limit value or the lower limit value, a normal portion is erroneously detected by properly selecting the correctness determination criterion. It can be sufficiently distinguished from the pixel value. As shown in FIG. 3, as a result of the loss portion determination, after the loss portion protruding from the upper limit value is specified, the loss portion is excluded and the correction calculation is performed to determine the authenticity. In this case, the authenticity determination is performed by excluding the loss portion once, and when the result is determined to be a genuine note,
This may end the processing. If not all the parts that are considered to be loss parts are excluded once, the predetermined processing is repeated as described above.

FIG. 7 shows a pattern of input pixel values after renormalization. The format of this figure is exactly the same as that described using FIG. 2 and FIG. As shown in this figure, when the correction amount is calculated by using the portion excluding the loss portion and the normalization processing is performed, it can be seen that the deviation between the input pattern and the average pattern is sufficiently reduced. .

FIG. 8 shows a state in which the upper limit value and the lower limit value are set and the authenticity is determined after the normalization as shown in FIG. As shown in this figure, since the input pattern is sufficiently close to the average pattern as compared with the example described in FIG. 3, the number of pixels below the lower limit value is considerably reduced. However, when the loss portion is relatively large, the portion determined as the loss portion becomes wider as the input pattern approaches the average pattern in this way. The loss part excluded before the first normalization is narrower than the loss part appearing as shown in FIG. Therefore, the input pattern and the average pattern are still deviated due to the influence of the loss portion that cannot be excluded. Here, when it is determined that the bill is a counterfeit note, the same process of determining the loss portion and excluding the region is repeated again.

In FIG. 9, after the judgment as shown in FIG.
The results of further normalization are shown. As can be seen from this result, the average pattern and the input pattern are close to each other because the loss part is largely excluded.

FIG. 10 shows a pattern of input pixel values indicating the upper limit value and the lower limit value in the authenticity determination in that case. As shown in this figure, since most of the loss parts are excluded, the pixel value protruding from between the upper limit value and the lower limit value is sufficiently reduced, so that the medium has a relatively large loss part. It becomes possible to judge that it is a genuine note. It should be noted that in order that the processing for performing the above-described normalization and determination of true / false determination is not repeated indefinitely, for example, after normalization,
The number of pixels determined to be a loss portion before that is compared with the number of pixels determined to be a loss portion next, and it is compared whether or not the number is increased. If the increase is equal to or more than a certain value, the normalization and the authenticity determination process are repeated, and if it does not increase or decreases, it is determined to have converged. This can prevent unnecessary repetition of processing. Such a determination can also be made by a method of comparing the increase rate of the number of pixels forming the loss portion with a fixed reference value.

<Effect of Concrete Example 1> According to the above example, when the input pixel value of the medium is normalized, the correction amount without the influence of the loss portion can be calculated, so that the medium which is partially heavily contaminated. It is possible to perform a stable authenticity determination even for a medium that is partially torn or damaged. As a result, it is possible to sufficiently reduce misjudgment in which a genuine note is determined to be a counterfeit note and vice versa. Furthermore, by stabilizing the normalized result and bringing it closer to the average pattern, the standard for authenticity determination can be set more strict than before.
Therefore, there is an effect that the risk of erroneously discriminating a counterfeit note as a genuine note can be reduced. This makes it possible to provide a highly reliable service that is convenient for general users.

<Second Specific Example> The system configuration of the second specific example is exactly the same as that shown in FIG. In the specific example 2, the condition for determining whether the medium is correct, that is, the reference data stored in the loss portion determination reference data storage unit 7 shown in FIG. 1 is different.
FIG. 11 is an explanatory diagram of a fitness determination condition. In this figure, the horizontal axis represents the area area corresponding to the number of pixels exceeding the range of the upper limit value and the lower limit value. This graph is drawn so that it gets larger toward the right. The vertical axis represents the average density after normalization. In this specific example, the division as shown in this figure is used to determine whether a loss portion, a correct portion other than that, a counterfeit note, or the like.

FIG. 12 is an explanatory diagram for explaining whether the correctness determination condition is satisfied. The contents shown in FIG. 11 indicate the judgment as shown in FIG. Here, for example, when the area area is larger than M _max, it is determined as a fake bill. That is, there is a case where the damaged portion covers a remarkably large area, a case where the bill is originally a fake bill, and a case where the bill is actually damaged so that it cannot be exchanged even if the bill is brought into the bank. In the latter case, for example, a banknote or the like that is half or more torn is applicable. In such a case, since it is handled as a fake bill, the maximum value of this threshold is set. Moreover, the range area of the region is given, that is, when a range of M _{mi n} or M _max determines the loss portion of this. If the absolute value of the normalized density is P _th or more shown in the figure, it is a loss part,
In the following cases, since it corresponds to the level of the genuine bill, it is judged as the genuine part. If the area of the region is M _min or less,
Since it does not correspond to any of the above, it is judged to be the original part. By providing such a condition, it becomes possible to make a stricter true / false decision. The average normalized density means normalization of the density (x _i ) of each pixel in the area using a statistical average value (μ _i ) and standard deviation (σ _i ) of each pixel {(x _i − μ _i ) / σ _i }
It is the average value within the area of the value. This means that the greater the absolute value, the greater the degree of contamination. However, the normalization performed here is normalization of each pixel using an average value or a standard deviation, and the feature amount depending on the medium, that is, the pixel value of the pixel value as performed in step S2 of FIG. This is different from the normalization in which the integrated value or the like is used as the correction value.

<Operation> The operation of the second specific example is generally performed according to the flowchart shown in FIG. However, the reference data and the like used here are different in each step. Therefore, the contents are shown below. FIG. 13 is an explanatory diagram of used data in the process of FIG. 4 in the second specific example. As shown in this figure, in the second specific example, the correction amount calculation in step S1 in FIG. 4, the upper and lower limit comparison in step S3, and the step S4
In the authenticity determination performed in step S5, the renormalization convergence determination performed in step S5, the labeling performed in step S6, the loss portion determination performed in step S7, and the correction amount calculation performed in step S8, the reference data shown on the right side, respectively. To use.

14 to 21 show further specific operation flowcharts in the second specific example. FIG.
6 is a flowchart illustrating a process of calculating a correction amount. The following data are used as parameters here. i: pixel number counter p: pixel value cumulative counter z: renormalization number counter c: correction amount S: correction amount calculation reference value

First, in step S11, the pixel number counter i, the pixel value accumulation counter p, and the renormalization number counter z are initialized. Next, in step S12,
The pixel number counter i is incremented. Further, in step S13, the pixel value x _i of the input pattern with the pixel number _i is read. Then, in step S14,
The pixel value x _i is added to the pixel value cumulative counter p. In step S15, if the pixel number i is smaller than the final number N, the process returns to step S12, and if the pixel number i is the final number N, the process proceeds to step S16. With these, cumulative addition of all pixel values is performed. In step S16,
Based on the result thus obtained, the correction amount (c = S / p) is calculated using the pixel value cumulative value p and the correction amount reference value S.

The above processing corresponds to step S1 in FIG. Now, the process corresponding to step S2 of FIG. 4 will be described. FIG. 15 is a specific flowchart of this normalization process. Here, the following are used as parameters. i: Pixel number counter c: Correction amount

In step S17, the pixel number counter i is initialized. In step S18, the pixel number counter i is incremented. In step S19, the pixel value (x _i ) of the input pattern of the pixel number counter i is read.
In step S20, step S1 when calculating the correction amount
6, or the pixel value (x _i ) is normalized using the correction amount (c) obtained in step S83 described later. In step S21, if the pixel number i is smaller than the final number N, the process returns to step S18, and if the pixel number i is the final number N, the process proceeds to step S22. This completes the normalization process for each pixel.

FIG. 16 shows a flowchart for the upper and lower limit comparison and the authenticity determination. This is step S shown in FIG.
This is a process corresponding to S3 and S4. The following parameters are used here. i: Pixel number counter n: Positive determination pixel counter z: Renormalization number counter wa _i : Upper limit error determination flag (1/0) wb _i : Lower limit error determination flag (1/0) M: Pixel number reference value (true False discrimination reference value) A _i : upper limit reference value B _i : lower limit reference value

In step S22, the pixel number counter i and the positive determination pixel counter n are initialized. Next, in step S23, the pixel number counter i is incremented. Further, in step S24, the erroneous determination flags (upper limit wa _i , lower limit wb _i ) are initialized. In step S25, the pixel value x _i of the input pattern with the pixel number _i is read. In step S26, the lower limit reference value (B _i ) is read. In step S27, the pixel value x _i
Is compared with the lower limit reference value (B _i ), and if the pixel value x _i is smaller than the lower limit reference value (B _i ), the process proceeds to step S30, and if not, the process proceeds to step S28. Step S28
Then, the upper limit reference value A _i is read. In step S29, the pixel value x _i as compared to the upper reference value A _i, the pixel value x _i
Is larger than the upper limit reference value A _i , the process proceeds to step S32, and if not, the process proceeds to step S31. In step S30, as a result of the determination in step S27, it is considered that the noise is added to the allowable value or more, and the lower limit error determination flag of the pixel is set (wb _i = 1). Step S3
At 1, the pixel noise is considered to be within the allowable range, and the positive determination pixel counter n is incremented. Step S3
In No. 2, it is considered that noise is added to the allowable value or more, and the upper limit erroneous determination flag of the pixel is set (wa _i = 1).
In step S33, if the pixel number i is smaller than the final number N, there are unprocessed pixels, and the process returns to step S23. If the pixel number i is the final number (N), the process proceeds to step S34.

FIG. 17 shows the processing after step S34. In step S34, the positive determination pixel number n is compared with the pixel number reference value M. If the positive determination pixel number n is equal to or larger than the pixel number reference value M, it is considered that the inspection medium is a genuine note, and the process proceeds to step S35. move on. If the positive determination pixel number n is smaller than the pixel number reference value M, the possibility of normalization failure is suspected due to the influence of the loss portion, and the process proceeds to the renormalization process after step S36. In step S35, as a result of the authenticity determination, the inspection medium is determined to be a genuine note. In step S36, since renormalization is performed, the renormalization count z is incremented.

From here, the renormalization convergence determination of step S5 shown in FIG. 4 is performed. When the number of times of renormalization counted in step S36 reaches the limit value stored in advance, renormalization is stopped and it is determined that the bill is a counterfeit note. If the limit value has not been reached, renormalization is continued. This avoids the risk of the inspection of abnormal bills such as counterfeit bills entering an infinite loop. After that, labeling processing is performed. This corresponds to the process of step S6 of FIG. The labeling process and the loss part determination process performed next are independently performed for the upper limit erroneous determination flag (wa _i ) and the lower limit erroneous determination flag (wb _i ). Therefore, in the following description, in order to avoid duplication, upper and lower erroneous determination flags (wa _i , wb)
_i ) will be collectively described as w _i . The labeling method follows the contents of the following documents, for example. "Practical image processing to learn in C language (p99-p101, p105), Yagi et al., Ohmsha". The following parameters are used in these processes. i, j: Pixel number counter m: Label number counter k: Neighbor list number counter n: Pixel number counter l _max : Maximum label number w _i : Misjudgment flag / label number list L: Label number initial value N _H : Neighbor pixel Number list number H _k : Neighborhood pixel number list

In step S37, the pixel number counter i and the label number counter m are initialized. Step S
At 38, the pixel number counter i is incremented.
In step S39, the erroneous determination flag w _i is checked for erroneous determination region search, and if the pixel is an erroneous pixel (w _i = 1), the process proceeds to step S40 and labeling is started. _i = 0) The process proceeds to step S55 to inspect the next pixel. In step S40, the erroneous determination flag (label number list w _i ) is overwritten with the label number m. In step S41, the pixel number counter j and the pixel number counter n
Is initialized. In step S42, the pixel number counter j is incremented. At step S43, examines the label number list w _j of the pixel number j, if the label number is a m, the process proceeds to step S44 shown in the following figure 18, is inspected and labeled eight neighboring pixels. If the label number is not m, the flow advances to step S54 to inspect the next pixel.

FIG. 18 shows the processing after step S44. In step S44, the neighbor list number counter k is initialized. In step S45, the neighbor list number counter k is incremented. In step S46, the neighboring pixel number list (H _k ) of the neighboring list number k is read. In step S47, the neighboring pixel number list (H _k ) is the pixel number difference data of eight neighboring pixels. Therefore, the neighboring pixel number is obtained by adding it to the inspection target pixel number j. At step S48, the examined erroneous determination flag of neighboring pixel number j (w _i), if found to be erroneous pixel also neighboring pixels (w _i = 1) proceeds in the vicinity of the pixel connection process in step S49, the otherwise In this case, the process proceeds to step S51 to check the next neighboring pixel. In step S50, the pixel number with the label number m is added to the pixel number counter n. In step S52, the neighborhood list number k is the final number (N _H = 8 in the case of 8 neighborhood pixels).
If it is smaller, the process returns to step S45, and if the neighbor list number k is the final number (N _H = 8), step S45.
Go to 51. In step S52, the pixel number counter is checked, and if there is even one number in which the level number is entered (when n ≠ 0), the process proceeds to step S54, and the search for the same label number and the neighboring pixel connection are repeated. If there is no pixel in which the label number is entered (when n = 0), the process proceeds to step S53 to continue the next area search.
In step S53, the label number m is incremented. Although step S54 is shown in FIG. 17, when the pixel number j is smaller than the final number N, the process returns to step S42, and when the pixel number j is the final number N, the step S54.
Return to 41. In step S55, the pixel number i is the final number N
If it is smaller, the process returns to step S38 and the pixel number i
Is the final number N, the process proceeds to step S56.

FIG. 19 shows the processing after step S56. After that, the loss portion determination process of step S7 of FIG. 4 is performed. In step S56, the maximum label number l
_{The max} is stored and the labeling process ends. The following parameters are used in the following processing. i: Pixel number counter m: Label number counter n: Pixel number counter p: Normalized density accumulation counter l _max : Maximum label number w _i : Label number list h _m : Loss part determination flag (1/0) M _max : Pixel Number upper limit reference value M _min : Pixel number lower limit reference value P _th : Average normalized density reference value

In step S57, the label number counter m is initialized. In step S58, the pixel number counter i, the pixel number counter n, and the normalized density cumulative counter p are initialized. Further, in step S59, the pixel number counter i is incremented. In step S60, erroneous determination region Check the label number list for search (w _i), the pixel label number (w _i) extracts the feature quantity proceeds to step S61 if m. If the pixel label number (w _i ) is not m, the process proceeds to step S64 to inspect the next pixel. In step S61, the pixel number counter n is incremented to count the number of pixels in the area with the label number m. In step S62, data (pixel value x _i , pixel average value μ _i , pixel standard deviation σ _i ) necessary for normalizing the normalized pixel density is read. In step S63, the normalized density is calculated by normalizing the pixel value x _i using the average (μ) and the standard deviation (σ), and is accumulated in the normalized density cumulative counter p. In step S64, the pixel number i is the final number N.
If it is smaller, there is an unprocessed pixel, so step S
It returns to 59 again and searches. If the pixel number i is the final number N, it is considered that the feature extraction of all the pixels in the area is finished, and the process proceeds to the damage determination processing of the area of the label number m after step S65.

FIG. 20 shows the processing after step S65. In step S65, the area (pixel number n) of the region is compared with the pixel number upper limit reference value (M _max ), and if the region area exceeds the pixel number upper limit reference value, it is considered that the contamination of the inspection medium exceeds the allowable range. It proceeds to step S66. If not, the process proceeds to step S67. In step S66, the inspection medium is regarded as a fake bill. In step S67, the area of the region (the number of pixels n) is compared with the pixel number lower limit reference value (M _min ),
If the area is small enough to fall below the pixel number lower limit reference value, the area is regarded as the normal portion and the process proceeds to step S69. If not, it is determined that there is still a possibility of loss, and the process proceeds to step S68. In step S68, the average (| p | / n) of the absolute values of the cumulative value p of the normalized densities is calculated and compared with the average normalized density reference value (P _th ). Absolute value of average normalized density value (| p
If | / n) is equal to or greater than the average normalized density reference value (P _th ), the stain level is high, and the region is regarded as a loss part. On the other hand, if not, it is regarded as the original part and the process proceeds to step S69. In step S69, when the area is regarded as the normal portion, the loss portion determination flag h _{m is set} to 0. In step S70, when the area is regarded as a loss portion, the loss portion determination flag h _{m is set} to 1. Step S
In 71, if the last label number l _{ma x)} is smaller than that determined by the label number i is step S56, step S for inspection region of the next label number because of the non-examination region
Go to 72. If the label number i is the final number (l _max ), the loss portion determination process is terminated and the process proceeds to step S73. Note that FIG. 21 shows steps after step S72.

In step S72, the level number counter m is incremented, and the process returns to step S58 again. After step S73 in FIG. 21, step S8 shown in FIG.
It corresponds to the correction amount calculation process excluding the damaged part. In the above labeling and loss determination, the upper and lower erroneous pixel processing is performed with an erroneous determination flag (label number list w _i ) and a loss determination flag (h
_m ) was used for the explanation. The data actually output in the loss portion extraction process (upper limit) includes the upper limit erroneous determination flag (label number list wa _i ) and the loss portion determination flag (h
a _ma ), and the data output in the loss portion extraction process (lower limit) are the lower limit erroneous determination flag (label number list wb _i ) and loss determination flag (hb _mb ). Both upper and lower limit data are used in the following description. The following parameters are used in the following processing. i: Pixel number counter ma: Upper limit label number counter mb: Lower limit label number counter p: Input pixel value cumulative counter q: Average pixel value cumulative counter c: Correction amount wa _i : Upper limit label number list wb _i : Lower limit label number list ha _m : upper limit loss determination flag (1/0) hb _m : lower limit loss determination flag (1/0) T _i : average pattern pixel value

In step S73, the pixel number counter i,
Input pixel value cumulative counter p, average pixel value cumulative counter q
Is initialized. In step S74, the pixel number counter i is incremented. In step S75, the label number (ma, mb) is read from the label number list (wa _i , wb _i ) of the pixel number i. In step S76,
Label number (ma, mb) read in step S75
Load the loss portion determination flag (ha _ma, hb _mb) respectively.
At step S77, the check the read loss portion determination flag in step _{S76 (ha ma, hb mb)} , if the test pixel are both Tadashibu _{(ha ma = 0 and hb mb} = 0) , the process proceeds to step S78 correction When the feature quantity for quantity calculation is extracted, and either one or both are loss parts ( _hama =
1 or hb _mb = 1), the process proceeds to step S82 to check the next pixel. In step S78, the pixel value x _i of the input pattern with the pixel number _i is read. In step S79, the pixel value x _i is cumulatively added to the input pixel value cumulative counter q. In step S80, the pixel value T _i of the average pattern of the pixel number _i is read. In step S81, the pixel value T _i is cumulatively added to the average pixel value cumulative counter p. Step S8
In 2, if the pixel number i is smaller than the final pixel number N, the process returns to step S74, and if the pixel number i is the final pixel number N, the process proceeds to step S83. Step S83
Then, the correction amount (c = q / p) is calculated using the input pixel cumulative value p and the average pixel value cumulative value q. And again in FIG.
The process returns to step S17, which is the process for normalization shown in FIG.

<Effect of Concrete Example 2> As described above, since the loss judgment is performed using the area area of the area extracted as the candidate of the loss area and the average normalized density of the area, the loss area and other areas are determined. The feature difference of 1 is likely to occur, and the loss portion can be accurately detected. Further, when the area area of the loss portion is equal to or larger than a certain value, the processing is considered to be a fake note and the processing is stopped. It can be reduced. In addition, when the number of times of renormalization is counted and it is repeated over a certain amount, the medium is determined to be a fake bill, so when inspecting a fake bill, there is a risk that the renormalization repetition process falls into an infinite loop. It can be avoided.

In the above specific example, the integrated value of the pixel values of the input pattern is used as the correction amount, but a method of calculating and normalizing the correction amount using both the average value of the pixel values and the standard deviation may be used. Can be adopted. Further, in the method of discriminating the loss portion, other than the method of using the area of the candidate region and the normalized density as the condition for the discrimination of the fitness, various other discrimination conditions can be combined in various ways. Further, as described in the above specific example, in order to set the limit on the number of times of renormalization, in addition to setting the upper limit of the number of times itself, the number of pixels protruding from the range between the upper limit value and the lower limit value increases or decreases by a certain amount or more. When it disappears, that is, when the rate of change becomes equal to or less than a certain value, it is possible to perform the convergence determination such as stopping the processing.

[Brief description of drawings]

FIG. 1 is a system functional block diagram of the present invention.

FIG. 2 is a pattern of input pixel values.

FIG. 3 is a pattern of input pixel values.

FIG. 4 is a system operation flowchart of the present invention.

5 is an explanatory diagram of an operation subject of the process of FIG. 4;

FIG. 6 is an explanatory diagram of used data in the process of FIG.

FIG. 7 is a pattern of input pixel values.

FIG. 8 is a pattern of input pixel values.

FIG. 9 is a pattern of input pixel values.

FIG. 10 is a pattern of input pixel values.

FIG. 11 is an explanatory diagram of a fitness determination condition.

FIG. 12 is an explanatory diagram of judgment of a condition for judging whether the condition is right or wrong;

FIG. 13 is an explanatory diagram of data used in the process of FIG. 4 in the second specific example.

FIG. 14 is an operation flowchart (part 1) of the second specific example.

FIG. 15 is an operation flowchart (part 2) of the second specific example.

FIG. 16 is an operation flowchart (part 3) of the second specific example.

FIG. 17 is an operation flowchart (part 4) of the second specific example.

FIG. 18 is an operation flowchart (part 5) of the second specific example.

FIG. 19 is an operation flowchart (Part 6) of the second specific example.

FIG. 20 is an operation flowchart (part 7) of the second specific example.

FIG. 21 is an operation flowchart (part 8) of the second specific example.

[Explanation of symbols]

 1 Image Reader 2 Normalization Section 3 Authenticity Discrimination Section 4 Loss Discrimination Section 5 Correction Amount Calculation Reference Data Storage Section 6 Authenticity Discrimination Reference Data Storage Section 7 Loss Section Discrimination Reference Data Storage Section 8 Medium 9 Reading Line 10 Loss Section

Claims (7)

[Claims] 1. An input pixel value obtained by reading the medium on a pixel-by-pixel basis for all pixels is summed along a predetermined line on the medium, and the integrated value thus obtained is compared with correction amount calculation reference data, A normalization unit for normalizing the input pixel value, comparing the input pixel value after normalization with true / false discrimination reference data, and storing the number of pixels satisfying this criterion and the pixel number not satisfying the criterion. , A true / false discrimination unit that compares the number of pixels satisfying the criterion with the true / false discrimination reference data to determine whether the medium is true or false, and refers to a pixel number that does not satisfy the criterion, connects neighboring pixels to each other, These are regarded as a series of areas, and a loss section determining unit that determines whether or not the series of areas is a loss section by referring to the loss section determination reference data is provided, and the normalization section determines the areas determined as loss areas. The excluded input pixel values are summed, and the integrated value thus obtained is again Compared to positive amount calculation reference data, authenticity discrimination system medium characterized by normalizing the input pixel value. 2. The true / false determination unit compares the normalized input pixel value with the true / false determination reference data, and stores the number of pixels satisfying the criterion and the pixel number not satisfying the criterion. The number of pixels satisfying the criterion is compared with the true / false determination reference data to determine whether the medium is true or false, and the loss portion determining unit refers to a pixel number that does not satisfy the criterion and connects neighboring pixels to each other. , Considers these as a series of areas, and refers to the loss area determination reference data to determine whether or not the series of areas is loss areas, and the normalizing section excludes the areas determined as loss areas from the input pixels. The values are summed, the integrated value thus obtained is again compared with the correction amount calculation reference data, and the input pixel value is normalized, and the true / false determination unit, the loss portion determination unit, and the normalization unit,
The authenticity discrimination system for a medium according to claim 1, wherein each operation is repeated until the normalized result converges. 3. The true / false determining unit, the loss determining unit, and the normalizing unit repeat respective operations until the normalized result converges, and the true / false determining unit presets the operation. 3. The medium authenticity discriminating system according to claim 2, wherein the medium is discriminated as a counterfeit bill when the medium cannot be discriminated as a genuine bill even after repeating the number of times. 4. When the rate of change in the number of pixels in the area determined to be a loss portion is equal to or lower than a certain value, the true / false determination unit, the loss portion determination unit, and the normalization unit converge the normalized results. The authenticity discrimination system for a medium according to claim 2, wherein it is determined that the determination is made. 5. The loss area determination unit refers to a pixel number that does not satisfy a criterion, connects neighboring pixels to each other, regards these as a series of areas, and determines the number of pixels forming this area to be constant. 6. The authenticity discrimination system for a medium according to claim 1, wherein the loss portion determination reference data is set so as to determine the area as a loss portion when the value is equal to or larger than the reference value. 6. The loss area determination unit sets a fixed upper limit reference value and a fixed lower limit reference value, and when the number of pixels forming a series of regions is between the upper limit reference value and the lower limit reference value, The authenticity discriminating system for a medium according to claim 5, wherein the area is discriminated as a loss portion and is discriminated as a counterfeit note when the upper limit reference value is exceeded. 7. The loss area determination unit sets a fixed upper limit reference value and a fixed lower limit reference value, and when the number of pixels forming a series of regions is between the upper limit reference value and the lower limit reference value, an average normal value is set. The authenticity discrimination system for a medium according to claim 6, wherein if the digitized pixel value is within a certain range, the portion is discriminated as a normal portion instead of a loss portion.