JP3423140B2 - Banknote identification method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a bill discriminating method, and more particularly to a bill discriminating method for suppressing the influence on the discriminating accuracy due to various stains on bills to be discriminated.

[0002]

2. Description of the Related Art As a technique prior to the present invention, Japanese Patent Laid-Open No. Sho 60
-215293 is available. In the publication, the banknote is divided into a plurality of zones, and the detection data for each zone is compared with reference data that is obtained in advance for each zone.
In the banknote identification method for identifying the banknote based on the comparison result in each zone, the reference data is set in plural corresponding to the front and back of the banknote, the orientation, and the positional deviation at the time of identification, and for one banknote. Then, the data of each zone is summed up and stored as reference pattern data with a ratio value to the total value, and a total value of the detection data is obtained, and a ratio value to this total value is calculated as detection pattern data, It is determined whether or not the detection pattern data is within the allowable value range of the reference pattern data, and the absolute value of the difference between the reference pattern data and the detection pattern data is calculated for each zone and totaled, There is disclosed a bill identifying method characterized by performing bill identification by determining whether the total value of the distance calculation is smaller than an allowable value.

[0003]

In the conventional method of absorbing the variation of bill identification due to stains, distortion, and other reasons by using the threshold value, the variation of identification is allowed to some extent. There is a problem that it is mistakenly recognized as a genuine note, which is a cause of deterioration in identification accuracy.

Therefore, the present invention aims to provide a method in which the characteristic amount of the bill used for the discrimination is completely different from the conventional method for discriminating the bill with high accuracy without being affected by dirt, distortion, and the like. And

[0005]

According to the present invention, instead of the conventional method of pattern-comparing the sensor input waveform of the identified bill with the reference waveform as it is, it is read from a plurality of locations on a plurality of genuine bills as a reference. Based on the first step of creating a reference waveform based on the sensor input waveform of the bill, and the sensor input waveform of each bill of the plurality of genuine bills and the reference waveform, A second step of extracting the regular fluctuation component, and a data group in which the distribution of the irregular fluctuation component of each bill is arranged in time series based on the irregular fluctuation component of each bill extracted in the second step. The third step of creating a fluctuation component estimation model consisting of, and based on the sensor input waveforms read from a plurality of locations on the bill to be identified and the reference waveform, the variation component for correction of the bill to be identified. Extract 4 and step, and correcting fluctuation component of the object to be identified banknotes extracted by said fourth step,
A fifth step of calculating an irregular variation component of the bill to be identified based on the variation component estimation model, and an irregular variation component of the bill to be identified calculated in the fifth step, The sixth step of correcting either the sensor input waveform or the reference waveform of the bill to be identified and the corrected sensor input waveform are compared with the original reference waveform or the sensor input waveform and the corrected reference waveform, It is a bill identifying method including a seventh step of determining the authenticity of the bill to be identified.

For this reason, fluctuation components such as stains and distortions are extracted from the sensor input waveforms of a plurality of genuine bills in advance, and an estimation model of the fluctuation components is created by learning with an autoregressive model, a multiple regression model, or a neural network model. To do.

As described above, since the input waveform itself or the reference waveform is corrected by using the fluctuation component of the sensor input waveform and the true or false of the bill to be discriminated is discriminated, the discrimination accuracy is hardly affected by the fluctuation such as dirt. The accuracy is high.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment in which a self-regressive (AR) regression model is used to create an estimation model in the bill identifying method of the present invention will be described in detail below with reference to the drawings. (First Embodiment) This method is roughly divided into a preprocessing and a bill insertion processing. FIG. 1 shows a conceptual diagram of an embodiment in which a sensor input waveform of an identified bill is corrected by a fluctuation component model.

The pre-processing is a processing for creating a fluctuation component estimation model from a sensor input waveform of a standard genuine bill by learning, and as shown in FIG. The real bill (new bill) of the bill is sensed by the image sensor as one of the sensors to obtain the sensor input waveform of the brightness and density of the image and letters on the bill, and the reference waveform (eg input average waveform) from each input waveform To get

Next, in step S2, a variation component such as dirt or distortion is extracted from the input waveform of the genuine bill for each genuine bill by using the reference waveform. Then, learning data is created using the data of the fluctuation component extracted in step S3.

In the learning here, for example, FIG.
It means that the data of the fluctuation component obtained as shown in (A) is regarded as a periodic time series signal and is applied to a well-known autoregressive model equation to obtain the coefficient of the equation.

The learning data in this case is comparable to the time-series waveform data of the fluctuation component when several banknotes are arranged and continuously input. The periodic time series signal of the fluctuation component thus created is learned by the method of autoregressive analysis in step S4, and as a result of the learning, an estimation model of the fluctuation component for one bill is created.

After performing the preprocessing in this way, the process shifts to bill insertion processing for making a genuine / counterfeit determination when a bill is actually inserted. As shown in FIG. 1 (B), in the bill insertion process, first, the image sensor input waveform from the identified bill inserted in step S11 is input.

Next, in step S12, a difference between the waveform signal of the bill to be identified and the reference waveform is calculated to extract a variation component such as a stain or a distortion component. In step S13, based on the data of the fluctuation component obtained in step S12, the fluctuation component of the identified banknote is estimated using the estimation model of the fluctuation component using an autoregressive model, and the estimated fluctuation component is calculated. Here, the estimation method will be described. Since the estimation model has been obtained by the pre-processing, the fluctuation component estimated by the formula of the autoregressive model and the fluctuation component of the input bill to be identified is calculated ( Step S13).

The waveform of the fluctuation component of the bill to be identified thus obtained is used to correct the image sensor input waveform of the bill to be identified. This correction method is performed by subtracting the estimated fluctuation component from the image sensor input waveform of the bill to be identified or by adding the estimated fluctuation component to the reference waveform to perform the correction.

Finally, in step S15, the matching degree of the corrected image sensor input waveform of the bill to be identified and the reference waveform is calculated by using a determination method such as cross-correlation or pattern matching. If it is obtained, it is judged as true,
Otherwise, it is determined to be false and the identification process ends. (Second Embodiment) In the second embodiment, steps S1 to S1.
Up to 3, the same processing as in the first embodiment is performed.

The difference is that the reference waveform is corrected using the estimated value of the fluctuation component calculated in step S16,
Finally, in step S17, the degree of coincidence between the image sensor input waveform of the identified bill and the corrected reference waveform is calculated using a determination method such as cross-correlation or pattern matching to obtain a certain degree of coincidence. If it is determined to be true, otherwise it is determined to be false, and the identification process ends.

In the pre-processing or post-processing, not only an auto-regressive model but also an estimation model using a multiple regression model or a neural network model can be used. For example, explaining the estimation method of the estimation model of the fluctuation component of the data when using the multiple regression model, the position of the target data as the factor data that is the target of the other variable, the variation of the data, the ink density, the paper Degradation degree and the like can be mentioned, and the formula of the multiple regression model of the fluctuation component is created based on these.

In this case, the position of the target data may be input from an encoder or the like when taking in a bill. The data variation may be obtained as the variance of the difference value between the input sensor signal and the reference waveform.

Further, since the density of the ink printed on the bill constitutes the factor data as the variation thereof,
It can be calculated. Further, the degree of deterioration of the paper can be calculated by using the transmittance of the white portion of the banknote (if the input data is obtained from the transmission sensor, the sensor signal can be used together).

Of course, it is also possible to similarly obtain an estimation model of dirt by using a neural network instead of the autoregressive model. Even if the multiple regression model and neural network model are used as described above, the dirt and distortion components as the fluctuation component are extracted from the input bill, and the estimation model of the dirt component as the fluctuation component is obtained from the learning result in the preprocessing. When the above is created, this estimation model is used for the bill insertion process and is utilized for the correction of the input waveform or the reference waveform of the identified bill.

[0022]

As described above, the present invention replaces the conventional method of comparing the inputted pattern of the banknote itself with the standard pattern and judging it as a counterfeit note when the width of the difference is a certain value or more. By extracting the fluctuation component such as the dirt component from the pattern of the input bill, creating an estimation model of the fluctuation component using this, and utilizing this estimation model to correct the input waveform or the reference waveform of the identified and identified bill, Since the authenticity of the bill is determined by comparing the corrected input waveform or the reference waveform, it is expected that the effect of accurate authenticity determination can be achieved without being affected by the fluctuation component of the identified bill.

[Brief description of drawings]

FIG. 1 is a schematic explanatory diagram of a first embodiment of a bill identifying method of the present invention.

FIG. 2 is a schematic explanatory diagram of a second embodiment of a bill identifying method of the present invention.

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-5-101250 (JP, A) JP-A-3-210692 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G07D 7/00-7/20 G06F 17/60-19/00

Claims (2)

(57) [Claims] 1. A sensor detects a desired area of paper failure.
The banknote based on the sensor input waveform obtained by detecting
In the banknote recognition method of determining the authenticity of, read from a plurality of locations on the bill of a plurality of genuine notes as a reference
Based on the first step of creating a waveform to be a reference, the plurality of the bill <br/> sensor input waveform genuine note and the reference waveform based on the bill sensor input waveform, respective bill
A second step of extracting a random variation component of each, said second
For the irregular fluctuation component of each banknote extracted by the step
Based on the time-series distribution of the irregular fluctuation component of each bill.
A third step of creating a fluctuation component estimation model consisting of data groups arranged from a plurality of locations on the object to be identified banknotes
Based on the read-out sensor input waveform and the reference waveform
And a fourth component for extracting a correction fluctuation component of the bill to be identified .
Step, and the identified person extracted in the fourth step
And correcting fluctuation component of another banknotes, on the basis of said variation component estimation model, the a fifth step of calculating a random variation component of the identified banknotes, calculated by the fifth step
The sixth step of correcting the sensor input waveform of the identified banknote using the irregular fluctuation component of the identified banknote to be identified, and comparing the corrected sensor input waveform with the reference waveform to identify the identified object. a seventh step of determining the authenticity of a bill, become more bill validation process. 2. A sensor detects a desired area of paper failure.
The banknote based on the sensor input waveform obtained by detecting
In the banknote recognition method of determining the authenticity of, read from a plurality of locations on the bill of a plurality of genuine notes as a reference
Based on the first step of creating a waveform to be a reference, the plurality of the bill <br/> sensor input waveform genuine note and the reference waveform based on the bill sensor input waveform, respective bill
A second step of extracting a random variation component of each, said second
For the irregular fluctuation component of each banknote extracted by the step
Based on the time-series distribution of the irregular fluctuation component of each bill.
A third step of creating a fluctuation component estimation model consisting of data groups arranged from a plurality of locations on the object to be identified banknotes
Based on the read-out sensor input waveform and the reference waveform
And a fourth component for extracting a correction fluctuation component of the bill to be identified .
Step, and the identified person extracted in the fourth step
And correcting fluctuation component of another banknotes, on the basis of said variation component estimation model, the a fifth step of calculating a random variation component of the identified banknotes, calculated by the fifth step
A sixth step of correcting the reference waveform using the irregular fluctuation component of the identified banknote to be identified, and comparing the corrected reference waveform with the sensor input waveform of the identified banknote ,
A seventh step of determining the authenticity of the identification banknotes, become more bill validation process.