JPS6154580A - Pattern matching circuit - Google Patents

Abstract

PURPOSE:To increase the accuracy of a pattern matching circuit by preparing the mean value of plural data and the standard deviation value as the reference data and calculating the resemblance degree (marks obtained) according to the value obtained by normalizing the difference between the detection data and the mean value with the standard deviation value. CONSTITUTION:A dictionary memory 6 stores the marks Sx obtained to the input data at points A-D of each card respectively. These marks Sx are decided as follows. That is, plural sheets of reference cards are prepared and both the mean value (m) and the standard deviation value (sigma) are obtained for each point. Then nx=(x-m)/sigma is calculated to an input (x), and the error between the input (x) and the value (m) is normalized with the value (sigma). Then Sx=A[1-k (10nx-1)**j] is calculated, where A, j and k show constants and **j shows the j-square respectively. The value Sx is set at the mark obtained against the input (x). Thus the deterioration of the pattern matching factor can be suppressed even in case the reference data has variance.

Description

Detailed Description of the Invention (1) Technical Field of the Invention The present invention examines the similarity between detected pattern data detected by a sensor and input as a digital signal by an AD converter and reference data stored in advance. The present invention relates to a pattern matching device.

(2) Background of the technology In automatic banknote exchange machines and various vending machines, it is necessary to determine how many banknotes a banknote has been processed by two, for example, nine people. In that case.

A pattern matching circuit is used, which scans, for example, an input banknote with a diode sensor,
Detected light information as detected pattern data + M
By performing pattern matching with 'I' data, '7 matching' is performed.

At this time, it is required to perform pattern matching with almost no errors, especially for devices that handle paper 1711 or the like.

(3) Prior art and problems Conventional pattern matching circuits divide nine banknotes horizontally, for example, into MI [lil] areas, and measure the average brightness of each area (called a zone) using a sensor. Detect. The pattern matching circuit is stored in advance with reference data for the brightness of each zone for each type of banknote, such as 10,000 yen bills, 5,000 yen bills, and 1,000 yen bills. Regarding the standard data and the input detected data,
Compute the similarity for each zone of Mf[lil. In this case, the degree of similarity is expressed, for example, by a score that is inversely proportional to the absolute value of the difference between the reference data and the detected data. Then, the scores are added up for the M zones and used as the degree of similarity between the 10,000 yen bill and the detected data. Do the above calculations.

The same process is performed for 5,000 yen bills and 1,000 yen bills, and a method is used in which the type with the highest score (similarity) is output as the type of input banknote.

However, the brightness of banknotes varies by type and by zone within each type. The degree of this variation depends on the age of the banknote, its two colors, its brightness, and whether or not it forms a crease when folded.

Therefore, if the similarity for each species and each zone is simply calculated in inverse proportion to the absolute value of the difference between the reference data and the detected data, as in the above method, the brightness in a specific zone for each species If the variation in banknotes is large, even if the input banknote and its type match, there will be a large difference in the data in that zone (and the similarity of that zone will drop significantly, leading to the possibility of misclassifying five types. In other words, in the conventional pattern matching method, the standard data includes the degree of data dispersion for each type and for each zone.

(4) Purpose of the Invention The present invention solves the above problems by normalizing the degree of similarity by the standard deviation of reference data.

It is an object of the present invention to provide a pattern matching circuit that can suppress a decrease in pattern matching rate due to data variations.

(5) Structure of the Invention In order to achieve the above object, in a pattern matching circuit that performs pattern matching between detected data and dictionary data, the degree of similarity between the detected data and the dictionary data is calculated based on the variation in the dictionary data. It is an object of the present invention to provide a pattern matching circuit characterized in that it has a similarity setting means for setting the degree of similarity according to the following, and a determination means for determining the type of the detected data based on the output of the similarity degree setting means.

(6) Examples of the invention Hereinafter, one embodiment of the present invention will be described in detail.

FIG. 1 is an overall configuration diagram of a banknote pattern matching circuit according to the present invention. First, sensor 1° amplifier 2. ΔD
Converter 31 Representative value detection circuit 4. Dictionary data block selection circuit5. The dictionary data memories 6 are connected in series in this order. The dictionary data memory 6 consists of M zone memories from 21 to 2M, and each zone memory costs ¥10,000.
(equivalent to a 10,000 yen bill), ¥5,000 (equivalent to a 5,000 yen bill),
It is divided into 1,000 yen (equivalent to 1,000 yen bill) types, and each type is roughly A.

It is divided into four types: B, C, and D. That is,
Each zone memory stores a total of 12 types of dictionaries. and 6
U: serves as a memory for storing scores for each of N blocks in which the output levels 1 to 256 of one representative value DAT are grouped into four levels, for example. Next, the dictionary data memory 6 is connected to a denomination-specific register 7. In this case, the outputs of the M zone memories Zl-ZM of the dictionary data memory 6 are connected to the corresponding registers of the denomination-specific register 7 for each type. The output of the type register 7 is input to the judgment logic circuit 8, and the judgment result is outputted as C2S.

In the banknote pattern matching circuit configured as described above, the output of the sensor 1 is first amplified by the amplifier 2, and then converted into a digital signal by the ΔD converter 3.
It is input to the representative value detection circuit 4. In the representative value detection circuit 4, the average value of the sampling data in each zone is calculated for each M zone, and is output as the representative value of each zone. Next, the representative value for each zone is input to the dictionary data block selection circuit 5, and blocks L of 1 to N correspond to the output level of each representative value of the zone memory in the dictionary data memory 6 corresponding to each zone. Select one. In the denomination ff data memory 6, the scores stored in the memory corresponding to two selected blocks for each of the 12 denominations are added to the corresponding registers of the denomination classification register 7. For example, if the output level of the representative value of the first zone is 10, the third block of the zone memory Z1 in the dictionary/data memory 6 is selected. As a result, the type ¥1 of the zone memory Z1 in the dictionary data memory 6 is stored in ΣA of the register ¥10000 in the denomination level 7.
The score of oooo's A's 3-brosoku on the memory -5 is added. Similarly, register ¥10 in the denomination-specific register
-30 is added to ΣB of 000, and the same operation is performed thereafter up to D of register ¥1000. The above operations are performed for all representative values of M zones. As a result, 8, for example, register ¥1ooo in register 7 by denomination
ΣA of o contains the type ¥100 in the dictionary data register 6.
00 A-Evening scores will be tallied.

After addition over all zones as above.

The judgment logic circuit 8 compares the total values of each register in the denomination type register 7 and selects the type with the largest value.
10000. ¥5000. Select from ¥1000 and output as CR$. In addition, for each CF3, A,
There are four types: B, C, and D.

The way bills are input to sensor 1 is (from the left side when facing up), (from the right side when facing up), (from the left side when facing down)
, (,L!: from the right side of the direction), so dictionary data corresponding to each is required.

Next, a method for creating each dictionary data in the dictionary data memory 6 will be described. First, for each type, a plurality of reference patterns of that type are input using the sensor 1. Then, the average value m and standard deviation σ of these plurality of output levels are calculated for each zone. Next, output level x (x
=1.2. ..., 256') is input, the n value in units of σ is calculated as follows.

nX=(x-m)/σ (1) nX calculated in this way is calculated when the average value of the output level of the reference pattern in a certain zone of the species is m and the standard deviation is σ. When detected data of output level X is input at the time of , it is shown that the relative distance between the detected data and the reference pattern in consideration of the variation is nX.

Next, using nx calculated by formula (1), the output level
The score SX in a zone of a certain type when detected data of is input is calculated as shown in the following equation.

3, =A' (1-k (10・X-1)')...
-(2) Here, A is a reference point, and j and k are constants.

When formula (2) is expressed in a diagram, it becomes as shown in Fig. 2. in this case.

By appropriately setting A and j, the shape of the curve in FIG. 2 is determined. This results in 2, e.g. output level
The closer the value is to the average value m, the higher the score.

The standard deviation σ can be set as the base 21i, and the score can be set to decrease as the distance from the average value m increases.

By doing this, even if the output level The scores are also different, and the score calculations are normalized to account for the variations. Calculate the score as above.

By performing the test for all output levels from 1 to 256 for each zone of each type and storing the results as dictionary data in the dictionary data memory 6, it is possible to input detected data at a certain level in a certain zone. Sometimes, it is not necessary to calculate equations (1) and (2) for each type of zone each time (it is possible to immediately calculate the score by drawing a table. In the case of Fig. 1, Since the dictionary and calligraphy data are made up of four output levels each as one block, the score is the average value of the four levels.In addition, the average value m, standard deviation σ, equation (1), and (2 ) Calculation of dictionary data using the formula is performed by an arithmetic circuit included in the dictionary data memory 6. Using such dictionary data, the score of the detected data is calculated, and the t and e scores for all zones are calculated as follows. By counting the denominations for each denomination in the denomination register 7, the construction value of the register for the most similar denomination will have the highest score.

FIG. 3 is an example of pattern matching when dictionary data and detected data are similar. It is a combination of the average values of dictionary data by species for each of the 9 broken line zones, and 1
The dashed line 0 connects values that are apart by +σ from the average value of the dictionary data, and the broken line 11 connects values that are similarly separated by −σ. A solid line 12 connects the representative values of the detected data. In this case 1, for example, the second zone and the fourth zone
The detected data of the zone is based on the average value of the dictionary data.
Their output levels are equally far apart. Therefore, in the case of the conventional pattern matching circuit, the scores (degrees of similarity) are the same for the second zone and the fourth zone. However, as can be seen from the figure, in the case of the fourth zone, the value of σ is larger and the variation in dictionary data is larger. This is considered to indicate that the variation in the input detected data is also large. Therefore, in this case, it is necessary to give a lighter score to the fourth zone. According to the invention.

Each score is nX of the above formula (1) with standard 2% deviation σ as the unit.
Since it is determined by the value of , the iU point (similarity) of the fourth zone is larger than that of the second zone. Since the scores normalized by the standard deviation σ are used as a reference in this way, more accurate pattern matching can be performed. In the case of Figure 3, since the detected data is within the range of the average value of the dictionary data ±σ over all zones,
The score of the detected data is high, and there is a high possibility that the detected data is of this type of dictionary data.

FIG. 4 is an example of pattern matching when dictionary data and detected data are not similar. The broken line 13 connects the average value of dictionary data for each species in each zone, the broken line 14.15 connects the values of ±σ from the average value, and the solid line 16 connects the representative value of the detected data. It is something. In this case, the score of the detected data becomes very low, and there is almost no possibility that the detected data is of the type of this dictionary data.

Although the embodiment in FIG. 1 has been described above in the case of a banknote pattern matching circuit, it goes without saying that the present invention can be applied to other pattern matching circuits. In addition, in the dictionary data memory 6 of FIG. 1, each dictionary data is divided into blocks of output levels of representative values every four levels, but if the memory capacity etc. are limited, there is no need to make it professional.

Also, 1M.

Although equation (2) is used as the equation for point calculation, equation (2) is just an example, and other equations can be considered in consideration of the characteristics of the dictionary data. The important points of the present invention are that the points assigned to the dictionary data and the values normalized based on the dispersion thereof are used as standards.

(7) Effects of the Invention The present invention uses dictionary data in which points are assigned based on values normalized by the variation in reference data (standard '! - IK! = deviation) in a pattern matching circuit. It is possible to perform pattern matching with greater flexibility and higher precision against data variations.

[Brief explanation of drawings]

Fig. 1 is an overall configuration diagram of a banknote pattern matching circuit according to the present invention, Fig. 2 is an explanatory diagram of the principle of score calculation according to the present invention, and Fig. 3 is a diagram showing a case where the detected pattern and dictionary data are similar. FIG. 4 is a diagram showing an example of pattern matching when the detected pattern and dictionary data are not similar. 4...Representative value detection circuit, 5...Dictionary data block selection circuit, 6...Dictionary data memory, 7...Denomination specific register. 8... Judgment logic circuit, Zl, Z2. ...
zone memory. Agent Patent Attorney Calibration Hiroshibe”;’, =f4i=2+
Figure 1 Figure 2 (Representative values) Figure 3 Figure 4

Claims (2)

[Claims] (1) In a pattern matching circuit that performs pattern matching between detected data and dictionary data, a similarity setting means for setting a degree of similarity between the detected data and the dictionary data according to variations in the dictionary data; A pattern matching circuit characterized by comprising a determining means for determining the type of the detected data based on the output of the similarity setting means. (2) The similarity setting means is a dictionary that allocates points according to the distance from the average value of the reference data (referred to as n value) in units of standard deviation of the reference data when creating the dictionary data. a storage means for storing data, a selection means for selecting a score of the detected data for each dictionary data stored in the storage means, and a totalization for totaling the scores selected by the selection means for each of the dictionary data. A pattern matching circuit according to claim 1, characterized in that it comprises means.