JP2000215313A - Method and device identifying data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a data identifying method which can identify even one-to- many data and also has a high identification success rate by reducing an erroneous rejection rate and an erroneous acceptance rate and to obtain its device. SOLUTION: In this data identifying method, the collation degree of each registered data with the other registered data is calculated about plural pieces of registered data registered with a registered data storing part 1 by a registration time collation degree calculating part 2a, threshold for each registered data is generated by a threshold generating part 6 on the basis of collation degree distribution calculated from the collation degree by a collation degree distribution calculating part 4, and the collation degree between collation data calculated by a collation time collation degree calculating part 2b and candidate data is compared with threshold for candidate data by a collation deciding part 8 to decide whether or not collation data being an identification object corresponds to the candidate data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data identification method and an apparatus for collating collation data to be identified with one of registered data and judging whether or not they correspond to each other. In particular, the present invention relates to identification of image data such as fingerprints and faces, and data such as feature amounts extracted from images.

[0002]

2. Description of the Related Art When an individual is identified by a personal identification device such as a fingerprint collation device, data such as a fingerprint of a person to be identified and its characteristic amount obtained by a sensor or the like are already registered in large numbers. In many cases, the degree of collation with certain data extracted from data such as a fingerprint or its characteristic amount is calculated, and the identity of the person is often determined based on whether or not the degree of collation exceeds a predetermined threshold.

For example, in a conventional data identification technique used for fingerprint collation such as Japanese Patent Application Laid-Open No. 58-176782, a threshold value for judging a person is already registered as a collation target. It is fixed to a constant value regardless of the characteristics of the data of another person or the data of the current person to be collated.

As another conventional data identification technique, there is a fingerprint matching apparatus disclosed in Japanese Patent Application Laid-Open No. 62-281080, in which a one-to-one matching fingerprint matching apparatus registers a fingerprint. Sometimes, the registrant repeatedly inputs a fingerprint a plurality of times, and calculates a threshold value used by the collator at the time of collation based on the average value and the variance value of the coincidence of all combinations of the respective feature amounts.

Next, such conventional data identification will be described. FIG. 6 is a block diagram showing the configuration of a conventional fingerprint matching device. In the figure, T1 is a keyboard for inputting a password, T2 is a camera for acquiring a fingerprint image,
T3 is a feature extraction circuit for extracting a feature amount based on a fingerprint image, and T4 is a feature memory for temporarily storing feature amounts in T51, T52,.
, T5 is a temporary memory group having temporary memories T51, T52,..., T5n, and T6 is a temporary memory T51, T5 of the temporary memory group T5.
T5n is a data selector for selecting any two of T52,..., T5n, T7 is a coincidence calculating circuit for comparing two data selected by the data selector T6 and calculating the coincidence, and T8 is a data selector T6 A determination level calculation circuit for calculating a determination level of a registered fingerprint from the degree of coincidence obtained by the degree of coincidence calculation circuit T7 for all combinations of data selected in the step T9; T9 is a feature amount from the temporary memory group T5;
A registration memory for recording a set of the judgment levels obtained by the judgment level calculation circuit T8 and a collation judgment circuit T10 for judging the collation based on the degree of coincidence of the feature amount and the judgment level at the time of the collation.

Next, the operation will be described. First, at the time of registration, when the registrant inputs a password from the keyboard T1 and puts a finger on the camera T2, the temporary memory selector T4
Is written into the temporary memory T51. By repeating this operation several times,
The data of the feature amount is sequentially stored in the temporary memories T51, T52,.
Written to T5n. A predetermined number of temporary memory groups T5
, T5n, the data selector T6 sets the temporary memory T5.
, T5n are sent to the coincidence calculating circuit T7, and all the combinations for selecting two of them are sent to the coincidence calculating circuit T7.
Calculates the degree of coincidence between the two data. The degree of coincidence is sent to the judgment level calculation circuit T8, and the judgment level is calculated from the average value and the variance value. When the judgment level is calculated, one of the feature amounts written in the temporary memories T51, T52,..., T5n and the judgment level calculated by the judgment level calculation circuit T8 are stored in the registration memory T9 corresponding to the personal identification number. Stored at the address.

On the other hand, at the time of collation, when the collator inputs a personal identification number from the keyboard T1 and places his finger on the camera T2, the feature amount of the fingerprint extracted by the temporary memory selector T4 is stored in the temporary memory group T5. T51, T52, ...,
Written to one of T5n. Further, the data selector T6 sends the feature amounts recorded in the temporary memories T51, T52,..., T5n to the coincidence calculating circuit T7. The registered feature value stored in the address corresponding to the input password in the registration memory T9 is sent to the coincidence calculating circuit T7 via the data selector T6, and at the same time, the registered feature value is determined. The level is the collation judgment circuit T1
Sent to 0. The coincidence calculating circuit T7 obtains the coincidence between the two feature amounts, and sends it to the collation determination circuit T10. The collation judging circuit T10 compares this with the judgment level sent from the registration memory T9, and outputs the collation result.

[0008]

Since the conventional fingerprint matching device as a data identification device is configured as described above, the threshold value used to determine the identity is uniformly set to a constant value regardless of the person to be matched. Fixed. However, the actual fingerprint collation varies from person to person, with some people tending to take high values and others tending to take low values. Therefore, as in the conventional fingerprint collation device, if the threshold for determining the identity is fixed to a constant value regardless of the person who collates, if the threshold is set higher, it is easier to refuse others. Instead, the person is often rejected by mistake, and if the threshold is set lower, the person may be erroneously accepted instead of making the person easier to accept. In other words, it makes it difficult for the person to be rejected accidentally,
It is difficult to make it difficult for another person to be erroneously accepted, and as a result, there is a problem that the identification success rate is reduced.

[0009] Considering that collation is performed using a threshold value obtained from a plurality of personal feature data, a one-to-one correspondence is considered.
However, there is also a problem that it is not possible to correctly determine the identity of a one-to-many collation, even though it is possible to determine whether the user is the principal in the collation. The reason is that the threshold value obtained from only the feature amount data of the principal does not take into account any relative relationship with the data of the other person registered at present, so that the determination of “the principal” is made at the threshold value This is because even if they can do it, they cannot judge that they are not others.

Further, in order to obtain a highly reliable threshold value only from the statistics of the data input by the registrant, a large amount of data is required, and the registrant must repeatedly input the fingerprint during registration. There is also an operational problem that does not occur.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and enables not only one-to-one but also one-to-many data identification, and reduces the identification error rate to reduce the identification success rate. It is an object of the present invention to obtain a data identification method and an apparatus for improving the data.

[0012]

According to the data identification method of the present invention, a collation degree which is a degree of coincidence with other registered data is calculated for each of a plurality of registered data. Generates a threshold for each registered data based on the collation degree distribution of the collation degree with other registered data calculated, and compares the collation data to be identified with the collation data among a plurality of registered data. The degree of collation with one candidate data as a candidate is calculated, and by comparing the obtained degree of collation with a threshold for the candidate data, it is determined whether the collation data to be identified corresponds to the candidate data. It was done.

In the data identification method according to the present invention, the degree of collation with other registered data is obtained for each registered data from the combination with all registered data other than the currently focused registered data. is there.

[0014] In the data identification method according to the present invention, a degree of collation with other registered data is determined for each registered data from a combination with a part of a plurality of other registered data other than the currently focused registered data. It is what we asked for.

In the data identification method according to the present invention, for each registered data, a collation distribution of collation with other registration data is obtained from a frequency ratio distribution of collation.

In the data identification method according to the present invention, for each registered data, a collation distribution of collation with other registered data is obtained from a function approximating a frequency ratio distribution of collation. .

In the data identification method according to the present invention, a threshold value for each registered data is determined from a collation degree distribution of collation degree with other registered data obtained for each registered data and a target value of a false acceptance rate. It was done.

In the data identification method according to the present invention, the threshold value for each registered data is set as a value indicating that the candidate data compared with the collation data does not correspond to the collation data.

In the data identification method according to the present invention, the threshold value for each registered data is set as a threshold value indicating that the candidate data compared with the collation data corresponds to the collation data.

In the data identification device according to the present invention, for a plurality of registration data registered in the registration data recording unit, the registration-time collation degree calculation unit indicates, for each registration data, the degree of coincidence with other registration data. The collation degree is calculated and recorded in the collation degree recording unit. Based on the collation degree recorded in the collation degree recording unit, the collation degree distribution of each registered data with other registered data is compared with the collation degree distribution. Created by the calculation unit, the collation degree distribution is recorded in the collation degree distribution recording unit, and the threshold value generation unit generates a threshold for each registered data based on the collation degree distribution and records it in the threshold recording unit, The collation degree between the collation data to be identified and one candidate data which is a collation candidate among a plurality of registration data already recorded in the registration data recording unit is calculated by the collation degree calculation unit at the time of collation. , At this time, By comparing the output collation degree with the threshold value for the candidate data among the threshold values recorded in the threshold value recording unit, it is determined whether the collation data to be identified corresponds to the compared candidate data. The determination is made by the determination unit.

In the data identification device according to the present invention, the continuation determining unit determines whether or not to continue the collation based on the determination result of the collation determining unit, and the candidate number generating unit records the data in the registration data recording unit. It is designed to determine which of the registered data is to be collated with the collation data to be identified as candidate data.

In the data identification device according to the present invention, data for determining a candidate number for specifying candidate data in the candidate number generation unit is input from the identification number specification unit.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration of the data identification device according to the first embodiment of the present invention. In the first embodiment, a data identification device that performs one-to-many search matching will be described. In FIG. 1, reference numeral 1 denotes a registration data recording unit that records a plurality of registration data to be compared with collation data to be collated. Reference numeral 2a denotes a registration-time matching degree calculating unit that calculates a matching degree indicating the degree of matching of the registered data recorded in the registered data recording unit 1 with other registered data, and 2b denotes an input. A collation time calculation unit for calculating a collation degree between the collated data thus obtained and candidate data which is a candidate among the registered data recorded in the registered data recording unit 1. Note that the basic operation of the registration-time collation degree calculating unit 2a and the collation-time collation degree calculation unit 2b is to calculate a plurality of collation degrees or one set of collation degrees by calculating the collation degrees from two data. The only difference is that they perform the same operation in principle.

Reference numeral 3 denotes a collation degree recording unit for recording the collation degree calculated by the registration collation degree calculation unit 2a. Reference numeral 4 denotes a matching degree distribution calculating section for creating a matching degree distribution with other registered data for each registered data based on the matching degree recorded in the matching degree recording section 3; The collation degree distribution recording unit records the collation degree distribution created by the calculation unit 4. Reference numeral 6 denotes a threshold generator for generating a threshold for each registered data based on the collation distribution for each registered data recorded in the collation distribution recording unit 5, and 7 is generated by the threshold generator 6. A threshold recording unit that records the threshold value.

Reference numeral 8 denotes a threshold value corresponding to the current candidate data from among the plurality of threshold values recorded in the threshold value recording unit 7, and the collation degree of the candidate data obtained by the collation degree calculation unit 2b at the time of collation. This is a collation determining unit that compares and determines whether the collation data corresponds to the candidate data. Reference numeral 9 denotes a continuation judging unit for judging whether or not to continue the collation based on the judgment result of the collation degree judging unit 8, and 10 receives the judgment result from the continuation judging unit 9 and matches the collation data. A candidate number generation unit that specifies the number of candidate data to be executed.

Next, the operation will be described. First, the operation at the time of registration will be described. At the time of registration, the input registration data is recorded in the registration data recording unit 1. Note that, at the time of this registration, the input operation may be performed a plurality of times, and data such as the maximum value and the average value thereof may be recorded as registration data. The registration-time collation calculating unit 2a calculates the collation of the plurality of registered data recorded in the registration data recording unit 1 with the other registered data for each registered data, and calculates the obtained collation with the collation. The information is recorded in the recording unit 3.

FIG. 2 is a schematic diagram showing all combinations of the collation degrees of the two registration data held by the collation degree recording unit 3. Next, referring to FIG. The operation of the calculation unit 2a will be described. For example, at present, the registration data recording unit 1 has (n-1) identification numbers 1 to n-1.
It is assumed that a number of pieces of registered data have been registered, and one piece of new registered data with n as an identification number is added thereto, and the number of registered data becomes n. Until new registration data is added, it is registered (n-1)
For each of the registered data 1 to (n-1), it can be considered that the degree of collation with other registered data has already been calculated, and the degree of collation is as shown in FIG. , Is stored in the matching degree recording unit 3 as a symmetric (n-1) × (n-1) square matrix.

When the new registration data with the identification number n is recorded in the registration data recording section 1, the registration collation degree calculating section 2a checks the newly registered data of the identification number n and the other data. Only the degree of collation with the registered data will be calculated. The degree of collation of the registered data of identification numbers i and j is x _ij
, X _n1 , x _n2 ,..., X _nn between the newly registered data with the identification number n and the other registered data, and (n−1) different degrees of matching as the symmetric components x _1n ,
x _2n ,..., x _nn−1 are calculated and they are (n−1) ×
It is added to the (n-1) matrix and recorded in the collation degree recording unit 3 as an n × n square matrix.

FIG. 3 is an explanatory diagram showing an example of a frequency rate distribution of the degree of collation of two pieces of registered data with other registered data, which is calculated by the collation degree distribution calculating section 4. FIG. FIG. 7 is an explanatory diagram showing an example of a matching degree distribution when a frequency distribution distribution of matching degrees between two registered data calculated with a matching degree distribution calculating unit 4 with other registered data is approximated by a normal distribution. The operation of the matching degree distribution calculation unit 4 will be described below with reference to FIGS.

First, the matching degree distribution calculating section 4 creates a frequency rate distribution of the matching degree for each registered data among the registered data recorded in the matching degree recording section 3.
That is, the data of the identification number 1, the matching degrees _{x 12, x 13, ...,} x 1n, the data of the identification number 2, matching degrees x _21, x _23, ..., from x _2n, · · · · , _{Xii -1} ,..., _Xii-1 ,
x _{ii + 1,} ..., from x _in, collating the data of the identification number n of x _n1, x _n2, ..., from x _nn-1, generates a frequency index of each matching degree. FIG. 3 shows an example of the frequency ratio distribution of the degree of matching with the other registered data for the registered data whose identification numbers are i and j. As shown in the drawing, the degree of collation with other registered data is distributed near a low value, indicating that the form of the distribution differs depending on the registered data.

Next, a matching rate distribution is obtained by approximating the frequency ratio distribution of the matching degree with other registered data obtained for each registered data by an appropriate function. FIG. 4 shows the registration data of the identification numbers i and j shown in FIG.
The distribution of the matching degree when the frequency rate distribution of the matching degree with other registered data is approximated by a normal distribution is shown. When the distribution of the degree of collation of the registered data with the identification number i is obtained by the normal distribution, the average _mi and the variance σ _i ² with the other registered data are calculated using the following equations (1) and (2), respectively. Then, using them, the collation distribution f _i (x) is approximated by the following equation (3). Here, the function approximating the frequency distribution of the matching degree is not limited to the normal distribution, and any other function may be used.

[0032]

(Equation 1)

When these matching degree distributions are obtained for all the registered data, the matching degree distributions are recorded in the matching degree distribution recording unit 5.

Next, the operation of the threshold generator 6 will be described with reference to FIG. The threshold generation unit 6 calculates a threshold based on the collation distribution obtained for each registration data recorded in the collation distribution recording unit 5. The threshold generation unit 6 obtains a threshold for each registered data as follows based on the target value of the false acceptance rate and the distribution of the matching degree. That is, assuming that the target false acceptance rate is far and the threshold for the registration data of the identification number i is t _i ,
The relationship represented by the following equation (4) holds between the registered data of the identification number i and the matching degree distribution f _i (x) of the other registered data. Therefore, the threshold value t _i can be obtained as a solution of the equation (4). In this case, the threshold value t _i is obtained from the degree of collation with other registered data and the target false acceptance rate. Does not correspond ".

[0035]

(Equation 2)

Finally, the threshold value t _i obtained by the threshold value generation unit 6 is recorded in the threshold value recording unit 7, and the registration operation is completed.

Next, the operation at the time of collation will be described. At the time of collation, the collation at the time of collation calculation unit 2b compares the candidate data selected from the registered data recorded in the registered data recording unit 1 based on the designation of the candidate number generation unit 10 with the inputted collation data. Find the degree of collation. The obtained degree of collation is sent to the collation judging unit 8 and compared with a threshold corresponding to the candidate data recorded in the threshold recording unit 7. The collation determination unit 8 determines whether the candidate data corresponds to the collation data by this comparison, and outputs the identification result.

Here, when the collation data and the candidate data correspond, the search collation ends, and when an identification result indicating that they do not correspond is obtained, the continuation determining unit 9 determines whether or not to continue the collation. I do. As a result, if the search is not continued, the search and collation ends, and if the collation is continued, the continuation determination unit 9
The trigger signal is sent to the candidate number generation unit 10. The candidate number generation unit 10 receiving this trigger signal determines a candidate number for selecting candidate data for the next collation. These operations are repeated until the continuation determination unit 9 determines that the search is not to be continued due to conditions such as terminating the search a certain number of times or ending the search of all data.

As described above, according to the first embodiment, in order to set a dedicated threshold for each registered data based on the degree of matching with the other registered data, for example, Dynamically based on the characteristics of the current registration data, such as increasing the threshold for data with a high degree of collation and lowering the threshold for data with a low degree of collation with other registered data. Because the threshold can be set, compared to the case where the threshold is set uniformly regardless of the nature of the registration data,
It is possible to reduce the false rejection rate, which is not correctly identified as a principal, and the false acceptance rate, which is mistakenly identified as a principal, but as a result, it is possible to improve the identification success rate, Further, since the threshold value is calculated for each registered data by using the combination matching degree with other registered data in the calculation of the threshold value, it is possible to obtain an effect that there is no need to input many data at the time of registration. .

Further, when the threshold value is calculated, the collation degree distribution of the collation degree with other registered data and the target value of the false acceptance rate are used. Candidate data does not correspond ". Therefore, there is an effect that the present invention can be applied to search collation in which collation is performed one after another with other registered data as candidates and it must be determined that the data does not correspond to collation data.

Embodiment 2 In the first embodiment, the registration-degree-of-registration calculating unit 2a calculates the degree of matching of each registered data with all other registered data. However, the matching-degree distribution calculating unit 4 calculates the matching degree distribution. In this case, in a situation where the number of registered data is very large, an approximate value of the matching degree distribution can be obtained without necessarily obtaining all the matching degrees.

For this reason, in the second embodiment, the registration-degree-of-registration calculating unit 2a does not calculate the degree of comparison between each registered data and other registered data, but calculates one of the registered data. It is determined from the combination with the department.

When the n-th new registration data is registered, the matching-degree-of-registration calculating unit 2a calculates not all matching degrees with other registered data but all matching degrees. The matching degree is obtained for a number m (m <n) in a range that can be sufficiently approximated to the statistics obtained in (1). For example, n = 10
In the case of a sufficiently large value such as 00, even if a value such as m = 0.8 × n, there is no large difference between the statistic of the m matching degrees and the n statistic values, and based on this, the matching degree distribution calculating section 4 and the threshold value obtained by the threshold value generation unit 6 have substantially the same values as those obtained from the n matching levels.

For the combinations for which the degree of collation was not calculated at the time of registration, when the degree of collation was determined by the collation determining unit 2b at the time of collation, the collation was recorded in the collation degree recording unit 3b.
Interpolate by recording in

As described above, according to the second embodiment, since the time required for data registration can be reduced, an effect that is effective when it is desired to perform data collation immediately after registering new registered data. Is obtained.

Embodiment 3 In the first embodiment, the case where the matching rate distribution calculating unit 4 approximates the frequency rate distribution of the matching degree to an appropriate function (normal distribution) is described. In this case, the frequency rate distribution may be used as it is.

Therefore, in the third embodiment, the frequency distribution of the matching degree is used as it is in the creation of the matching degree distribution in the matching degree distribution calculating section 4.

In this case, the threshold generation unit 6 calculates a threshold for each registered data as follows. That is, assuming that the target false acceptance rate is far and the threshold value for the registration data of identification number i is t _i , each matching degree distribution of the registration data of identification number i and other registered data (frequency rate distribution of matching degree) ) The minimum threshold value t _i that satisfies the following equation (5), which indicates the relationship between g _i (x) and the false acceptance rate, far:

[0049]

(Equation 3)

As described above, according to the third embodiment, when the collation degree distribution unit 4 creates the collation degree distribution,
Since the frequency ratio distribution of the matching degree is used as it is, the effect that the threshold value can be obtained even when handling data having a distribution in which the matching degree cannot be approximated by an appropriate function is obtained.

Embodiment 4 FIG. In the data identification device shown in the first embodiment, the threshold value generation unit 6 calculates a threshold value for each registered data from the degree of collation with other registered data and the target false acceptance rate for each registered data. Therefore, within the range of the target false acceptance rate, it indicates that “the collation data currently being matched does not correspond to the candidate data”, but the left side of the equation (4) is very much smaller than the false acceptance rate far. May be provided with another threshold value indicating that “the collation data currently being collated corresponds to the candidate data”.

Hereinafter, such a fourth embodiment of the present invention will be described.
Will be described. First, the operation at the time of registration will be described. Here, "verification data does not correspond to candidate data"
A threshold t _i indicating that the operation will be described when searching matching using both thresholds u _i representing that "matching data is equivalent to the candidate data." Note that, when newly registered data is registered, the operation from the time when the matching degree distribution f _i (x) between the newly registered data and other registered data is obtained and recorded in the matching degree distribution recording unit 5 is described. Since it is as shown in the first embodiment, the description is omitted here.

When the obtained collation degree distribution f _i (x) is recorded in the collation degree distribution recording unit 5, the threshold value generation unit 6 calculates the target false acceptance rate far based on the equation (4). T threshold
seeking _i, at the same time, using a very small value far2 compared to the false acceptance rate _far, calculate a threshold u _i from the following is the formula (6). The small value f used at that time
ar2 is, for example, 1/1000 of the false acceptance rate far
A degree value is used. The threshold value u _i thus obtained is recorded in the threshold value recording unit 7 together with the threshold value t _i , and the registration operation ends.

[0054]

(Equation 4)

Next, the operation at the time of collation will be described. The operation at the time of collation is the same as that of the first embodiment except for the operation of collation judging unit 8, and therefore, only collation judging unit 8 will be described here. The collation determination unit 8 determines whether the collation levels of the collation data and the candidate data correspond to each other. This determination is first made by comparison with a threshold value t _i . If the matching degree is smaller than the threshold value t _i , it is determined that the matching data does not correspond to the candidate data. On the other hand, if the collation degree is larger than the threshold value t _i , it is then compared with the threshold value u _i . At this time, if the matching degree is larger than the threshold value u _i , it is determined that the matching data and the candidate data correspond. When the matching degree is smaller than the threshold value u _i ,
Only when it is determined that there is a possibility that the collation data and the candidate data correspond, the continuation determining unit 9 is left to determine whether to continue the search. These operations are repeated until the continuation determination unit 9 determines that the search is not to be continued due to conditions such as terminating the search at a fixed number of times or ending the search of all data.

As described above, according to the fourth embodiment, since the threshold value “the collation data currently being collated corresponds to the candidate data” can be set, the candidate data corresponding to the collation data during the retrieval collation can be set. As a result, an effect is obtained that the continuation determining unit 9 can be provided with a material for determining whether or not to continue the search.

[0057] In the above description has dealt with the case of obtaining the threshold value u _i which is a reference for determining that the "matching data and candidate data currently being collation corresponding" Equation (6), the threshold value The expression for determining u _i is not limited to this expression (6), and from that meaning, the average value or average of the degree of collation between data entered at different times with the same identification number in each registered data. The sum of the value and the variance may be used.

Embodiment 5 FIG. In the first embodiment, 1
Although the data identification device that performs one-to-many search collation has been described, the present invention can also be applied to a data identification device that performs one-to-one search collation.

FIG. 5 shows such a fifth embodiment of the present invention.
Is a block diagram showing a configuration of a data identification device according to
Corresponding parts are denoted by the same reference numerals as in FIG. 1, and description thereof is omitted. In the figure, reference numeral 11 denotes an identification number designating unit for designating an identification number of collation data at the time of collation. As a method of designating the identification number of the collation data, a number such as a password, a character string such as a password, or an ID card in which such information is written may be used. . The data identification device according to the fifth embodiment is different from each of the above embodiments in having the identification number designating unit 11 instead of the continuation determining unit 9.

Next, the operation will be described. Here, the operation at the time of registration is the same as that of the first embodiment, so that the description thereof will be omitted, and only the operation at the time of verification will be described. At the time of one-to-one matching, in order to designate a unique identification number attached to data to be matched, the identification number designating section 11 inputs a password or password from a keyboard or the like, or reads an ID card with a card reader. The identification number designation unit 11 associates the data thus input with an identification number used inside the data identification device. The candidate number generation unit 10 specifies the identification number specified by the identification number specification unit 11 as the identification number of the candidate data for collation. The collation-time collation-degree calculating unit 2b reads the registration data corresponding to the identification number designated by the identification number designation unit 11 from the registration data recording unit 1 as candidate data, and calculates the collation degree with the collation data.

The collation degree obtained in this way is sent to the collation judging section 8 and read out of the thresholds recorded in the threshold recording section 7 for the registration data corresponding to the identification number. Is compared to When the collation determination unit 8 outputs the identification result based on the comparison, the operation at the time of collation ends.

As described above, according to the fifth embodiment, the discrimination number designation section 11 designates the identification number of the candidate data to be collated from now on, so that one-to-one search / collation becomes possible. Is obtained.

[0063]

As described above, according to the present invention, the collation degree distribution is obtained by using the collation degree distribution for each registered data obtained from the collation degree of the combination of the registered data and other registered data. Since the threshold for each data is determined from the required false acceptance probability, for example, for data with a high degree of collation with other registered data, the threshold is increased, and the threshold for other registered data is increased. The threshold can be dynamically set based on the characteristics of the current registration data, such as lowering the threshold for data with a lower collation degree. Compared to the case where a threshold is set, the false rejection rate in which the user is not correctly identified as a person or the false acceptance rate in which the person is mistakenly identified as the other person can be suppressed. The rate can be improved There is a result.

According to the present invention, the threshold value is calculated by using the degree of collation in combination with other registered data. Therefore, in order to obtain the threshold value for each registered data, a number of data are inputted at the time of registration. This has the effect of eliminating the need for it.

According to the present invention, since the threshold value for each registered data is obtained based on the degree of matching with other registered data, the threshold is obtained from the statistics of a plurality of registered data having the same identification number. With the threshold value, it can be determined that the matching data corresponds to the candidate data, but it cannot be determined that the matching data does not correspond to the other registered data, but it can be determined that the candidate data does not correspond to the matching data.
This has the effect of enabling many-to-many matching.

According to the present invention, since the degree of collation is determined from a combination with a part of other registration data,
The time required for data registration can be shortened, and there is an effect that confirmation and collation can be performed immediately after new registration data is registered.

According to the present invention, since the frequency distribution of the matching degree is used as it is for the creation of the matching degree distribution in the matching degree distribution calculating section, the matching degree cannot be approximated by an appropriate function. There is an effect that a threshold value can be obtained even when handling data having a distribution.

According to the present invention, since the threshold value is set as the threshold value indicating that the collation data currently being collated and the candidate data correspond to each other, the candidate data corresponding to the collation data can be identified during the retrieval collation. Thus, there is an effect that the material for determining whether to continue the search can be provided to the continuation determining unit.

According to the present invention, since the data for determining the candidate number of the candidate data in the candidate number generation unit is input from the identification number designation unit, one-to-one search and collation can be performed. There is an effect that it becomes possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a data identification device according to a first embodiment of the present invention.

FIG. 2 is a schematic diagram showing all combinations of collation degrees of two registered data held by a collation degree recording unit according to the first embodiment.

FIG. 3 is an explanatory diagram illustrating an example of a frequency rate distribution of the degree of matching between two pieces of registered data and other registered data, calculated by a matching degree calculation unit according to the first embodiment.

FIG. 4 shows an example of a matching degree distribution calculated by a matching degree calculating unit according to Embodiment 1 when a frequency distribution of matching degrees of two pieces of registered data with other registered data is approximated by a normal distribution. FIG.

FIG. 5 is a block diagram showing a configuration of a data identification device according to a fifth embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a fingerprint matching device which is a conventional data identification device.

[Explanation of symbols]

1 registration data recording unit, 2a registration collation degree calculation unit, 2a
b Collation degree calculation unit at the time of collation, 3 collation degree recording unit, 4 collation degree distribution calculation unit, 5 collation degree distribution recording unit, 6 threshold generation unit, 7 threshold recording unit, 8 collation judgment unit, 9 continuation judgment unit, 10 candidates Number generation unit, 11 identification number designation unit.

 ────────────────────────────────────────────────── ─── Continued from the front page (72) Inventor Koichi Sasakawa 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F-term (reference) 3E040 DA02 5B043 AA09 BA02 BA04 EA09 FA03 FA07 FA09 GA02 5L096 AA06 BA02 BA15 BA18 FA35 GA51 HA08 JA03 KA15

Claims (11)

[Claims] 1. A data identification method for determining whether or not matching data to be identified corresponds to one candidate data which is a candidate for matching among a plurality of registered data already recorded, For each of the plurality of registered data, a matching degree indicating a degree of matching with other registered data is calculated for each of the registered data, and from each of the registered data, a matching degree with the other registered data is calculated. A matching degree distribution with other registered data is created, and a threshold value for each registered data is generated based on the obtained matching degree distribution, and the matching data to be identified and the registered plural Calculating the degree of collation with one candidate data which is a candidate for collation among the registered data of the above, comparing the obtained collation degree with a threshold for the candidate data among the thresholds, and collating the identification target. data , Data identification method characterized in that it is determined whether or not corresponding to the candidate data. 2. The data according to claim 1, wherein for each registered data, the degree of collation with other registered data is obtained from a combination with all registered data other than the currently focused registered data. Identification method. 3. The method according to claim 1, wherein a degree of collation with each of the registered data is obtained from a combination with a part of a plurality of registered data other than the currently registered data. Item 2. The data identification method according to Item 1. 4. The data identification method according to claim 1, wherein for each registered data, a collation degree distribution of collation degree with other registered data is obtained from a frequency ratio distribution of collation degree. 5. The data identification method according to claim 1, wherein for each registered data, a collation distribution of collation with other registration data is obtained from a function approximating a frequency rate distribution of collation. . 6. The method according to claim 1, wherein a threshold value for each registered data is obtained from a collation degree distribution of collation degrees with other registered data obtained for each registered data and a target value of a false acceptance rate. Data identification method described, 7. The threshold value for each registered data is set as a value indicating that candidate data compared with the collation data does not correspond to the collation data. 2. The data identification method according to claim 1. 8. The method according to claim 1, wherein the threshold value for each registered data is set as a value indicating that the candidate data compared with the matching data corresponds to the matching data. 2. The data identification method according to claim 1. 9. A registration data recording unit for recording registration data, and a collation degree indicating a degree of matching of each of the plurality of registration data recorded in the registration data recording unit with other registration data. Based on the collation degree recorded in the collation degree recording unit that records the collation degree calculated by the registration collation degree calculation unit that calculates the collation degree calculated by the registration collation degree calculation unit, A matching degree distribution calculating unit that calculates a matching degree distribution with the other registered data for each registered data; a matching degree distribution recording unit that records a matching degree distribution calculated by the matching degree distribution calculating unit; A threshold generation unit configured to generate a threshold for each of the registration data based on a collation distribution recorded in a collation distribution recording unit; a threshold recording unit configured to record a threshold generated by the threshold generation unit; Recording section And one candidate data is already candidates for matching among the plurality of registration data being recorded,
A matching degree calculating unit that calculates a matching degree with matching data to be identified; a matching degree calculated by the matching degree calculating unit; and a threshold value recorded in the threshold value recording unit. And a comparison determination unit that determines whether the comparison data to be identified corresponds to the compared candidate data by comparing the comparison data with a threshold value for the candidate data. 10. A continuation judging unit for judging whether or not to continue collation based on the judgment result of the collation judging unit; and receiving the judgment result of the continuation judging unit, and recorded in the registration data recording unit. 10. The data identification apparatus according to claim 9, further comprising a candidate number generation unit that determines which of the plurality of registered data is to be compared with the matching data to be identified as candidate data. apparatus. 11. A candidate number generating unit for determining which registered data is to be compared with matching data to be identified as candidate data from a plurality of registered data recorded in a registered data recording unit; 10. The data identification device according to claim 9, wherein the candidate number generation unit includes an identification number designation unit to which data for determining a candidate number for designating candidate data is input.