DE102010000985A1 - Method for separating group of people into several target audiences, involves linking basic pattern with each target audience by calculating target distribution and configuring associations of basic patterns to target audience - Google Patents

Abstract

The method involves inputting desired relative size of each target audience into data processing device. The group of basic patterns derived from identification image is selected, and the standard distribution value is determined. A basic pattern is linked with each target audience by calculating target distribution, calculating deviations of computed target and configuring associations of basic patterns to target audience. The biometric identification image is detected and checked on correspondence with basic pattern to assign to target audience.

Description

The invention relates to a method for the random allocation of a group of persons into a plurality of target groups, wherein the total size of the group can be indefinite and wherein the relative size of the target groups can be predetermined.

In many applications it is necessary to divide a group of people according to random principles into several target groups.

For example, subjects in a test group must be divided into groups for testing a new pharmaceutical preparation. The individual subgroups then receive, for example, the active substance to be tested in different dosages, while a comparison group is tested with placebos or already known substances. A division of the total group at random makes it possible here that the assignment is free of subjective selection criteria of those persons who lead the test and certainly can not be influenced by the subjects.

Another example is gambling. In the case of a tombola, for example, the persons have a ticket that is provided with an individual identifier such as a number. From the total number of tickets awarded then the winners are drawn in a known manner and assigned to the winning ranks.

Another application example is the conduct of surveys for the purpose of market research. Here, too, it may be a requirement not to specifically select the test persons according to specific criteria from the outset, but according to random principles, whereby in the sense of the present invention a randomness does not necessarily mean a selection according to the laws of stochastics, but that it is a splitting method which can neither be influenced by the subjects in the group to be split nor by persons who act as test leaders.

In the examples given, however, it is always the case that for a non-subjective division of the total group into several target groups via a random selection, the total size of the group of persons or the lots or other codes held by them must be established before they are divided into the target groups or profit classes , Only then can it be achieved by establishing specific criteria for the target group assignment that the desired size distribution of the target groups is achieved in practice.

If, for example, lots are assigned to the starting group whose final digits are assigned in exactly equal proportions, so that there is an occurrence probability of exactly 10%, then by bundling several final digits to a target group, their size can be increased to 20% -90%. However, if people who have already received a lot do not take part in the draw, then individual target groups are understaffed. If subsequently additional individual persons are given lots, even though the raffle has already started, then the desired target group distribution is likewise not reached.

In selection systems such as a raffle, therefore, all the lots must first be issued before the draw, ie the division of the group of losers into several target groups according to the winning ranks, can begin, otherwise the target group size to be specified via the lots will not be reached.

If, on the other hand, spontaneously generated random numbers are selected, then these can be supplemented as desired, so that persons who subsequently join the initial group can be classified into a target group via the random number. The drawback, however, is that the theoretically calculated target group distribution only sets in practice with very large starting groups. For small group sizes, however, one or the other target group may be more or less frequently represented than might be expected according to the stochastic laws.

A random selection in the sense of a stochastic result thus has the disadvantage that either the group to be distributed must be determined from the outset in its size and with its content, for example the total amount of all lots, so that a selection in the sense of a raffle is to be made. or that a random allocation of criteria for the division of the group into several target groups with a small group size can lead to an irregular distribution.

It is therefore an object of the present invention to provide a method for dividing a group of persons into several target groups, which makes it possible, without knowing the size of the groups, to be able to indicate a relative size for each target group and to ensure that, taking into account a certain amount Tolerance threshold - the specified group size is reached.

This is achieved by a method having the features of patent claim 1.

The basic idea underlying the invention is to read in a biometric identification image of the persons in the group to be divided by means of an electronic reading device, to fix certain pattern structures and to assign the person to one of the target groups on the basis of the detected pattern structures.

"Identification image" in the sense of the present invention means that it is an electronic image of the selected body features or a data set with the selected body features, which can be obtained by machine during scanning of the body. It does not have to be an optical image, but it can also be an already machined data set that contains the data extracted from the person's body. For example, in the case of using a fingerprint, the identification image may be either the optical representation of the fingertip in a digital data format, or it may just be a formatted data set containing the features obtained from the fingerprint. A gene sequence obtained from body tissue or body fluids would also be an "identification image" in this sense.

The use of a biometric identification image has the following advantages:
There are inherent basic patterns within the identification images that occur in any case in the group of people to be divided, or where non-existence is an additional characteristic feature per se.

A basic pattern exists with each member of the group of persons, regardless of whether the person is aware that they are participating in a selection process or not. In a figurative sense, almost every person in the whole group to be split up has their personal raffle ticket right from the start. It does not matter if the group grows beyond its original size or how much it grows.

According to the invention, it is thus possible to divide even a small group of persons into several target groups whose respective relative size corresponds to the specification. Likewise, it is possible according to the invention to divide very large groups. It is decisive, however, that with the method according to the invention using a data processing system and an electronic reader for biometric features, the group size can increase even further during the already ongoing division and can not be influenced by the participants or the line managers of the selection procedure.

• – die geeignet sind, verfälschungssicher Personen voneinander zu unterscheiden,
• – die mit einer elektronischen Leseeinrichtung erfassbar sind und
• – die in einer hinreichend großen Zahl von Ausprägungen mit genügend hoher Auftretenswahrscheinlichkeit vorkommt, um überhaupt eine Aufteilung in mehrere Zielgruppen vornehmen zu können.

As a biometric identification image, such expressions of the body are suitable

• - which are suitable for distinguishing people from each other in a tamper-proof manner,
• - Which are detectable with an electronic reading device and
• - which occurs in a sufficiently large number of occurrences with a sufficiently high probability of occurrence in order to be able to make a division into several target groups at all.

The advantage of the invention is further that, at least for certain ethnic groups, the frequency distribution of the occurrence of these basic types of identification images is known, so that the method of the invention can make the individual target group assignment based on the given frequency distributions.

In contrast to conventional selection procedures or competitions, which are based, for example, on the previous issuing of passes or lots, unwanted transfer from one person to another is excluded in the method according to the invention because it is based on the processing of biometric identification images. Thus, the method of the invention is tamper-proof.

The method according to the invention in its simplest embodiment is described as follows:
As part of a sweepstakes in a point of sale, the participants are to receive prizes such as shopping vouchers or prizes. The relative distribution of profit groups is predetermined. High-quality prices should be a special feature and therefore rarely awarded. The total size However, it is unknown at the beginning of the procedure, since it is typical for such actions that at the beginning only a small group is present and, depending on the frequency of the customer, more or fewer people join.

The biometric identification image used is an electronic fingerprint scan of any fingertip.

In the simplest case, the lead person has set the target group sizes before the start so that, with a certain tolerance, they correspond exactly to the distribution of the basic pattern types on the fingerprint. In this simple case, only a 1: 1 assignment of the basic pattern to the target groups Z _{i is} made in the analysis device.

In the current process, each participating person is then fed to the output group of a reading device which records the biometric identification image. For example, a known fingerprint scanner is used.

Subsequently, the detected feature image is analyzed in the data arrangement device and assigned to a specific basic pattern group W _{pattern, i} , wherein the abovementioned, relative distribution of the occurrence probability of the individual basic patterns W _{pattern, i is} natural and is therefore known within certain limits.

In the case of a person from the group of raffle participants, therefore, only the basic pattern and not the individual features of the fingerprint are detected in the analysis device in this embodiment of the method, and the target person is classified into one of the winning groups on the basis of this basic pattern or a different handling instruction will be used Target group is assigned.

The advantage is that each additional person also possesses one of the basic patterns on their fingerprint, so that it is irrelevant whether the raffle is started only with a small group of, for example, 10 people, or whether the group in the course of performing the raffle action still growing to several thousand participants.

In the manner known from mathematics and computer science, condition definitions, group names and actions are indexed below.

The index "i" indicates the assignment of the person or the selected features of their identification image to the respective target group Z _i and identifies the actually measured size distributions W _{ist, i} or the expected or desired probabilities W associated with this target group Z _{i should, i} .

The index "1-n" is used in the following as an abbreviation to denote the sequence from the first to the nth element of a group. The index i also serves as a running variable for the current value within the sequence of possible values 1 to n. B. "H _1-n " as an abbreviation for the sequence of actions H ₁ ... H _n .

• – Der Index i gibt die Nummer der Zielgruppe Z_i an, z. B. i = 3.
• – Z_1-6 bedeutete, dass es sechs Zielgruppen von Z₁ bis Z₆ gibt.
• – W_soll,i kennzeichnet die relative Größe der 3. Zielgruppe Z₃ innerhalb der Gesamtzahl der Zielgruppen Z₁ bis Z₆.

In a simple example, where the process is performed with six target groups Z _1-6 , the terminology used is explained:

• - The index i indicates the number of the target group Z _i , z. I = 3.
• - Z _1-6 meant that there are six target groups from Z ₁ to Z ₆ .
• - W _{soll, i} denotes the relative size of the third target group Z ₃ within the total number of target groups Z ₁ to Z ₆ .

In the simple case, which is first described here, the indices "i" of the target groups Z _i also agree with the indices of the respectively assigned basic pattern classes; otherwise, another index such as eg. For example, "j" can be used. For example, W _{pattern, 3} marks the relative size of the third basic pattern class in fingerprints.

The simplest case in the method according to the invention is therefore as follows:
A person has their finger scanned. The data processing device determines therefrom a fingerprint of the fifth basic pattern class W _{pattern, 5} . This basic pattern class W _{pattern, 5} is also associated with the fifth target group Z ₅ . For the fifth target group Z5 there is also an associated instruction H ₅ .

But this is not necessarily so. However, in the case of the other method variants described below, a plurality of basic pattern classes can also be assigned to a target group. For example, then a fingerprint that corresponds to the third basic pattern W _{pattern, 3} , and also a fingerprint that corresponds to the fourth basic pattern W _{pattern, 4} , into the second target group Z ₂ , depending on how the Zurodnugnsregelen were set up.

Preferably, the biometric characterization feature is a fingerprint as described in the above embodiment. Fingerprints are, as known in criminalistics, to distinguish people of any origin from each other. One of five basic patterns, the so-called pattern, as well as further characterization features, so-called minutiae, are always present within the fingerprint.

Even if a fingerprint due to a previous injury of the fingertip is no longer in the natural form, the person can still be included in the starting group by z. B. just another finger is displayed.

• – eine Aufnahme der Iris des menschlichen Auges,
• – körpereigene Gerüche
• – biometrische Merkmale im Gesichtsfeld wie beispielsweise Abstände zwischen Mund, Nasenspitze, Nasenwurzel oder Augen oder
• – eine Gensequenz.

Other identification images can be:

• A photograph of the iris of the human eye,
• - body odors
• - visual field biometric features such as distances between the mouth, tip of the nose, root of the nose or eyes or
• A gene sequence.

Preferably, it is provided in the method according to the invention to perform a duplicate check. For this purpose, a certain fingerprint is always used in the course of the implementation, z. B. the index finger of the right hand. The scanned fingerprint and / or the individual features of the fingerprint captured thereon are stored in a database.

This procedure, familiar from the field of criminalistics, serves to prevent the same person from having twice the participation in the competition. If, for example, a competition is involved in a competition, this prevents the person from receiving multiple prizes.

If a person who has previously participated reaches the reader device once again, a comparison with all data records previously stored in the database is carried out in the analysis device before being assigned to the target groups. If a match is found with a previously stored data record, then the analysis device no longer assigns to one of the target groups, which means in practice that the person can not participate in the raffle a second time.

The method according to the invention can have further advantageous embodiments, which make it possible to approximate the target group distribution more precisely to the output-side predetermined target probability. Further advantageous embodiments serve to define feature combinations in order to be able to process target probabilities which deviate to a greater extent from the natural distribution of the basic pattern types or whose number is greater or smaller than the number of basic patterns.

First, the preferred additional process steps will be described, which serve to improve the accuracy:
In an advantageous embodiment of the method according to the invention not only the assignment of the detected identification image is made to one of the target groups in the processing device, but it is running in the entirety of the group distribution of the basic pattern and possibly the distribution of the other used for characterization individual features ongoing Updated by writing the calculated values to a database. It then becomes a comparison between the determined probabilities for the occurrence of each feature combination W _{i, i} with the empirical-based distributions assumed at the beginning of the procedure.

For example, it is known that certain basic patterns of a fingerprint occur more frequently within certain ethnic groups than in other ethnic groups. As a result, different distributions can result according to which geographical location or within which population group the procedure is carried out.

The underlying distribution of the basic patterns and / or the individual features is then adjusted after a configurable number of passes and stored in adapted form again in the database, so that when re-performing the method in the same environment improved accuracy is achieved.

A further advantageous additional variant of the method according to the invention provides for a check, referred to herein as delta correction, of the deviation of the actual probabilities W _{i, i} for the occurrence of specific feature combinations of the desired probabilities W _{soll, i} and thereafter with the aid of the data processing device To influence the target group size such that their actual distribution of the desired target distribution W _{soll, i is} approximated. This correction also begins after a manually configurable number of passes.

The number of passes after which a correction is made may be a fixed amount.

However, it is also possible to specify a random number of passes through an additional random generator within the data processing device.

Due to the fact that the correction takes place automatically and without operator intervention, there is once again manipulation security in the method according to the invention, even if the stochastic distribution of the occurrence of certain feature combinations is actively manipulated by the intervention of the data agreement system in order to achieve the desired target group size. However, since the occurrence of a correction value can not be predicted per se and also the automatically performed correction can not be triggered by the operator at will, there is a high degree of certainty that a subjective influence by the operator is avoided.

In the so-called delta correction method, therefore, during the execution of the method, that is to say, for example, during the performance of the raffle action, it is checked how large the deviations of the probability W determined over the entirety of the participating persons _{are, i} to W _{soll, i} for the respective target groups are.

The values W _{i, i are} calculated as the ratio of the number of persons assigned to the respective target group Z _i to the total number of all persons who have participated in the pass so far in all target groups.

If the deviations lie within a configurable tolerance field, then it is checked in a normal manner in which of the target groups Z _1-n the person is to be classified on the basis of the checked features and the action H _i corresponding to the respective target group Z _i becomes the plurality of possible Actions H _{1-n are} made.

In simple terms, this means that the measurable distribution actually existing within the target group coincides with the theoretically expected probabilities for the occurrence of the defined characteristics. Therefore, the captured identification images need only on the basis of the established in the previous procedure criteria, eg. B. the occurrence of certain basic patterns, are assigned to the individual target groups.

If, on the other hand, when performing the delta-correction method in the main procedure, it is determined that the measured distributions W _{is 1-n} outside the allowable tolerance range, then the action where the negative deviation is between W _{, i} and W is made _{should, i is} greatest.

In other words, that target group is searched out whose previously occurring frequency is smallest compared to the theoretical and / or previously stored empirical values defined before the start of the main procedure, and this target group is then artificially replenished by the data processing device as follows:
If, for example, the third target group W _{soll, 3 has} an occurrence probability corresponding to the expected frequency of occurrence of the basic pattern W _{pattern, 3} in the fingerprints of W _{pattern, 3} = 27.9%, and then it is determined that in the previously evaluated overall group of If the characteristic combination for the third group is only with a probability of W _{, 3} = 22%, then further persons are placed in this underrepresented third target group until they are again in the third group Tolerance limit is. This applies even if the features given in the fingerprint of the persons actually do not lead to a classification in exactly this target group.

The usual allocation procedure is thus temporarily bypassed in the automatic process.

In the case of the described method variant with the self-correction, subprints are also conceivable in order to improve the accuracy and to increase the security against tampering.

For example, it could be provided that the underrepresented group is successively filled up until the measured distributions W _{i, i} again agree with the target probabilities W _{soll, i} .

Furthermore, it can be provided not to classify each subsequent person in the underrepresented group, but to continue to carry out the usual distribution according to the criteria established in the previous procedure.

Furthermore, it can be provided that after the first determination of a correction requirement, only one or a few persons are artificially classified as an underrepresented group, but that the preset test interval in the data processing system is automatically shortened until the measured actual distribution returns to the target distribution and that only then is the user preset value for the number of passes resumed.

In addition to the group distributions, which are based on a body-individual characteristic, a superimposition of a purely time- and / or random-controlled process may be possible, in which an additional action is taken in the data processing device after a predetermined time or a random number of persons.

If the procedure is applied to subjects in scientific studies, an additional random sample can be selected.

In the case of the application of the method according to the invention thereby a single price can be played.

The inventive method is explained with further advantageous embodiments below using exemplary embodiments, with reference to as 1 to 3b attached flow charts is taken. Showing:

1 a pre- and post-processing procedure surrounding the main program,

2 the flow of the main program in a simple embodiment and

3 . 3b the flow of the main program in a preferred embodiment.

4 a bar chart as a scheme for the size distribution;

5 a fingerprint in a coordinate system.

A method for randomly dividing a group of persons into a plurality of target groups in its simplest form preferably has pre- and post-processing procedures (cf. 1 ) between which the actual main procedure (cf. 2 and 3 ) is carried out.

In the described embodiment of the method according to the invention, a fingerprint is detected and evaluated as a biometric identification image.

I. Preparations

First, preparations are made that require user input:
The relative size of the target groups within the total size of the group of persons is defined by first defining a number of n desired target probabilities W _{soll *, i} and then each one at the beginning of the method by a user input in the used data processing system Target probability W _{soll *, i is assigned} a percentage distribution value which corresponds to the desired relative size of the respective target group Z _i within the total group Z _1-n .

The user is also prompted to enter a tolerance value ΔW or to confirm a suggested tolerance value ΔW. The tolerance value indicates .DELTA.W percentage or in percentage points, how much _is the respective target group probability _{W, i to} those of _W distribution may deviate _i, which is to be expected based on associated base patterns or individual characteristics.

Subsequently, in a separate iterative process, a connection is established between the desired target group probabilities W _{soll *, i} and the features which can be derived by machine from the fingerprint. These are, on the one hand, the five basic types and, on the other hand, the individual characteristics, the so-called minutiae.

• – rechter Bogen (right arch), der mit einer Wahrscheinlichkeit W_pattern,1 von etwa 31,7% auftritt;
• – linker Bogen (left arch) mit W_pattern,2 = 33,8%;
• – Wirbel (whor1) mit W_pattern,3 = 27,9%;
• – Flacher Bogen (arch) mit W_pattern,4 = 3,7% und
• – Gewölbter Bogen (tented arch) mit W_pattern,s = 2,9%.

The basic types known from dactyloscopy according to the HENRY classification are (cf. A. Senior: "A HIDDEN MARKOV FINGERPRINT CLASSIFIER" in: "In proceedings of the 31st Asilomar conference on Signals", Systems and Computers, 1997, pp. 306-310 .):

• Right arch, which occurs with a probability W _{pattern, 1} of about 31.7%;
• - left arch with W _{pattern, 2} = 33.8%;
• - vortex (whor1) with W _{pattern, 3} = 27.9%;
• - Flat arch (arch) with W _{pattern, 4} = 3.7% and
• - Arched arch (tented arch) with W _{pattern, s} = 2.9%.

A sixth basic pattern type W _{pattern, 6} can be defined in the method according to the invention in order to catch noncategorizable fingerprints.

If more than n = 6 target groups desired, then the method in its simple form can not be performed, but it must be made additional differentiations, which are described in further advantageous embodiments of the method according to the invention in the section "main program type 3".

If the user selects six target groups Z _1-6 with a desired target group probability, which corresponds approximately to the occurrence probabilities listed above of the basic patterns W _{pattern, i} , then the assignment can be determined simply by applying the basic patterns W _{pattern, i} to the Target groups Z _{i are} mapped.

If the probabilities W _{soll, i} theoretically to be expected on the basis of the known basic pattern distribution lie in the tolerance field of the desired target group distributions W _{Soll *, i} , then the pre-process is completed and the main program can be started.

The tolerance threshold with which the possible distributions W _{soll, i} deviate from the user-desired distributions W _{soll *, i} can be as large as the previously defined tolerance values that were decisive when setting up the group assignments. They can also be defined differently.

If the tolerance fields can not be reached in this way, first of all combinations of the basic pattern probabilities W _{pattern, 1-n are} run through in the data processing device and it is checked whether the desired distributions W _{Soll *, 1-n} can be achieved with such combinations. The same applies if fewer target groups are defined than there are basic pattern types. Even then, it is automatically checked whether the desired W _target can be achieved with a summary of two or more basic pattern probabilities W _Pattern .

If this does not lead to success, additional individual features on the fingerprint are used in addition to the basic patterns. It is then the relevant criterion, due to which a fingerprint is assigned to a specific target group or not, formed by combining one or more basic patterns with the additional individual features. This special pre-process will be explained in more detail below.

II. Main Program Form 1

The main program within the procedure is explained by the example already given at the outset, where the exit group is a group of people visiting a point of sale and invited to participate in a raffle.

At the beginning of the procedure, the target group probabilities have been defined, which for the example of application means that the relative size of the winning classes is fixed.

For example, it may be determined to play a gain of the first class Z ₁ with a probability of 3%, a second class Z ₂ of 4%, a third class Z ₃ of 30%, etc., where in the rest group Z ₄ , the largest has relative size and includes the remaining 63%, consolation prizes are played or just no profits are spent. As a tolerance, for example, a deviation of ΔW = 10% based on the said value W _{target is} allowed.

The assignments defined in the preparation procedure are therefore: Target group Z _i W _{target, i} (tolerance field) Assigned basic pattern W _pattern , Z ₁ 3% (2.7 ... 3.3%) arched bow 2.9%. Z ₂ 4% (3.6% ... 4.4%) flat arch 3.7% Z ₃ 30% (27% ... 33%) whirl 27.9% Z ₄ 63% (56.7 ... 69.3%) right arch 31.7% > = 65.5% left arch 33.8% (not categorizable) ?

In the in 2 sketched main procedure of the method in its simplest embodiment, the participants are then invited to place their finger on a fingerprint reader. In the fingerprint reader, the fingerprint is optically detected, then analyzed in a module within the fingerprint reader or additional computer, and assigned to one of the five basic types of fingerprint patterns or the fall-out class.

In a subsequent processing step, the person is assigned to one of the target groups due to the classification of their identification image into one of the basic pattern groups W _pattern .

The basic patterns "right arc" and "left arc" are combined with the non-categorizable ones, ie linked with a logical OR function. The group Z ₄ is thus assigned all recorded fingerprints that have the basic pattern "right arc" or "left arc" or that should eventually be unreadable.

Subsequently, the action belonging to the target group is carried out, for example a corresponding prize voucher is automatically issued by the data processing system. Subsequently, the procedure can be repeated as often as desired with other persons.

The 3a and 3b represent the preferred method according to the invention 3a shown first part corresponds to the duplicate check the previously based on 2 described procedure. This is followed by a branch at which it is checked whether a number of passes has already been made and thus a correction check should be carried out. If not, the run will follow as in the procedure 2 also finished.

If a correction test is required, then immediately in 3b 2, in which a computer-controlled intervention, referred to here as delta correction, is implemented in order to reach the target group distributions W _{soll, i} .

First, the degree of deviation is determined. The actual size W _{ist, i of} the target group Z _i into which the fingerprint would actually have to be classified after evaluation of the basic pattern and / or its individual features is compared with the predefined size W _{soll, i} for precisely this target group.

If it is determined that the amount of allowable deviation ΔW is exceeded, then the data processing device automatically intervenes in the process by overwriting the original target group assignment that would apply based on the detected master pattern and / or the detected individual features.

For this purpose, the deviations of all actual occurrence probabilities W _{ist, i} measured in the previous method are calculated for the respective target values W _{soll, i} for the target groups.

Then, that target group is preferably selected, in which the measured deviation is greatest in magnitude and which is determined as a negative deviation. This practically means that this target audience is smaller than it should actually be. By actively feeding participants into this under-represented target group, it is virtually filled up until the actual and setpoint values match and the permissible tolerance is maintained again.

The previous description of the method according to the invention presented cases in which the desired sizes of the target groups W _should either coincide within the allowable tolerance with the basic pattern distributions W _pattern or where by an iterative to be determined combination of different basic pattern classes approach to the desired size distribution was achieved.

However, if the sizes of the target groups are selected by the user in such a way that no assignment can be achieved by evaluating the basic patterns alone, or if more target groups are defined than there are basic pattern classes at all, then it is necessary to use the individual features which have additionally appeared on the fingerprint ,

These individual traits, which are referred to as minutiae in dactyloscopy, are, for example, certain characteristic branching points where ridge lines or bifurcate or where inclusions are present in a papillary line.

The fact of occurrence of one or more minutiae already constitutes an additional distinguishing feature per se. In addition, the topographical location of the minutia can be used in the overall image of the fingerprint or, at the location of minutiae, a tangent can be applied to the papillary line and its angle relative to a fixed coordinate system can be determined.

Using the additionally defined individual characteristics, an additional criterion is introduced in order to assign recorded fingerprints to a specific target group. Preferably, it is placed on the basic pattern. Only when with a combination of basic patterns on the one hand and individual characteristics on the other hand, no criteria can be defined to achieve the desired target group sizes can also be passed in the context of the process, the assignment solely on the basis of the frequency, the location or the angular orientation of the respective individual characteristics and define papillary lines.

4 shows three columns 410 . 420 . 430 each with 100%. The blocks 421 ... 425 in the left column 410 correspond to the naturally predetermined relative distribution W _pattern among the five basic patterns of fingerprints. The fallback class for uncategorizable impressions, sometimes added as the sixth class, is not shown in this case.

The right pillar 430 Contains the target group distribution desired by the user W _{target *} for five target groups 1-5

The middle pillar 420 represents the finally calculated by the computer feasible target group distribution W _Soll . The small deviations, which in 4 between the blocks to the bottom 415 . 435 as well as with the small blocks on the top 411 . 431 and 412 . 432 can be seen either lie within the tolerance range specified by the user, or the user is asked in this regard to either accept these small deviations despite exceeding its specified tolerance or to have the definitions generated again.

If the so-called delta correction method is used, then the correction of the remaining deviations from the desired idea can also be corrected later in the course of the method.

In the example shown, the largest deviation between the natural probability of occurrence of a basic pattern of class no. 3 and the user-desired size of the associated target group no. 3 occurs in the middle blocks 413 . 423 and 414 . 424 on.

To make up for this, the block is based on the algorithm underlying the method in this preferred embodiment 414 as subset according to block 424b completely assigned to the target group no. In addition, the block 413 so split that one half 413b still with the block 414 is linked so that a subset corresponding to the block 424a the target group no. 4 is added and finally the fourth target group one the block 424 in the middle column 420 corresponding size possesses.

The one around the lower half 413b correspondingly shortened residual group 413a Finally, this corresponds to the target group No. 3, which in the diagram after 4 as a block 423 is shown.

The representation in the form of blocks is only an illustration here. Although the distributions can also be visualized in this way in the context of the interaction with the user, this is not absolutely necessary for carrying out the method. The actual division of the groups takes place in the data processing system purely mathematically, but not geometrically.

To the in 4 Illustrated division of a pattern pattern (pattern) group into subgroups, a second, individual feature is used as additional condition when this pattern group occurs, which corresponds to a logical AND operation.

In the present example, the class of the basic pattern No. 3 was split in half. For the additionally introduced condition, therefore, a distribution of 50:50 must be given for the fulfilled additional feature or the non-fulfilled characteristic. For this purpose, for example, the number (even / odd) of the features can be used.

Example:

• a) entweder der Grundmuster-Klasse Nr. 4 zuzuordnen sein ODER
• b) er ist der Grundmuster-Klasse Nr. 3 zuordnen, jedoch in diesem Fall nur dann, wenn er zusätzlich auch (logisches UND) eine gerade Anzahl von Verzweigungen der Papillarlinien aufweist.

In order for the captured fingerprint to be assigned to the target group 4, he must

• a) be assigned to either basic pattern class no. 4 OR
• b) it is assigned to the basic pattern class no. 3, but in this case only if it additionally has (even logical AND) an even number of ramifications of the papillary lines.

If the detected fingerprint corresponds to the basic pattern class No. 3 AND at the same time has an odd number of ramifications of the papillary lines, then it is to be assigned to the target group No. 3.

The count of features and the even / odd number discrimination is the easiest way to add an individual feature to split a basic pattern class in half.

If divisions other than 50:50 are required, you must resort to screenings. For this purpose, the detected identification image can be divided into several sectors, which are based either on a Cartesian coordinate system, that is rectangular, or on a polar coordinate system.

5 shows the image of a fingerprint. The center of the coordinate system was automatically placed by the recognition software on a characteristic point of the fingerprint.

The absolute position and angular orientation of the fingerprint in the coordinate system is irrelevant to the present method, since only a random selection based on the characteristics of the fingerprint is to be taken and a recognition of the fingerprint is not essential, unless a duplicate check is desired.

Four quadrants are formed by the superimposed coordinate system, so that the occurrence of certain features in one of the quadrants has a theoretical probability of 25%. As a result, a basic pattern class in the ratio 75:25 or even 50:50 can be divided.

Furthermore, in 5 a tangent to the characteristic point created, z. B. the point of greatest curvature. Their angle α in the coordinate system can also be used as a differentiating aid by forming correspondingly finely divided angle groups.

If, for example, one wants to achieve a division of 1/6 to 5/6 or 1/3 to 2/3 from this additional feature, then the full circle can be subdivided into six sectors of 60 ° or 3 sectors of 120 ° each. To make the assignment, it is then checked in which sector the tangent falls.

II. Main Program Form 2

• 1.) Prüfung auf Wahl der Größen W_soll,i der Zielgruppen Z_i ähnlich den natürlichen Häufigkeiten des Auftretens der Grundmuster;
• 2.) Automatische Prüfung der Möglichkeit der Zusammenfassung mehrerer Grundmusterklassen W_pattern,i und Zuordnung zu einer Zielgruppe Z_i zur Erreichung der gewünschten Größenverteilungen W_soll,i.
• 3.) Automatische Hinzuziehung individueller charakteristische Merkmale zur Aufteilung einzelner oder mehrerer Grundmusterklassen und Kombination der aufgeteilten Grundmusterklassen mit anderen Grundmustern; dabei Rückgriff auf Anzahl, Lage und Ausrichtung bestimmter charakteristischer Merkmale, Bildung eines geeigneten Rasters.

The previously presented process variants were:

• 1.) Checking on the choice of the quantities W _{soll, i of} the target groups Z _i similar to the natural frequencies of occurrence of the basic patterns;
• 2.) Automatic examination of the possibility of combining several basic pattern classes W _{pattern, i} and assignment to a target group Z _i to achieve the desired size distributions W _{soll, i} .
• 3.) Automatic inclusion of individual characteristic features for the division of single or several basic sample classes and combination of the divided basic pattern classes with other basic patterns; recourse to number, location and orientation of certain characteristic features, formation of a suitable grid.

With these methods described above, very precise target group distributions can already be achieved.

A certain disadvantage of the previously described methods can also be seen in the fact that not every basic pattern of a fingerprint ever has the possibility of being able to be assigned to specific target groups when the basic patterns are split in blocks.

If only one person's participation in the allocation procedure is envisaged, this disadvantage will not be significant. If, however, persons are allowed multiple participation, then these persons will automatically be assigned to the same or very similar target groups or profit classes, which are just linked to the basic pattern of their fingerprint.

Since - in the case of the process application in a raffle - just the naturally very understaffed basic pattern classes are suitable for allocation to the upper profit margins, this would mean that people whose fingerprints correspond to a rare basic pattern, always the main profits, whereas the rest of those who have one of the basic patterns that averaged about 30%, then would always be grouped only in the lower ranks or completely out of the profit ranks.

To counteract this disadvantage, a further variant of the method according to the invention is described below, in which a screening or segmentation is again carried out. But not only with the goal to be able to define the desired target group sizes, but also to be able to solve sub-areas from each basic pattern type and to assign subsets of a specific target group from each basic pattern type.

• – In der Vorprozedur wird jedem Grundmuster gemäß der erwarteten Wahrscheinlichkeit des Auftretens eine Menge von Ganzzahlen zugeordnet (z. B. Fingerabdruck des Types „Schleife” = 1–28, „Wirbel” = 29–45, usw.), deren Summe 100 ergibt.
• – Um jedem Grundmuster die Zugriffsmöglichkeit auf alle Gewinne zu erlauben werden die Grundmusterblöcke in einprozentige Teile zerlegt und dann zufällig den Gewinnen zugeordnet, so dass z. B Gewinn der Kategorie 1 mit 30% Gewinnchance aus sechs Werten besteht, die dem Grundmuster „Schleife” zugeordnet sind, plus 5 Werten aus „Wirbel” etc. Dazu muss der Gewinn allerdings mit mehr als 6% definiert sein wegen der sechs Eingabe-Grundmuster, ansonsten ist dies nicht möglich.
• – In der Hauptprozedur wird ein Fingerabdruck analysiert und dessen Grundmuster samt den ermittelten individuellen Merkmalen als Eingabeparameter an die Gewinn ermittelnde Prozedur übergeben.
• – Es werden nun die Primfaktoren des dem Grundmuster zugeordneten Werts errechnet (z. B. für „Schleife”: 28 = 2·2·7) und gemäß der Primfaktoren eine weitere Zerlegung vorgenommen.
• – Wenn der kleinste Primfaktor 2 ist, so wird die Anzahl der Merkmale als Entscheidungsgrundlage genutzt:
• • Ist die Anzahl gerade, so wird die untere Hälfte der Eingabemenge zurückgegeben, ansonsten die obere (1–14 bzw. 15–28).
• • Bei jedem nachfolgenden Durchgang wird der Fingerabdruck in zwei gleichgroße Segmente unterteilt bzw. geviertelt, geachtelt etc. Oder es wird ein Segment zufällig ausgewählt und es wird dort die gerade/ungerade Anzahl der Merkmale analog zum ersten Durchgang als Kriterium herangezogen.
• – Sollte der kleinste Primfaktor ungleich 2 sein, so werden die Winkelausrichtungen der Merkmale betrachtet:
• • Der Vollkreis wird in eine Anzahl von Segmenten (= Faktor) entsprechend dem Primfaktor unterteilt und das Segment zurückgeliefert, das die meisten Merkmale zugeordnet hat. Die Zuordnung eines Merkmals zu einem Segment geschieht dann, wenn der Winkel des Merkmals in dem des Segments liegt (z. B. Primfaktor 7: 360'/7 = 51,42° pro Segment, also 0–51,42°, 52,43°–102,85°, ...)
• • Sollte diese Berechnung keinen eindeutigen Rückgabewert haben (2 oder mehr Segmente mit der selben Anzahl von Merkmalen) so wird dasjenige Segment ermittelt, dessen durchschnittlicher Merkmalswinkel am nächsten zum mittleren Winkel des Segments liegt (z. B. Segment 1: mittlerer Winkel = 25,71°, Segment 2: mittlerer Winkel = 42,01°, ...)

This procedure is performed as follows:

• In the pre-procedure, a set of integers is assigned to each master pattern according to the expected probability of occurrence (eg fingerprint of the type "loop" = 1-28, "swirl" = 29-45, etc.), the sum of which equals 100 ,
• - To allow each basic pattern access to all profits, the basic pattern blocks are divided into one-percent and then randomly assigned to the profits, so that z. B win of category 1 with 30% chance of winning consists of six values assigned to the basic pattern "loop", plus 5 values of "vortex" etc. However, the profit must be defined with more than 6% because of the six input basic patterns otherwise this is not possible.
• In the main procedure, a fingerprint is analyzed and its basic pattern together with the determined individual features are passed as input parameters to the profit-determining procedure.
• - Now the prime factors of the value assigned to the basic pattern are calculated (eg for "loop": 28 = 2 · 2 · 7) and a further decomposition is carried out according to the prime factors.
• - If the smallest prime factor is 2, then the number of characteristics is used as the basis for the decision:
• • If the number is even, the lower half of the input quantity is returned, otherwise the upper half (1-14 or 15-28).
• • For each subsequent pass, the fingerprint is divided or quartered into equal segments, equalized, etc. Or, a segment is randomly selected and the odd / even number of features is used as a criterion, analogous to the first pass.
• - If the smallest prime factor is not equal to 2, the angular orientations of the features are considered:
• • The full circle is divided into a number of segments (= factor) according to the prime factor and returns the segment that has assigned the most characteristics. The assignment of a feature to a segment occurs when the angle of the feature lies in that of the segment (eg prime factor 7: 360 '/ 7 = 51.42 ° per segment, ie 0-51.42 °, 52, 43 ° -102.85 °, ...)
• • If this calculation does not have a unique return value (2 or more segments with the same number of characteristics), the segment is determined whose average feature angle is closest to the middle angle of the segment (eg segment 1: mean angle = 25, 71 °, segment 2: mean angle = 42,01 °, ...)

In this way, a value is determined from the set (eg type "loop": 1-28, eg result 21). This now determines the assigned profit.

If the delta correction is not necessary, ie if no profit (no target group) has exceeded a deviation value ΔW (eg 10%), the profit is played out, otherwise the profit which has the largest negative deviation is played again , so too rarely played out.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited non-patent literature

• A. Senior: "A HIDDEN MARKOV FINGERPRINT CLASSIFIER" in: "In proceedings of the 31st Asilomar conference on Signals", Systems and Computers, 1997, pp. 306-310 [0074]

Claims (10)

Method for dividing a group of persons into a plurality of target groups Z _1-n , wherein the total size of the group can be indefinite and wherein the relative size W _{soll, i of} the target groups Z _{i is} predeterminable, comprising the following steps: a) input of a desired relative size W _{soll *, i of} each target group Z _i in a set of n target groups Z _1-n in a data processing device; b) selecting a group of basic patterns which can be derived from an identification image detectable by the persons by means of at least one electronic reading device and determining a standard distribution value W _{pattern, 1-n} for the number and / or the topographic position and / or orientation of the respective basic patterns in the identification image from a first data memory; c) linking at least one basic pattern from a group of n basic patterns, each with a target group Z _i, by means of an iterative optimization method running in the data processing system, in which i. Configurations of assignments of individual basic patterns to individual target groups and / or combinations of several basic patterns to a target group are formed; ii. in each case a relative nominal distribution W _{soll, 1-n} for each target group Z _{1-n is} calculated from the standard distribution values W _{pattern, 1-n of} the basic pattern retrieved from the first data memory; iii. the deviations of the calculated target distributions W _{soll, 1-n} from the user-desired desired distributions W _{soll *, 1-n are} calculated; iv. the one of configurations of assignments of the basic patterns to the target groups is selected, in which the calculated deviations are smallest; d) detecting a biometric identification image in the persons by means of an electronic reading device; Checking the identification image for correspondence with at least one of the basic patterns and associating the person with the target group Z _{i associated} with the respective basic pattern. A method according to claim 1, characterized in that after completion of the division of the group, the actually measured frequencies W _{pattern, i of} the occurrence of the respective basic pattern are written in the first data memory. A method according to claim 1 or 2, characterized in that during the current division of the group, the respective number of persons in the target groups Z _{i is} updated in a second data memory and wherein the deviations between the measured relative magnitudes is each target group W _{, i} and Target distribution variables W _{soll, i are} determined, wherein when falling below an allowable tolerance value .DELTA.W the person is assigned to the target group Z _i , which is linked to the basic pattern determined in her and wherein when a tolerance value .DELTA.W is exceeded, the person is assigned to that target group Z _j in which the largest negative deviation ΔW exists. Method according to one of Claims 1 to 3, characterized in that, in the case of a number n of target groups which is smaller than the number of possible basic patterns, the data processing means iteratively combine the variables W _{pattern, i} of a plurality of basic patterns in such a way that the data sets for the respective combinations of calculated distributions with the desired target group distributions W _{soll *, i are} compared and the combinations of basic patterns are selected and linked to the target groups, in which the deviations between the calculated distributions W _{soll, i} with the desired target group distributions W _{soll *, i} in the tolerance field. Method according to one of Claims 1 to 4, characterized in that with a number n of target groups which is smaller than the number of possible basic patterns, configurations of links of basic patterns with target groups are iteratively checked in the data processing device, in which the variables W _{pattern , i} are combined by a plurality of basic patterns, wherein the distributions calculated for the respective combinations are compared with the desired target group distributions W _{soll *, i} and wherein the combination is selected and assigned to a target group in which the deviations between the calculated distributions W _{soll, i} with the desired target group distributions W _{soll *, i} are lowest. Method according to one of claims 1 to 5, characterized in that in the identification image, the number and / or the topographic position in a grid and / or orientation or at least one additional feature is determined. A method according to claim 6, characterized in that a combination of a basic pattern and at least one additional feature formed and the combination with one of the target groups Z _{i is} linked. Method according to one of claims 1 to 7, characterized in that a fingerprint is used as the identification image and the electronic reading device is a fingerprint scanner Method according to one of claims 1 to 8, characterized in that the basic patterns correspond to the five basic types of fingerprints for which a natural basic pattern distribution W _{pattern.1-5 is} given. A method according to claim 6 to 9, characterized in that in the identification image, a grid is placed, wherein the number of grid units at least the desired number n of target groups Z _1-n corresponds and wherein each target group Z _i with the number and / or orientation at least one basic feature and / or an additional feature in at least one zuegordneten grid unit is linked.

Images