WO2011082793A1

WO2011082793A1 - Method for testing value documents

Info

Publication number: WO2011082793A1
Application number: PCT/EP2010/007688
Authority: WO
Inventors: Shanchuan Su; Dieter Stein
Original assignee: Giesecke & Devrient Gmbh
Priority date: 2009-12-16
Filing date: 2010-12-15
Publication date: 2011-07-14
Also published as: DE102009058439A1

Abstract

The invention relates to a method for testing value documents of the same specified type. Characteristic data is ascertained for one of the value documents, said characteristic data describing specified characteristics of the value document. A class is ascertained for the value document from a number n of classes which are specified for value documents of the specified type, wherein n is greater than 1. A transformation of at least some of the characteristic data or a transformation of the processing data ascertained from the characteristic data is carried out dependent on the ascertained class in order to at least partially compensate for a dependence of the characteristic or processing data on the ascertained class. In order to test the value document, at least one specified criterion which is preferably independent of the classes and which relates to the resulting characteristic data or processing data is tested, and a signal that describes the result of the test is formed.

Description

Method for checking value documents

The present invention relates to a method for checking value documents and a method for providing data for the examination method.

Under value documents are understood card and preferably sheet-shaped objects that represent, for example, a monetary value or an authorization and therefore should not be arbitrarily produced by unauthorized persons. They therefore have features which are not easy to manufacture, in particular to be copied, whose presence is an indication of the authenticity, i. the manufacture by an authorized agency. Important examples of such value documents are chip cards, coupons, vouchers, checks and in particular banknotes.

Value documents must be checked to see if they are genuine. As falsifications of value documents become better and better over time, it is necessary to constantly improve the examination of the authenticity of value documents.

Further, banknotes can be so badly damaged or soiled by heavy and / or careless use that they are no longer usable. This condition of banknotes is often called "unfit for circulation." Banknotes are therefore also checked for their condition in order to withdraw banknotes that are no longer fit for circulation and, if necessary, replace them with new ones. Here, too, the demands on the quality of the condition test increase.

The present invention is therefore based on the object of specifying a method for checking value documents, which allows a precise examination of value documents. The object is achieved by a method for checking value documents of the same predetermined type, in which for one of the value documents characteristic data are described which describe predefined properties of the value document, a class for the value document of a number n of classes specified for value documents of the specified type is determined, where n is greater than 1, depending on the determined class, a transformation of at least some of the property data or processing data determined from the property data is carried out to at least partially match a dependency of the property or processing data on the determined class and, for checking the value document, at least one predetermined criterion, preferably independent of the classes, relating to the resulting transformed property data or processing data is checked and a signal describing the result of the test is checked is formed.

Value documents of the same predetermined type are understood to mean value documents which are considered to be of the same type in relation to the criterion used for the test. Value documents of the same specified type are, in particular, value documents which have predefined properties which must lie within tolerances specified for the respective type. In principle, they can have other properties in which they differ more. For example, banknotes of a same given type may have predetermined picture elements that are characteristic of that type. If features determined only from the recorded property data are used for the check, for example for the echo-checking, then all the value documents that contain them can be used Have features in a given tolerance range, to be regarded as the same type.

In a preferred embodiment of the method, the value documents of the same predetermined type can be banknotes of the same currency and denomination. Depending on the type of currency, the predetermined type can be additionally given by the fact that the banknotes in substantially predetermined areas, ie. H. only in one, for example, by a central bank, given range of permissible fluctuations also varia- ble, have properties. For example, the pressure on the banknotes can be the same except for predetermined areas.

As property data, in principle any data can be used which describe predetermined properties, in particular predetermined physical properties, of at least parts of a value document to be checked. This can be raw data from sensor elements or from these calculated data.

It is now provided for the value documents of this same type in each case to perform a classification, ie an assignment to one of a number n of classes, in particular to one of at least two predetermined classes, whereby value documents of the same type are distinguished. Depending on the class determined, a transformation of at least some, or a part of the property data or of processing data determined therefrom is then carried out in a compensation step in order to at least partially compensate for a dependence of the property or processing data on the determined class. To it can be provided that, for n-1 of the classes, transformations are performed which are specified for the respective class, while no transformation takes place for the remaining class of the n classes; or it can be provided that transformations are carried out for all classes which are specified for the respective class. The transformations are respectively specified for the classes, which can be done, for example, by specifying procedural steps for carrying out the transformation and values for parameters used in the method steps; For example, a Tramformationsfunktion be provided by means of which the property or processing data can be converted depending on the parameters in transformed property or processing data.

This approach is based on the idea that there can be systematic differences for value documents of the same given type, which on the one hand enable a division into classes and which on the other hand can express themselves in the property data. A dependency of the property or processing data on the determined class is therefore to be understood as meaning in particular that the value documents of this class have property or processing data which systematically differs from those of value documents of other classes, but of the same type.

The property or processing data present after the compensation step are then used to test the value document, in which the at least one predetermined criterion, which is preferably independent of the classes, is examined, which relates to the resultant tiansformed property data or processing data. In particular, reference data and allowable deviations from the Enter reference descriptive parameters that are specified for the value documents of the given type. By compensating the class-dependent influences, it is possible to keep the permissible deviations small, which results in a precise examination. In addition, there is the advantage that only one set of reference data and permissible deviations describing parameters are necessary.

In principle any data can be used to determine the class. For example, you can use data that describes other properties of the value document than the property data. Thus, magnetic properties could be used to determine the class and optical properties to test the value document. Preferably, however, the determined property data are used to determine the class. This has the advantage that it is just the differences in those property data that can be used for the classification, which are compensated for later testing.

Furthermore, characteristic vectors can be determined from the property data when determining the class, and preferably the feature vectors can be used as processing data. Preferably, these feature vectors are also used in the later test

The classes can be predefined in such a way that they reflect possible systematic deviations of the property data, that is property data of value documents of one of the classes systematically differ from the property data of value documents of the other classes. Preferably, the classes are predetermined in such a way that value documents are assigned to each of them that are similar in terms of at least one aspect of the production of the value documents. This can be done by the Classification production-related systematic differences in the given properties and thus property data of the value documents are recorded. In principle, the aspect of the production can be chosen arbitrarily, but suitably. According to a first preferred embodiment, as aspect of the production, only the respective manufacturer of the value documents can be used and preferably each of the classes can be assigned only value documents produced by a single manufacturer. Depending on the type of value document, the manufacturer can be determined by checking at least the property data and / or in particular the serial number in the case of the EURO.

According to a second preferred embodiment, production batches in which value documents have been produced can be used as an aspect of the production. A batch is understood to mean value documents that are produced during the same production process. Value documents from different batches can, for example, be distinguished by differences in the materials used therein, in particular, for example, the paper or printing inks, which are also produced in batches and not in continuous processes, and / or the means used in the production, for example the printing plates, and / or differentiate the manufacturing process parameters used in the manufacture. The exact reasons for the occurrence of differences play no role in the context of the present invention. A class only needs to comply with at least one production batch; however, value documents of the same type and of at least two different batches of a single class can also be assigned. Method for specifying the classes and for determining a method for determining one of the classes for a value document, also referred to as classifying a value document in the context of the present invention, and transformations for the classes can in principle be arbitrarily determined.

Preferably, a classifier can be used to determine the classes or classification, and a transformation can be used as a transformation, which was determined by the following method, which is likewise the subject of the present invention. In this method for determining data for the performance of the method described here for checking value documents, preferably in the embodiment in which production batches are used as an aspect of the production in which value documents have been produced, value documents of the same predetermined type and from different Production batches provided for the value documents by means of a sensor property data that describe predetermined properties of the value documents, and determined by means of a Datenverarbeihmgseinrichtung from the property data values of predetermined feature vectors. Depending on the values of the feature vectors, the data processing device defines classes to which respective value documents of the given type are regarded as being similar in relation to the values of the feature vectors according to the predetermined criterion, and at least one class is generated by the data processing device. sifikator is determined, by means of which a value document of the given type as a function of this determined property data assigned to one of the classes and thus for the value document a class can be determined. For at least one of the defined classes, a transformation is carried out for at least one of the defined classes by means of the data processing device. niger of the property data or determined from the property data processing data determined by the dependence of the property or. Processing data from the determined class is at least partially compensated. The classes determined in this way can be used as the given classes for the procedure for checking value documents of the same type, also referred to below as test methods. Among the data for carrying out the test method, data describing the classifier, ie the at least one classification function and its parameters, also referred to as classifier parameters in the context of the invention, and the data of the transformation, in particular a transformation function and its parameters, are understood.

According to the described method for checking value documents, the value documents of the same predetermined type as above are also determined in the method for determining data for carrying out the test method described above, and in particular preferably banknotes of the same currency and denomination.

Preferably, to determine the classes, component analysis may be used in a space of feature vectors determined from the property data. The dependence of the feature vectors on the property data is the same as in the test method, if feature vectors are used in this case. This method has the advantage of a very simple implementation. Preferably, a Karnene-Loeve transformation or a principal component analysis (PCA) can be used for this purpose.

Also, to set the classes, a discriminant analysis in a space of feature vectors obtained from the property data or a subspace of the feature vectors is used. The advantage of this approach is that a very good separation of the classes can be achieved. In principle, one of the predetermined classes can be determined as desired. Preferably, however, a linear classifier is used as the classifier. For such a classifier, a classification function dependent on the property or feature vectors may be used with a functional form given by the linearity with respect to the feature vectors, for which in turn corresponding parameters defining the classifier are determined. Such classifiers not only offer the advantage that they are easy to determine, but can also be evaluated quickly during the test. As already explained at the outset, it is possible, in particular if the number of determined classes is a number n, for each at least n-1 of the classes to determine a transformation of the property data determined for value documents of the given type or processing data determined from the property data Depending on the property or processing data is compensated by the determined class.

In both methods, this transformation can only be determined for some of the property data and / or only for processing data, in particular feature vectors, and can also be carried out only for these.

However, it is also possible that the transformation is determined so that it can also be used all property data and / or processing data used for subsequent tests. In both methods, the property data preferably describe spatially resolved optical properties of the value document, particularly preferably an image in reflection or transmission in different spectral ranges.

Particularly preferably, the property data describes a color image of the value document.

If feature vectors are used, they preferably describe color properties of given evaluation areas of the image. When determining the classes, the color properties are then preferably used, which does not mean the recognition of special shapes, for example letters or numbers.

In preferred embodiments of the method, the transformation may then comprise a transformation in a color space.

The subject matter of the present invention is furthermore a device for processing value documents with at least one input pocket for

Recording of value documents, a transport device for transporting isolated value documents to at least one output tray, a sensor device arranged on a transport path given by the transport device with an evaluation device for carrying out the test method, in particular the determination of the property data, and a control device for controlling the transport device in response to signals the evaluation device. Preferably, the device may have at least two output compartments and the control device for controlling the transport device in dependence on the Signa len of the evaluation be formed so that depending on the signals for tested value documents, the value documents in one of the compartments, which is determined by the respective signal transported. The device, preferably the evaluation device, preferably has a memory in which all data necessary for the execution of the method are stored.

The invention will be explained in more detail below by way of example with reference to the drawings. Show it:

1 shows a schematic illustration of a bank note processing device, FIGS. 2a and b show a schematic representation of an optical sensor of the bank note processing device in FIG. 1 in the transport direction and from above onto a transport plane,

3 shows a schematic representation of an example of a value document to be examined in the form of a banknote,

4 is a simplified flow chart for a method for checking value documents that can be performed in the routine handling device in FIG. 1, and a simplified flow chart for a method for determining data for the method for checking value documents in FIG. 4. 1 shows a device 10 for processing value documents, in the example a banknote processing device, which serves inter alia for determining the state and the authenticity of value documents 12 in the form of banknotes. The apparatus 10 has an input tray 14 for input of value documents 12 to be processed, a separator 16, which can access value documents 12 in the input tray 14, a transport device 18 with points 20 and 20 'arranged successively along a transport path 22, and after Each of the points or an end of the transport path 22 following the two points respectively has an output pocket 26 or 26 'or 26 "along the transport path 22 given by the transport device 18 before the point 20 and after the singling 16 a sensor arrangement 24 A control device 30 is connected at least to the sensor arrangement 24 and the switches 20 and 20 'via signal connections and serves for the evaluation of sensor signals of the sensor arrangement 24 and control of at least the switches 20 u nd 20 'depending on the result of the evaluation of the sensor signals.

The sensor arrangement 24 comprises at least one sensor for this purpose; In this embodiment, only one optical sensor 32 is provided for the spatially resolved detection of color properties, which detects the optical document remitted from the value document. In other embodiments, still other sensors, e.g. be provided for other than optical properties.

During the advance of a value document, the sensor 32 captures an image of the value document in three spectral regions corresponding to the three Color channels red, green and blue, which is represented by corresponding sensor signals.

From the analog and / or digital sensor signals of the sensor 32, characteristic data are determined by the control device 30 during a sensor signal evaluation, which data are relevant for the checking of the bank notes with respect to their condition and / or authenticity. For this purpose, the control device 30 has an evaluation device 31, which is integrated in the example in the control device 30, but in other embodiments, but also part of the sensor assembly 24, preferably the sensor 32 may be.

In the present example, the state of a value document is regarded in particular as the state of fitness or fitness for circulation, ie the suitability to continue to be used as a means of payment. To assess the condition or authenticity, predetermined criteria are used. These relate in particular to the presence of soiling and / or color abrasion or stains as well as the presence of cracks, dog-ears and / or holes, and / or the absence of components of the value documents used in this example. In addition to a corresponding interface for the sensor 32, the control device 30 has a processor 34 and a memory 36 connected to the processor 34 in which at least one computer program with program code is stored, in the execution of which the processor 34 in a first function as the evaluation device 31 Sensor signals, in particular for determining the authenticity and / or a total state of a tested value document, evaluates and controls the device in a second function or corresponding to the evaluation, the transport device 18 controls. The control device 30 further has a data interface 37. The control device 30 can be operated in at least two operating modes. In a first operating mode, the working mode, it controls the device for checking value documents. In a second operating mode, the detection mode, it controls the device in such a way that value documents are singled out of the input compartment and occasionally transported past the sensor 32. Thereafter, the value documents are again output in a predetermined one of the output compartments 26, 26 'or 26 ", whereby the sensor 32 detects the properties of the value documents which are transported past it Control device 30 acquires property data acquired by sensor 32 and makes it available to each of the value documents via data interface 37. In the example, the data interface comprises means for describing data carriers, for example CDs, which can be read by a suitable data processing device.

In the working mode, the evaluation device 31, more precisely the processor 34 therein, after determining property data, can check a given criterion for the authenticity of the value document into which at least some of the detected properties and reference data are received.

Accordingly, the evaluation device 31 can check a criterion for the overall status of the value document into which the property data enter or depend on them. In particular, further reference data for determining a permissible state of the value document which are predefined and stored in the memory 36 can enter into the criterion. The overall state can be given for example by two categories "still fit for use" or "marketable" or "to be destroyed". Depending on the determined state and the authenticity, the control device 30, in particular the processor 34, controls the transport device 18, more precisely the switches, in such a way that the checked value document is transported for storage to corresponding output compartments in accordance with its determined overall state.

For processing documents of value 12 are separated into the input tray 14 as a stack or individually inserted value documents 12 of the Vereinzier 16 and isolated the transport means 18 which supplies the isolated value documents 12 of the sensor assembly 24 This detects optical properties of the documents of value 12, in the example, a color image , wherein sensor signals are formed, which reflect the corresponding properties of the value document. The control device 30 detects the sensor signals, determines in dependence thereon a state and the authenticity of the respective value document and, depending on the result, controls the points so that the examined value documents are fed to the output compartments in accordance with their ascertained state and authenticity.

The sensor 32 is formed as a line sensor which, during the advance of a value document past the sensor 32, comprises a sequence of line images which give an image of the value document in a direction transverse to the direction of the line. It comprises in the present example, in FIGS. 2a and 2b only extremely simplified schematic representation, a lighting device 38, in the example with two light sources, for illuminating at least a portion of a value document 12 with white light during its Vorwerkransport over its entire extent transverse to the transport direction T and as detection device 40 in the example three pointers steering cameras 42, 42 'and 42 "with not shown, arranged in the beam path in front of these red, green and blue filters for detecting red, green or blue portions of the remitted from the value document light of the Beleuchtungseir direction 38. Each of Line scan cameras each have a detector row with line-arrayed photodetection elements, in front of which the filter is arranged, which corresponds to the color component of the remitted light to be detected by the respective line scan camera.The sensor 32 can also comprise further optical elements, in particular for imaging or focusing The detector rows of photodetection elements are arranged parallel to one another, but a more detailed description of such a sensor, with a further row of IR radiation detectors, can be found in US Pat

WO 96/36021 AI of the applicant, the contents of which is incorporated by reference in the description.

To record a color image of a value document 12, it is transported past the sensor 32 at a constant speed, whereby intensity data are recorded at constant time intervals with the line scan cameras 42, 42 'and 42 "in an intensity-resolved manner in terms of color and resolution representing the properties of pixels 44 of a line image representing the line-shaped area of the value document 12 detected by the sensor 32. By juxtaposing the line images according to the chronological order of the detection, ie corresponding assignment of the image data, an image of the value document is then displayed. received with pixels.

An image captured by the sensor 32 is therefore composed of pixels arranged in a rectangular matrix. It is described by property data in the form of image data. In the illustration of the picture of a Value document 12 in FIG. 3 are shown for the sake of clarity only some of the pixels 44. Each of the pixels are assigned as color data n, g ,, bi for red, green and blue as property data next to a number i representing the position in the image. It is assumed here that the signal processing device 44, after calibration from detection signals of the detector lines 42, 42 'and 42 ", can generate and generate color values which can be used to a good approximation as color coordinates in the standardized CIE XYZ color space for ease of illustration to a vector V, given by the components (i, r ,, gi, bi) Ϊ = Ι, Ν where N is the number of pixels.

For checking a value document for the state or the authenticity, a criterion given for a respective type of value documents can be used for a sufficiently good state or the advance of a value document to be regarded as genuine, which depends on the above-mentioned vector. This can generally be formulated such that a respective test function K (P _j , V) is given, which depends on predetermined criteria parameters P _j (j = 1, .., m) and the vector. If the function assumes a given value for a given vector V, the respective criterion is considered fulfilled, otherwise not. An examination of the criterion can therefore be to calculate the value of the test function K for a given vector V and to compare it with a predetermined value G. If the value of K exceeds G, the criterion is met, otherwise not. The calculation of the value of the test function is understood to mean that the value is determined from the vector and the parameters by means of steps predetermined by the test function. To check the value documents, a program is stored in the memory 36 in a section serving as part of the evaluation device 31 and thus in this example in the control device 30, which executes the following steps of a method for checking value documents when executed by the processor 34. In particular, all the data necessary for carrying out the method are stored in the memory 36.

Using characteristic data of the sensor 32, the evaluation device 31 determines in a step S10 in a property data item for the value document to be checked, which describe predefined properties of the value documents. In this example, the sensor 32 acquires images of the value documents, more specifically the image or image data representing images, in the example full-area images with three color channels, namely red, green and blue (RGB channels); The type of image data has already been described above. The property data thus indicates properties of the value document as a function of the location on the value document. The property or image data are transmitted to the evaluation device 31 and detected by this. Depending on the type of sensor, a preprocessing of the detected data in the sensor 32 or the evaluation device 31 can be performed in this step, in which the image data are transformed, for example, to compensate for background noise.

The evaluation device 31 then determines the type of a value document to be checked in step S12 as a function of the property data detected by the sensor 32. In the present example, value documents are checked in the form of euro banknotes. The evaluation device 31 can therefore first perform a search or recognition of edges of the banknote in the picture. From the recognized borders, she can use the format of the value document and in the example of value documents in the form of Banknotes determine the denomination and thus the type from the set of the given possible value document types.

Then, in step S14, using the property data for the value document of the determined type to be checked, it obtains a class from a number n of classes given for value documents of the predetermined type, where n is a number greater than 1.

In a sub-step S16 of step S14, the evaluation device 31 positions predetermined regions of interest (ROI) in the image, which consist of the known position of predetermined regions on the value documents of the given type relative to the contours of the value documents and an outline determined in the image of the value document. For this purpose, the evaluation device may, in particular, first perform a search or detection of edges of the banknote in the image, or resort to results of step S12, in order then to position the ROI in the image, depending on the position of the edges in the image. to select appropriate property values. For example, twenty-four evaluation areas can be selected for a 10 Euro banknote. FIG. 3 shows, by way of illustration, an evaluation area 46 emphasized by hatching and comprising nine pixels 44.

In the present example, feature vectors are determined from the property data, which reproduce color properties of the evaluation areas; When determining the classes, the feature vectors and in particular these color properties are used, ie the classification takes place as a function of the color properties. In a sub-step S18 of step S14, the evaluation device 31 determines from the property data for each of the evaluation areas at least one value assigned to the respective evaluation area, which is referred to below as a feature value. In the example, three feature values are determined, a feature value for each of the color channels. For example, as the evaluation area 46 comprises nine adjacent pixels arranged in a given rectangle, the average value over each of the 10 pixel values of the color channel can be the property value for each of the color channels corresponding color channel can be calculated. With a number N of evaluation ranges, f * color channels result in N * f feature values, which will be referred to below as xi, i = 1,... N * f. The feature values for a banknote are referred to as the feature vector X of the banknote. Optionally, depending on its design, the value-value device 31 can carry out a transformation of the property values or of the feature vector into another color space, for example a device-independent color space, in particular a CIE-Lab color space. In a sub-step S20 of step S14, the actual classification of the value document takes place by means of a predetermined classifier. For the feature vector X or the feature values xi, i = 1,... N * f, the value of at least one classification function specified for the type of value documents, which depends on the feature vector, is determined and a class corresponding to the value is determined on the basis of the value , In the present example, a classifier is used, which is determined by the method described below for determining data for the present test method. In the step S22 following the step S14, the evaluation device 31 carries out a transformation of at least some of the property data or processing data determined from the property data as a function of the determined class. By means of this transformation, systematic differences between the property data for value documents assigned to different classes should be compensated to the extent that they do not play a significant role for further evaluation, in particular testing. In this exemplary embodiment, the property data, preferably all image data representing the image of the respective value document, are transformed. For this purpose, transformations assigned to the respective class, in the example transformer functions, are used, which are set so that the data transformed with the transformations are used for checking with the above-mentioned criteria for the condition and authenticity of the value documents of the given type independently of the determined class can. These criteria are independent of the classes, ie the same for all classes. In the present example, as will be described below, this transformation corresponds to a change in the color and / or brightness of the entire images or a corresponding transformation of the property or image data. The determination of the transformation or data for the transformation can be carried out with the method described below for determining data for the present test method.

In step S24, the evaluation device 31 checks the predetermined criterion relating to the transformed property data or processing data for checking the value document and forms a signal describing the result of the test. In this exemplary embodiment, the two aforementioned criteria are checked, one for the state of the value document and the other for the authenticity of the value document, and in each case formed speaking signals. Depending on the signals, the control device then controls the transport device.

To establish data for performing the described test method, in particular the classes and transforms, the following method can be used.

In step T10, value documents of the same type are provided, in the example banknotes of the same currency and denomination, but from different production batches.

In step T12, images of the value documents, more specifically the image representative property data are detected by the sensor 32. In this step, a sub-step corresponding to step S10 is performed for each of the predetermined value documents of the same type. In contrast to the step S10, however, the detected property data are transmitted via the data interface 37 to an external data processing device with a corresponding interface, which takes over the further processing of the data.

The following steps are performed by the external data processing device.

In step T14, the data processing device carries out the following positioning of predetermined evaluation ranges for each of the value documents and stores corresponding data for the respective value document. More precisely, the data processing device positions predetermined regions of interest (ROI), which correspond to the evaluation regions in step S16, in the image which is predefined from the known position. ner areas on the value documents of the given type relative to the outlines of the value documents and an ascertained in the image outline of the value document. For this purpose, the data processing device can, in particular, first carry out a search or recognition of edges of the banknote in the image in accordance with the procedure for edge recognition in step S12, and then, depending on the position of the edges in the image, the ROI in the image position, ie select appropriate property values. For example, for a 10-euro banknote 12, twenty-four evaluation areas can be selected. These are preferably selected so that they can also be used for testing for authenticity and / or for checking the state. In the example, one of the evaluation areas is area 46 in FIG. 2.

In step T16, the data processing device determines for each of the value documents from the property data for each of the evaluation areas at least one feature value assigned to the respective evaluation area. In the example, the determination of the feature values for each of the value documents takes place as in step S18. From the feature values, as in the test method, a feature vector X is formed for each of the value documents

In the optional step T18, depending on its design, the data processing device can carry out a transformation of the property values or of the feature vector into another color space, for example a device-independent color space, in particular a CIE-Lab color space. Such a transformation must then also be carried out in the test procedure. In the following step T20, the data processing device searches classes depending on the determined values of the feature vectors for the value documents, which are respectively associated with value documents relating to the property data or in the example of the processing data determined therefrom in the form of the values of the feature vectors according to a given criterion does not differ significantly from each other, ie it sets appropriate classes. In the example, the evaluation device determines clusters or clusters of the feature vectors in a corresponding space, which are separated from one another according to at least one predetermined criterion. Each of these clusters corresponds to a class. It has been shown that the classes can reproduce the origin of different production batches, if they differ sufficiently in their properties. In this case, a class or an accumulation banknotes can correspond to several production batches, but it is also possible that one class is assigned to only one production batch. In the example, a so-called "Principal Component Analysis" (PCA) or analysis based on the Karhunen Loeve transformation (KL transformation) are performed. By means of the PCA or KL transfommation, at least two, in the example, four main components are determined by means of which the largest portions of the scattering of the feature vectors can be represented. Clusters are then determined in the subspace formed by these main components. In the example, it is used as a criterion for the presence of two different clusters that the distance of the mean vectors for two different clusters must be greater than the sum of the covariances of the feature vectors of the clusters projected on the difference of the mean vectors.

In step T22, the data processing device sets at least one classifier for the accumulations or classes found, ie a classifier. tion function and classifier parameters for this, by means of which a value document of the specified type can be assigned to one of the classes. This at least one classifier, ie the corresponding classification function, is the classifier used in step S20, with the function used in step S20. For this purpose, the data processing device determines the classifier parameter values for the classifier or the predetermined classification function, by means of which a value document of the predetermined type, in the example of a banknote of the specified type, for the data corresponding to steps T12 to T16 or T18 is detected and returned to property values or Feature vectors have been processed, depending on the property values to one of the accumulations or one of the determined classes can be assigned. In particular, a reduction of the property values or feature vectors to a subspace in which the separable accumulations are easily recognizable can take place in this step. If a PCA or KL transformation is used, as classifier parameters, for example, for each of the classes, the mean vector of the accumulation and covariance data, in particular in the subspace, and threshold values for the maximum distance from the mean vector according to one, may be used, in addition to the principal components representing and subspace spanning the subspace determined by the covariance data distance measure, in the example, a Euclidean distance. For the classification in step S20, first the projection of the feature vector X on the subspace is determined. Then, for each of the classes, the distance of the projection from the mean vector of the class in the subspace is determined and compared with the maximum distance for the respective class. The class is assigned to the class for which the distance is smaller than the respective maximum value. If there are several such classes, the one is chosen where the distance is the smallest If no distance is determined for any of the classes that is less than the distance ximal distance for the class, the classification is aborted with an error message. In the example, n classes were determined.

In step T24, the data processing device determines transformation functions by means of which the image data or the feature values can be transformed to common reference values that can be used for a check using the predetermined criteria in step S24. In the present example, the transformation functions are functions of the property data and compensation parameters whose functional form, ie dependencies on the property data and compensation parameters, is specified. To determine the transformation functions, only the values of the compensation parameters are determined. The

In the present example, parameters can be used in particular to parametrize displacements and / or rotations and / or scalings in the color space by parameterizing the compensation parameters. In the example, for the sake of simplicity, the mean values of the cluster or class that originate from the largest number of batches are used as reference values. The transformation functions for the other determined classes, n-1 in number, are given on the one hand by a corresponding functional form and on the other hand by compensation parameters which include the difference vectors between the mean vectors for the respective clusters and the mean vector for the cluster Value documents of most batches are assigned. If only one batch is determined for all clusters or classes, the components of the cluster in which the smallest scatters were determined serve as reference values. The transformation or compensation parameter values are then stored and can be transferred to the evaluation device 31. In a variant of the first exemplary embodiment, in step T20, after determining a predetermined number of the most important main components, in the example four main components, and corresponding unit vectors in a subspace spanned by the main components, a linear discriminant analysis is performed on the basis of the projections of the feature vectors into the subspace. In particular, a Fisher discriminant analysis can be carried out. Depending on the available data, this can result in improved separation of clusters.

In the example, a linear classifier is used in step T22 and thus a function that is a linear function of the feature vectors

The linear classifier is given in the example for n classes by an n-dimensional vector B with threshold values, an N * fx n matrix A, the formula g (x) = Α ^Τ · Χ + Β. In this case, a class j (j = 1,..., N) is assigned the jth component of the vector B and the jth column of the matrix A. Each component of the vector g (X) therefore corresponds to one of the given classes. Further, the linear classifier is given by the rule that the feature vector X is assigned to the class whose component of the vector g (X) is larger than all the other components In the example, in particular the vector B and the matrix A are determined as the classifier parameters.

The following steps are unchanged.

Accordingly, the described linear classifier is used as a classifier in step S20

In another variant of the first embodiment, in step T20, as a criterion for defining the clusters, it may be used that the length of the difference between a feature vector of a respective cluster from the mean over the feature vectors of the cluster must be smaller than the smallest length of the differences between feature values different clusters. The other steps are unchanged.

Another embodiment differs from the first embodiment by the formation of the clusters in step T20 and the determination of the classifier. Here, with the feature vectors for the given value documents from different batches without a previous principal component analysis, a discriminant analysis, in the example a Fisher dispersion analysis, is carried out, in which the classes are determined such that a maximum distance of the clusters with minimal scattering in the clusters is reached. As a classification parameter, the parameters determined in the Fisher discriminant analysis, which correspond in particular to the matrix A and the vector B, are then determined in a step corresponding to step T22. In contrast, in the first variant of the first exemplary embodiment, the matrix A still contains portions that represent the projection in the subspace determined by the PCA or KL transformation. In other embodiments, alternatively, processing data may be transformed in step S14. For example, the characteristic values can be used as processing data.

Other embodiments may differ from the described embodiments in that, in steps S22 and T24, transformations of only the feature vectors but not all feature data are performed.

Claims

P a n t a n s p r e c h e

Method for checking value documents of the same type specified, in which

for one of the value documents, property data are determined which describe predefined properties of the value document,

for the value document, a class is determined from a number n of classes specified for value documents of the specified type, where n is greater than 1,

in dependence on the determined class, a transformation of at least some of the property data or processing data determined from the property data is performed in order to at least partially compensate for a dependency of the property or processing data on the determined class, and

for checking the value document, at least one predetermined criterion, preferably independent of the classes, which relates to the resulting property data or processing data, is checked, and a signal describing the result of the test is formed.

Method according to Claim 1, in which the value documents of the same predetermined type are banknotes of the same currency and denomination.

Method according to Claim 1 or 2, in which the property data are used to determine the class.

Method according to one of the preceding claims, wherein feature vectors are determined from the property data during the determination of the class, and preferably the feature vectors are used as processing data.

5. The method according to any one of the preceding claims, wherein the classes are predetermined so that each value documents are assigned to them, which are similar in relation to at least one aspect of the production of the value documents.

6. The method of claim 5, wherein as an aspect of the production of only the respective manufacturer of the value documents is used and preferably each of the classes are assigned only produced by a single manufacturer value documents.

A method according to claim 5, wherein production aspect batches in which value documents have been produced are used as an aspect with respect to the production.

8. The method according to any one of the preceding claims, wherein for the determination of the classes, a classifier and as a transformation transformation are used, which were determined by a method according to any one of claims 14 to 20.

9. The method according to any one of the preceding claims, wherein the property data describe spatially resolved optical properties of the value document, preferably an image in reflection or Trartsmission in different spectral ranges.

10. The method according to any one of the preceding claims, wherein the property data describe a color image.

11. The method of claim 4 in conjunction with claim 10, wherein the feature vectors describe color properties of predetermined evaluation areas of the image and preferably the color properties are used to determine the classes.

12. The method according to any one of the preceding claims, wherein the criterion comprises a criterion for the authenticity of the value document.

13. Method according to one of the preceding claims, in which the criterion includes a criterion for the state of the value document.

14. A method for determining data for carrying out a method according to one of claims 7 to 13, in which

- value documents of the same specified type and from different production batches are provided,

property data are determined for the value documents, which describe predefined properties of the value documents,

values of given feature vectors are determined from the property data,

depending on the values of the feature vectors, classes are defined to which value documents of the given type are assigned, which are considered to be similar with respect to the values of the feature vectors according to the predetermined criterion,

at least one classifier is defined, by means of which a value document of the given type can be assigned to one of the classes as a function of the property data determined for it, and thus a class can be determined for the value document, and

for at least one of the specified classes, a transformation for at least some of the property data or from the property Data determined processing data is determined by the dependence of the property or processing data of the determined class is at least partially compensated. 15. The method according to claim 14, wherein the value documents of the same predetermined type are banknotes of the same currency and denomination.

16. The method of claim 14 or claim 15, wherein the property data describe spatially resolved optical properties of the value document, preferably an image in reflection or transmission in different spectral ranges.

17. Method according to one of claims 14 to 16, in which a component analysis in a space of feature vectors determined from the property data is used to set the classes.

18. The method according to any one of claims 14 to 17, wherein for determining the classes a discriminant analysis in a space of feature vectors determined from the property data or in a subspace of the feature vectors is used.

19. The method according to any one of claims 14 to 18, wherein the classifier used is a linear classifier. 20. The method according to any one of claims 14 to 19, wherein, when the number of determined classes is a number n, a transformation is determined for a total of at least n-1 of the classes, by means of which for value documents of the given type determined property data or processing data determined by the property data common reference system can be transformed for all determined classes, which is given by the predetermined criterion for checking the value documents. 21. Apparatus for processing value documents with at least one input compartment for receiving value documents, a transport device for transporting isolated value documents to at least one output compartment, a sensor device arranged on a transport path given by the transport device with an evaluation device for carrying out a method according to one of

Claims 1 to 13, in particular the Ennittlung the property data, and a control device for controlling the transport device in response to signals of the evaluation device. 22. The device according to claim 21, which has a, preferably in the evaluation integrated memory in which all necessary for the implementation of the method data are stored.