Sensor and method for checking value documents, in particular bank notes, and value document processing apparatus

Abstract

The invention relates to a sensor and a method for checking value documents (2), in particular bank notes, and a value document processing apparatus (1). The sensor has a detection device (3) which is configured to detect electromagnetic radiation, which emanates from a value document (2) to be checked, in at least two different spectral ranges in a locally resolved manner, wherein a plurality of image points are obtained, to each of which at least two intensity values are assigned. Furthermore, the sensor has a processing device (7), which is configured i) to assign one feature point located in a feature space to a plurality of image points in each case, said feature point being defined by at least two feature point coordinates which are based on the intensity values assigned to the respective image point and / or adjacent image points, ii) for feature points which lie outside a feature region in the feature space specified for the respective image point by a curve or area, determining position information by way of which the position of the respective feature point relative to the feature region is characterised, and iii) assigning the feature points lying outside the feature region to different defect classes depending on the determined position information, said defect classes characterising different types of defects on the value document.

Description

[0001]Sensor and method for checking valuable documents, in particular banknotes, and a valuable document processing device. The present disclosure relates to a sensor and a method for checking valuable documents, in particular banknotes, as well as a valuable document processing device. A number of methods are known for mechanically checking the condition, also referred to as "fitness," of banknotes. However, these methods do not lead to equally reliable test results for all types of banknotes. For example, certain testing methods used for paper banknotes cannot simply be used to check so-called polymer banknotes. It is an object of the present disclosure to specify a sensor, a method, and a valuable document processing device for improved checking of valuable documents, in particular banknotes, which is / are particularly suitable for checking the condition of polymer banknotes.which have a window. This object is achieved by a sensor and a method according to the independent claims as well as a value document processing device with such a sensor. According to a first aspect of the present disclosure, a sensor for checking value documents, in particular banknotes, comprises: a detection device which is configured to detect, in particular transmitted and / or remitted, electromagnetic radiation emanating from a value document to be checked in a spatially resolved manner in at least two different spectral ranges, wherein a plurality of pixels are obtained, each of which is assigned at least two intensity values ​​which characterize the intensities of the radiation respectively detected in the at least two different spectral ranges; and a processing device which is configured toi) to assign a feature point located in a feature space to a plurality of pixels, which is defined by at least two feature point coordinates based on intensity values ​​assigned to the respective pixel and / or neighboring pixels of the respective pixel, ii) to determine position information for feature points that lie outside a feature area in the feature space specified for the respective pixel by a, in particular closed, curve or surface, by which the position of the respective feature point relative to the feature area is characterized, and iii) to assign the feature points lying outside the feature area to different defect classes depending on the respectively determined position information,by which different types of defects on the value document are characterized. Optionally, the processing device can further be configured to determine status information relating to the assignment of the feature points to the different defect classes. According to a second aspect of the present disclosure, a value document processing apparatus comprises: at least one processing device for processing, in particular transporting and / or checking and / or counting and / or sorting and / or destroying, value documents, in particular banknotes; and at least one sensor according to the first aspect of the present disclosure. A third aspect of the present disclosure relates to a method for checking value documents, in particular banknotes, in which i) signals emanating from a value document to be checked, in particular transmitted and / or remitted,electromagnetic radiation is detected spatially resolved in at least two different spectral ranges, whereby a plurality of pixels are obtained, each of which is assigned at least two intensity values ​​that characterize the intensities of the radiation detected in the at least two different spectral ranges, ii) a plurality of pixels is assigned a feature point located in a feature space, which is defined by at least two feature point coordinates that are based on intensity values ​​assigned to the respective pixel and / or neighboring pixels of the respective pixel, iii) for feature points that lie outside a feature range in the feature space specified for the respective pixel by a, in particular closed, curve or surface, position information is determined,by which the position of the respective feature point relative to the feature area is characterized, and iv) the feature points lying outside the feature area are assigned to different defect classes depending on the respectively determined position information, by which different types of defects on the value document are characterized. Optionally, state information regarding the assignment of the feature points to the different defect classes can be determined. A further aspect of the present disclosure relates to a computer program product, comprising instructions which, when executed by a computer, cause the computer to carry out the method according to the third aspect. Yet a further aspect of the present disclosure relates to a computer-readable storage medium, comprising instructions which, when executed by a computer, cause the computer toto carry out the method according to the third aspect. Aspects of the present disclosure are preferably based on the approach of spatially detecting the electromagnetic radiation transmitted and / or remitted from a region of interest (ROI) of a value document, such as a viewing window in a polymer banknote, in at least two different color channels, such as red, green, blue, and / or infrared, from the corresponding intensity values ​​for the obtained image points (pixels), determining a feature point, which is also referred to herein as a "feature vector," in a two-dimensional or higher-dimensional feature space, and then checking whether the feature points lie within or outside a feature range specified for the respective image point, preferably determined in advance in a training process.which is defined by a preferably closed curve (two-dimensional feature space) or surface (three- or higher-dimensional feature space). Pixels whose respective feature point lies within the respective specified feature range are classified, for example, as "regular pixels" or "fit pixels," whereas pixels whose respective feature point lies outside the feature range are classified, for example, as "irregular pixels," "outlier pixels," or "unfit pixels." For irregular pixels, the position of the respective feature point relative to the feature range is determined, and the irregular pixels or the associated feature points are classified into different defect classes depending on their respective position relative to the feature range, which are or can be associated with different types of defects at the respective location on the value document.such as graffiti (writing subsequently applied to the valuable document or paint splashes from the valuable document printing process), holes, creases, wear, scratches, color loss, and the like. The defect class assigned to the various defects or the position relative to the feature area relevant for assigning a feature point to the respective defect class is preferably determined in advance using valuable documents classified as "FIT" and / or valuable documents classified as "UNFIT" with corresponding defects. Optionally, status information can be generated based on the classification or assignment of the irregular pixels identified for the region of interest to the various defect classes.which characterize the condition or fitness of the respective value document in qualitative and / or quantitative terms. For example, the condition information contains details about the types of irregular pixels or types of defects identified in a value document at the corresponding location on the value document and / or the respective number of irregular pixels and / or the respective proportion of a certain type of irregular pixel in the total number of irregular pixels. Based on the classification or assignment of the irregular pixels identified for the region of interest to the different defect classes and any condition information obtained from this, value documents, in particular polymer banknotes, can be characterized in a particularly reliable and meaningful way, particularly with regard to their condition or fitness. Overall, this facilitates the examination of value documents,in particular polymer banknotes with a viewing window. When using the sensor or method for checking banknotes, in particular polymer banknotes that have a viewing window (so-called clear window), the region of interest from which the electromagnetic radiation emanating from the banknote is detected is selected such that it lies on a transparent section of the window or on an opaque section, for example a hologram. Preferably, the detection device is configured to detect the electromagnetic radiation transmitted from the region of interest on the value document in a bright field (i.e., when the value document is illuminated from the opposite side in the direction of the detection device) in a spatially resolved manner, so that a so-called bright field transmission image is obtained for each spectral range or spectral channel. Alternatively or additionally, it is also possible or preferred,to capture a dark-field transmission image (i.e., when the value document is illuminated from the opposite side not in or obliquely to the direction of the detection device) and / or a remission image. Preferably, the detection device comprises at least one radiation source, by means of which the value document is irradiated with electromagnetic radiation in different spectral ranges, in particular in bright-field, dark-field, or incident-light geometries. Preferably, the detection device is configured to capture the electromagnetic radiation emanating from the value document in different spectral ranges, which are in the visible (VIS), in particular red (R), green (G), and / or blue (B), and / or in the infrared (IR) spectral range, or are selected from these spectral ranges. It is further preferredthat the intensity values ​​assigned to the pixels in the different spectral ranges are normalized to the background of the value document. Preferably, the detection device is configured to detect electromagnetic radiation emanating from the background of the value document to be checked in a spatially resolved manner in at least two different spectral ranges, whereby a plurality of background pixels is obtained, each of which is assigned at least two background intensity values. Furthermore, the processing device is configured to normalize the intensity values ​​assigned to a pixel using the background intensity values ​​of the corresponding background pixels, for example by forming a quotient (for each pixel P and color channel K, the following then applies, for example: normalized intensity value (P, K) = intensity value (P, K) / background intensity value (P,K)). This achieves a so-called pixel-specific normalization of the respective intensities. The normalized intensity values ​​obtained in this way are used in the further process, in particular for determining or calculating the respective feature point coordinates. Through the normalization, influences on the respectively recorded intensity values ​​are eliminated or at least reduced, which can be caused, for example, by contamination, in particular of the sensor measurement window of the detection device, during the examination or processing of the value documents. Preferably, the processing device is configured to calculate one or more of the feature point coordinates of the feature points based on, preferably normalized, intensity values ​​assigned to the respective pixel and / or neighboring pixels of the respective pixel. The calculation includes, for example, the formation of a difference,a sum and / or a quotient. For example, if a feature point in the two-dimensional feature space is defined by two feature point coordinates f1 and f2 and the respective image point is assigned the intensity values ​​I, R , I G , I B and / or I IR in the red, green, blue or infrared spectral range, the feature point coordinates of the feature point assigned to this pixel can be calculated, for example, as follows: f1 = I R - I G and f2 = I G - I B Alternatively, it can also be provided to equate a feature point coordinate with one of the, preferably standardized, intensities, e.g. f1 = I IR, and to calculate the other feature point coordinate f2, e.g. by subtraction, as explained above, or by means of a local operation – described in more detail below – in which, in addition to the intensity values ​​assigned to the respective pixel, the intensity values ​​(I R , I G , I B and / or I IR ) in the respective spectral range or color channel. In this way, for example, using a maximum intensity value assigned to an immediately adjacent pixel, a contrast value can be determined for the respective pixel, which then gives the other feature point coordinate f2: f2 = contrast value. Alternatively, it is also possible to equate both feature point coordinates f1, f2 with one of the preferably standardized intensities, e.g., f1 = I VISand f2 = I IR , where I VISdenotes an intensity value that characterizes the intensity of the electromagnetic radiation detected in the visible spectral range. Preferably, the curve or surface by which the feature area in the feature space is specified for the respective pixel is described by a polynomial function in the feature space, in particular by an equation for an ellipse (two-dimensional feature space), an ellipsoid (three-dimensional feature space), or hyperellipsoid (n-dimensional feature space, n > 3). The polynomial function or the ellipse, the ellipsoid, or hyperellipsoid thus forms the outline of a kind of "fitness area" for the individual pixels or the feature points in the feature space assigned to the pixels. By selecting the polynomial function, in particular its degree and / or its coefficients, the position, orientation, shape, and / or size of the feature area can be specified in a simple and reliable manner. Preferably,The position information determined for the feature points lying outside the feature area is directional information and / or distance information, which (quantitatively) characterizes a direction or distance in which the respective feature point lies relative to the feature area. The distance information is quantitative, i.e., it goes beyond simply specifying whether the respective feature point lies within or outside the feature area. The quantitative distance information contains, for example, information about the distance of the respective feature point from the edge of the feature area and / or about the distance of the respective feature point from one or more points (e.g., center or focal point(s)) within the feature area. If the feature area is defined in the two-dimensional feature space, for example, by an ellipse, the directional information indicateswhich direction the feature point in question lies relative to the ellipse, e.g. in the x and / or y direction. The distance information indicates, for example, the distance or distances D1 or D2 that the feature point in question has to a focal point or to the focal points of the ellipse. Based on the distances D1, D2 of the feature point to the focal points of the ellipse, it can be determined particularly easily and reliably whether the feature point lies inside or outside the ellipse. If D1+D2 > D, then the feature point lies outside the ellipse; otherwise, the feature point lies inside the ellipse, where D = 2a denotes the largest diameter (i.e. the length of the major axis = 2 x length a of the semi-major axis) of the ellipse and can preferably be determined in advance in a training process. In principle, it is also possible for the distance information to characterize the distance of the feature point to the center of the ellipse. Alternatively,In addition, it can also be provided that the distance information characterizes a distance D3 and / or D4 of the feature point to the major axis or minor axis of the ellipse. In this case, the respective irregular pixel can be classified on the basis of the major axis distance D3 or minor axis distance D4 of the associated feature point, i.e., the defect class or type of defect is determined in this case based on D3 or D4. For example, it can be provided that an irregular pixel is classified as a "graffiti pixel" (corresponding to a location on the value document with paint splashes or a subsequently applied inscription) if D3 is greater than a predetermined threshold or reference value, which can preferably be determined in advance in a training process. Preferably, one or more of the following types of defects on the value document are characterized by the different defect classes and / or are assigned to theOne or more of the following types of defects on the value document are assigned to different defect classes: subsequently applied inscriptions (when examining used value documents), paint splashes (during an inspection subsequent to the value document's production), wear, color loss, creases, scratches, holes, and / or a defect classified as "unknown." Taking this assignment into account, the processing device can reliably infer, based on the respective defect class to which feature points were assigned, the presence and, if applicable, the number of a specific type of defect in the respective pixels to which these feature points were assigned, or at the corresponding locations on the value document. The status information preferably contains one or more of the following information: the number of feature points lying outside the feature range and / or the number of each defect classassigned feature points and / or the proportion of the number of feature points assigned to a defect class to the number of feature points lying outside the feature area. The presence of irregular pixels or defects at the corresponding locations on the value document can thereby be reliably characterized in a simple manner, both quantitatively and qualitatively. Alternatively or additionally, it can be provided that the status information contains an average of the distances that the feature points or irregular pixels (e.g., "graffiti pixels") assigned to a defect class have from the feature area, in particular from the main axis of the ellipse. Preferably, when determining the position information and / or assigning the feature points lying outside the feature area to different defect classes, it can also be averaged whether a systematic deviation relative tofeature area, in particular to the ellipse. This can be seen, for example, as an indication of systematic errors in security printing, e.g. due to fluctuations in print quality at the same position (x, y) in the region of interest. Such a systematic deviation relative to the ellipse can be detected and / or suspected, for example, if an irregular pixel always occurs at the same position (x, y) in the ROI for several value documents, for example 100 banknotes, and in particular the same type of defect (e.g. always blue color) is present. Then there is at least a suspicion that a systematic error occurred at the value document manufacturer, so that a corresponding warning message can be issued to the operator, displaying and / or specifying the erroneous position (x, y) in the ROI. Preferably, the processing device is configured to transfer the performed assignment (classification) of theto write the irregular pixels of the different defect classes into an “unfit map” and / or to create such an “unfit map” which shows the positions of the irregular pixels in the spatial space. Preferably, the processing device can further be configured to control a display device, for example a display, such that it displays an image of the valuable document or the ROI together with the “unfit map”, preferably superimposed on the image, i.e. the irregular pixels are preferably displayed in the displayed image of the valuable document. Depending on the application or configuration of the processing device, the display of the respective irregular pixels can be carried out, for example, for a large number of valuable documents and / or only for the most frequently irregular pixels. The respectively displayed image of the valuable document or the ROI and the location information (x, y) of irregular pixels is then available to the operator for further analysis.Preferably, the status information is used to check the value documents. For example, the processing device is configured to check the value document to be checked based on the status information of the respective value document, in particular based on the aforementioned number(s) and / or proportion(s) of feature points, e.g., to classify it with regard to its status, in particular as "FIT" or "UNFIT." This allows for highly differentiated or application-specific specification of criteria for classifying the status of value documents. For example, the processing device can be configured such that a value document is classified as "FIT" or "UNFIT" if the number of feature points or irregular pixels lying outside the feature area determined for a region of interest on the value document is smaller or larger than a predetermined threshold.Alternatively or additionally, a classification as "FIT" or "UNFIT" can occur if the number and / or proportion of feature points or irregular pixels assigned to a specific defect class is smaller or larger than a threshold value for the number or proportion specified for this defect class. It can also be provided to classify value documents as "UNFIT" if a specific type of defect is present, e.g., a possibly very small hole, by assigning at least one feature point lying outside the feature range to a specific defect class. It is also possible to specify multiple types of defects or defect classes, so that a classification as "UNFIT" occurs if a value document has at least two, preferably all, of the specified types of defects. To check the value documents based on the status information, the processing device can additionallyor alternatively, be configured to provide the status information of the value document(s) and / or information derived from the status information, in particular information relating to the status of the value document(s), for further use, e.g. for a later joint evaluation of a plurality of value documents, in particular for output, for example on a display, to an operator of the sensor or the value document processing device, for storage together with other information relating to the value document(s) (e.g. image data, serial number, etc.) for documentation purposes and / or for use in processing the respective value document, in which, for example, a sorting of the value document takes place depending on the status information. Preferably, the feature area specified for each pixel by a, in particular closed, curve or surface in theFeature space, in particular in advance in a training process, in which: i) electromagnetic radiation emanating from several test value documents, in particular classified as "FIT", is recorded in a spatially resolved manner in the at least two different spectral ranges, wherein for each test value document, a plurality of pixels are obtained, each of which is assigned at least two intensity values ​​that characterize the intensities of the radiation emanating from the respective test value document and recorded in the at least two different spectral ranges; ii) several pixels of each test value document are each assigned a feature point located in a feature space, which is defined by at least two feature point coordinates that are based on intensity values ​​assigned to the respective pixel and / or neighboring pixels of the respective pixel; and iii) for several of the pixelsIn each case, a curve or surface is determined within which the, in particular all, feature points obtained from the test value documents for the respective pixel lie. The curve or surface determined in this way specifies the outline of the feature area (so-called "fitness area") in the feature space for the respective pixel, which is used when checking value documents. Preferably, the feature area specified for a respective pixel in the feature space is described by a polynomial function, in particular an equation for an ellipse, an ellipsoid, or hyperellipsoid, which is determined by means of principal axis transformation from the feature points obtained from the test value documents for the respective pixel. The outline of the fitness area determined for the respective pixel can thus be determined in a particularly simple and reliable manner. Preferably, the training method is carried out usingof N test value documents, preferably with N = 50 to 150, in particular N = 100, in very good condition ("FIT" or "ATM-FIT"). Preferably, the electromagnetic radiation transmitted and / or remitted by a region of interest (ROI), such as a viewing window with a transparent section and / or an opaque hologram section, of the test value documents, in particular the bright field transmission, is recorded in at least two different color channels, such as red, green, blue and / or infrared, with spatial resolution, and from the corresponding intensity values, which are preferably normalized to the intensity of the background of the test value documents, a feature point (so-called "feature vector") is determined for each of the image points (pixels) obtained in a two- or higher-dimensional feature space (so-called "feature space"). For several or all pixels of the ROI, preferably N feature points are obtained in the feature space.Preferably, a polynomial function, in particular an ellipse, is determined for each individual pixel, within which all N pixel-specific feature points lie. The ellipse assigned to a pixel is preferably determined by principal axis transformation. Preferably, a check of the pixel-specific feature areas determined in this way, in particular the polynomial functions or ellipses, can be provided, in particular based on M test value documents (preferably with M = 50 to 150, in particular M = 100), which are classified as "UNFIT" at least with regard to the state of the viewing window and for which, as with the test value documents classified as "FIT", preferably M feature points are obtained in the feature space for several or all pixels of the ROI. Then, a pixel-by-pixel check is performed to determine whether all M feature points lie outside the polynomial function or ellipse determined for this pixel based on the "FIT" test value documents.If this is the case, the polynomial function or ellipse represents a clear separation between value documents to be classified as "FIT" or "UNFIT". If not, it can be provided that the respective polynomial function or ellipse is iteratively adapted to the respective N feature points until it enables the clearest possible separation between "FIT" and "UNFIT". Preferably, for each pixel of the ROI, the position of the two focal points of the respectively obtained polynomial function or ellipse can be determined and / or twice the length 2a of the semi-major axis can be set as the threshold value D for the sum of the distances D1+D2 of feature points during the value document verification. Alternatively or additionally, it can also be provided that for each pixel of the ROI, the distances D3 and / or D4 of feature points lying on the ellipse to the major axis or minor axis of the ellipse are calculated as reference points or values, which are then used when verifying the value documents.can be. Further advantages, features, and possible applications of the present invention emerge from the following description in conjunction with the figures. They show: Fig. 1 shows an example of a value document processing device with a sensor for checking value documents; Fig. 2 shows an example of a banknote with a viewing window; Fig. 3 shows an example of a 3x3 environment of a pixel to illustrate a local operation; Fig. 4 shows a first example of an ellipse; Fig. 5 shows a second example of an ellipse; Fig. 6 shows an example of a captured image of a viewing window of a training banknote; Fig. 7 shows an example of feature points obtained for training banknotes to illustrate the determination of the parameters of an ellipse; Fig. 8 shows an example of a diagram to illustrate different statistical distributions of the respectively determined distance D for FIT and UNFIT training banknotes; and Fig. 9 shows an example to illustratepossible classifications of different feature points. Figure 1 shows a schematic representation of an example of a value document processing device 1 with an input device 9, for example a so-called input compartment, for receiving a stack 10 of value documents 2, in particular banknotes, which are individually removed from the stack 10 by means of a separating device (not shown) and transported along a transport path 6 by means of a transport device 4. In the present example, the transport device 4 has a plurality of transport elements 4a-4c and switches 5a-c, which are only shown schematically. Furthermore, a sensor for checking the value documents 2 is provided, which has at least one detection device 3, which is configured to detect electromagnetic radiation emanating from a value document 2 to be checked in each case in at least two different spectral channels or spectral ranges, in particular in thevisible and / or infrared spectral range, spatially resolved. The electromagnetic radiation emanating from the value document 2 can, for example, be electromagnetic radiation transmitted, reflected, and / or remitted (i.e., diffusely reflected) by the value document 2. Preferably, the detection device 3 detects the electromagnetic radiation transmitted by the value document 2 in the bright field (so-called bright field transmission). For this purpose, an irradiation device 8 for generating visible and / or infrared electromagnetic radiation is provided on the side of the value document 2 or transport path 6 opposite the detection device 3, by means of which the value document 2 to be checked is exposed. Alternatively, the irradiation device 8 can also be arranged or aligned such that the detection device 3 detects the electromagnetic radiation transmitted by the value document 2 in the dark field.(so-called dark field transmission), whereby the irradiation device 8 then does not radiate directly in the direction of the detection device, but obliquely thereto. Alternatively or additionally, an irradiation device (not shown) can also be provided on the side of the value document 2 or transport path 6, on which the detection device 3 is also located, in order to detect the electromagnetic radiation reflected or remitted by the value document 2. The sensor shown in the example shown can in principle be used to check different types of value documents, but is particularly suitable for checking polymer banknotes, i.e. banknotes with a sheet-shaped substrate made of plastic and / or a hybrid composite, e.g., a cotton layer (core) and a thin plastic film (coating), and / or banknotes equipped with a so-called window or viewing window, wherein the window has one or more transparentSections and optionally also non-transparent sections, for example in the form of a hologram. The electromagnetic radiation detected by the detection device 3 with spectral and spatial resolution thus provides signals or intensity values ​​for each image point (pixel) in the at least two different spectral channels, which represent a measure of the spectral intensities of the respectively detected radiation. The detection device 3 can be, for example, a color camera or several monochrome cameras or two or more detector arrays (multi-track sensors) provided with different color filters for the spatially resolved detection of the electromagnetic radiation emanating from the value document 2 in the visible and / or non-visible (e.g. ultraviolet and / or infrared) spectral range. Optionally, further sensors (not shown), such as ultrasonic, magnetic and / or capacitive sensors, can be used to detect further properties of theValue documents 2 may be provided. Based on the electromagnetic radiation detected spatially resolved in at least two spectral ranges by the detection device 3 and, if applicable, based on the properties detected by further sensors, the value document 2 is checked in a processing device 7, in particular with regard to its condition (fitness), such as graffiti (writing or paint splashes subsequently applied to the value document), soiling, creases, wear, etc., and, depending on the result of the check, is output or sorted to one of several output compartments 11a-d. For this purpose, the transport elements 4a-c and / or switches 5a-c are controlled or actuated accordingly by the processing device 7 and / or a control device. Preferably, the processing device 7 is designed as a computer and / or the processing device 7 has a processor for data processing and a memory forStoring data. The processing or evaluation of the electromagnetic radiation detected spatially resolved in at least two spectral ranges in the processing device 7 is explained in more detail below using examples. Figure 2 shows a schematic representation of a banknote 2 with a foil element 12, which is also referred to as a window or viewing window and in the present example is provided with a hologram element 13. Depending on the design, the foil element 12 is preferably permeable or transparent to electromagnetic radiation, in particular in the visible and / or infrared spectral range, whereas the hologram element 13 is essentially impermeable or opaque to the electromagnetic radiation. In the example shown, the banknote 2 is transported in a transport direction X past the detection device 3 (see Figure 1), which extends in the Y direction, for example inForm of a color camera, and the electromagnetic radiation emanating from the banknote 2 during transport is recorded in a spectrally and spatially resolved manner. In this case, a large number of pixels are recorded for the banknote 2 in each color channel c (e.g. red, green, blue, infrared, etc.). with the coordinates (x, y) obtained (banknote pixels). Each pixel ^^ ^ ^ ௫ ,௬^ with the coordinates (x, y) thus contains two or more signal or intensity values ​​that characterize the intensities of the electromagnetic radiation emanating from the banknote 2 and detected in the respective color channels c. Preferably, the electromagnetic radiation can also be detected before and / or after the banknote 2 passes or has passed the detection device, whereby a plurality of pixels ^^ ௬ ^ of the background (background pixels). Each background pixel ^^ ௬ ^with the coordinate y thus contains two or more signal or intensity values, which characterize the intensities of the electromagnetic radiation of the banknote background recorded in the respective color channels c. For the banknote pixels ^^ ^ ^ ௫,௬^ at least two pixel-specific features ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^, which in the context of the present disclosure are also referred to as "features" or "feature point coordinates". Preferably, at least one of the features ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^ ^ from the intensity value of a channel c of the banknote pixel ^^ ^ ^ ௫,௬^ with the coordinates (x, y) and the corresponding background pixel ^^ ௬ ^ with the coordinate y calculated as follows: ( 1-1) which represents a normalization of the intensity values ​​of the banknote pixels with respect to the intensity values ​​of the background pixels. Alternatively or additionally, at least one of the features ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^ can be calculated by subtracting the intensities of different color channels i, j, preferably normalized to the background: (1-2) Alternatively or additionally, at least one of the features can be calculated by means of a local operation performed in a color channel c, in which the intensity values ​​in the neighborhood of the respective pixel are taken into account, such as a 3x3 environment illustrated in Figure 3, in which the pixel ^^ ^ ^ ௫ ,௬^ total of 8 neighboring pixels ^^ ^ ^ ௫ േ^,௬േ^^ For example, a contrast value can be determined using a local operation. To do this, the maximum intensity ^^ ^^௫the respective and then calculates the pixel-specific feature as follows: The pixel-specific features obtained in this or other ways ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^ are then analyzed using a given model, in particular a fitness model. Such a model is preferably a pixel-specific model m^ ^^ ^ , ^^ ଶ ^, which is preferably represented by a non-linear function F of two features ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^ which is calculated, for example, according to the above formulas (1-1) to (1-4) and / or in another way from the respective banknote pixel ^^ ^ ^ ௫,௬^ assigned intensity values ​​can be derived: ^^^ ^^ ^ , ^^ ଶ ^ ൌ ^^^ ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^^. (1-5) The two characteristics ^^ ^ , ^^ ଶcan be statistically dependent on each other or independent. The two-dimensional nonlinear function F is preferably a generic polynomial function (n-th degree polynomial with n ≥ 2), whose parameters are determined depending on the application (e.g., type of verification, such as condition or authenticity testing, banknotes or denominations to be verified, etc.), in particular by means of a pre-conducted training procedure. To illustrate the use of a generic polynomial function, the normal form of an ellipse (second-degree polynomial) is used below as a nonlinear function: where the parameters a and b denote half the length of the major axis (major semi-axis) and minor axis (minor semi-axis) of the ellipse, respectively. As illustrated in Figure 4, an ellipse E can be defined as a plane oriented in the direction of the axis of the first feature around a and in the direction of the axis of the second feature ^^ ଶThe ellipse E is the set of all points lying in a plane for which the sum of their distances (so-called focal distances) from the two foci fp1 and fp2 has the constant value 2a (double major axis), which is also called the ellipse radius D. The distance c is called the linear eccentricity and indicates the distance of the respective focal point fp1 or fp2 from the center of the ellipse E. The distances from the axes of the first and second feature ^^ ^ , ^^ ଶ The plane spanned is also referred to as "feature space" in the context of the present disclosure. If a pixel-specific feature point determined for the banknote 2 ^^^ ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^^ lies within the ellipse E, then the feature point ^^^ ^^ ^ ^ ^^, ^^^, ^^ ଶ^ ^^, ^^^^ and / or the corresponding banknote pixel with the coordinates (x, y) is classified as "FIT" or regular, if it lies outside the ellipse E, then the feature point ^^ ଶ ^ ^^, ^^^^ and / or the associated banknote pixel with the coordinates (x, y) are classified as "UNFIT" or irregular. This is illustrated by way of example in Figure 5. Preferably, when checking whether a feature point P( ^^ ^ , ^^ ଶ ^ inside or outside the ellipse E, two distances D1 and D2 (see Figure 5) are determined or used, namely the distance D1 of the If D1 + D2 ≤ D = 2a, then the feature point P ( ^^ ^ , ^^ ଶ ^ within or on the ellipse E, if D1 + D2 > D, then the feature point P ( ^^ ^ , ^^ ଶ^ outside the ellipse E. This allows a reliable and simple way to determine whether a feature point is to be classified as regular or irregular. If, unlike in the present example, the center of the ellipse E is not located at the origin of the coordinate system of the feature space and / or both axes of the ellipse E are not parallel to the axes of the coordinate system, then both selected features are preferably converted using an affine transformation, so that the above-mentioned formula (2-1) remains valid, i.e., the ellipse E can still be defined by the normal form. Depending on the selection of features, i.e., depending on the definition or calculation of the feature point coordinates ^^ ^ , ^^ ଶIn the feature space, different models, in particular fitness models for determining and / or testing the condition of banknotes, can be specified using such an ellipse E. For example, in a first feature space "Color," the features or feature point coordinates ^^ ^ , ^^ ଶ calculated from the difference between the intensity values ​​of two color channels (red, green, blue) such as = Intensity (red) – Intensity (green) and ^^ ଶ = Intensity (green) – Intensity (blue). Using a "Color Model" ellipse specified for this feature space, the presence of graffiti (paint splashes or inscriptions subsequently applied to the security document) in the transparent area of ​​the foil element 12 can be detected with particularly high reliability. In a second feature space, "Brightness," however, the following features or feature point coordinates can be used: ^ , ^^ ଶthe intensity values ​​of the electromagnetic radiation detected in the visible (VIS) or infrared (IR) spectral range are used: VIS intensity = and IR intensity = ^^ ଶ Using a "Bright Model" ellipse specified for this feature space, creases, wear and / or defects in the transparent area of ​​the film element 12 can be detected with particularly high reliability. In a third feature space, "Contrast", the following can apply: IR intensity = and contrast value = ^^ ଶ(For calculating a pixel-specific contrast value using a local operation, see above). Using a "Contrast Model" ellipse specified for this feature space, creases, wear, and / or defects on the hologram element 13 of the foil element 12 can be detected with particularly high reliability. Preferably, for the feature points lying outside the ellipse E, i.e., those classified as "UNFIT" or irregular, ^^^ ^^ ^ ^ ^^, ^^^, ^^ ଶ ^ ^^, ^^^^ each determines a position information, by which the position of the respective feature point ^^^ ^^ ^ ^ ^^, ^^^, ^^ ଶ^ ^^, ^^^^ relative to the ellipse E, and the feature points or the associated banknote pixels are assigned to different defect classes depending on the respectively determined position information, which are associated with different types of defects. Based on the assignment of the feature points to the different defect classes, condition information of the respective banknote is determined, which can then be used - if necessary together with other criteria - to check the condition of the respective banknote and, if necessary, for its sorting. In addition to the classification as "FIT" or regular or "UNFIT" or irregular, the irregular pixels are also classified with regard to their position relative to the ellipse or the type of defect. This is explained in more detail below. For example, a pixel classified as irregular in the "Color" feature space, for example,classified as a graffiti pixel if the major axis distance D3 (see Figure 5) is large or exceeds a predefined threshold value. Optionally, the color of the graffiti can be determined based on the minor axis distance D4 (see Figure 5). Alternatively or additionally, the color itself can be determined for a pixel classified as irregular using the RGB intensity values ​​and compared with a banknote color specified for this pixel or the corresponding location on the banknote. If a larger deviation is detected, the presence of graffiti can be concluded. A classification according to the type of irregular pixel can also be made on the basis of several major and / or minor axis distances D3 or D4 in several different feature spaces, in particular "color", "brightness" and / or "contrast". As already explained above, the window orFilm element 12 can have both an opaque window section, for example in the form of a hologram element 13, and a transparent window section (so-called see-through area). Preferably, when detecting the electromagnetic radiation emanating from the banknote 2, the region of interest (ROI) is placed either on the opaque or on the transparent window section. Alternatively, it can also be provided that an ROI is located in each of the two window sections. Preferably, the criteria for classifying the type of irregular pixels can be different in different ROIs of the window, e.g., with an ROI in the transparent window section or in the opaque (hologram) window section.For example, it can be provided that, in the case of an ROI located in the transparent window section, a pixel classified as irregular in the “brightness” feature space is classified as a graffiti pixel if the major axis distance D3 in this feature space is large or exceeds a predetermined first threshold. Furthermore, it can be provided that if the major axis distance D3 is small or falls below the predetermined first threshold and the minor axis distance D4 is large or exceeds a predetermined second threshold, the pixel is classified as a wrinkled or hole pixel. Alternatively or additionally, it can also be provided that a distinction is made in the “brightness” feature space between wrinkled pixels and hole pixels, for example on the basis of the minor axis distance D4.Alternatively or additionally, it can be provided that, for an ROI located in the opaque window section (hologram), a pixel classified as irregular in the "Contrast" feature space is classified as a wrinkled pixel or hole pixel if the major axis distance D3 in this feature space is large or exceeds a predefined threshold. If the major axis distance D3 is small or falls below the predefined threshold, the pixel is classified as a wear pixel. The classification or determination of the type of irregularity of a pixel can preferably be carried out step by step as follows. First, a check is carried out in the "Color" feature space to determine whether the feature point assigned to a pixel lies outside a predefined "Color Model" ellipse. If this is the case, this feature point or the corresponding pixel is classified as a graffiti pixel and, if applicable, its color is determined.If the feature point assigned to the pixel lies within the "Color Model" ellipse, then a further test is carried out in the "Brightness" feature space using a predefined "Bright Model" ellipse as follows: a) If the pixel belongs to the ROI in the transparent window section (see, for example, the areas of the foil element 12 outside the hologram element 13 in Figure 2), a check is carried out to determine whether the feature point in the "Brightness" feature space lies outside the "Bright Model" ellipse. If this is the case, it may be a hole, a crease, or wear at the relevant location on the banknote. A distinction is made between these defect classes preferably based on the major axis distance D3 of the feature point and / or based on VIS and / or IR intensities, with different thresholds preferably being predefined. If, for example, the VIS and / or intensity is very high orIf the corresponding thresholds are exceeded, then it can be concluded that there is a hole at that point on the banknote. If, on the other hand, the intensity is low or the corresponding thresholds are not met, then it can be concluded that there are creases and / or wear at that point on the banknote. If the feature point lies within the "Bright Model" ellipse, it cannot be concluded that the corresponding pixel has any brightness irregularity. b) If the pixel belongs to the ROI in the opaque window section (see, for example, the hologram section 13 on banknote 2 in Figure 2), a check is carried out to determine whether the feature point lies within the "Color Model" ellipse in the "Color" feature space. If this is the case, a further check is carried out in the "Contrast" feature space using a predefined "Contrast Model" ellipse.If the feature point lies outside the "Contrast Model" ellipse, then the presence of a crease, a hole, wear, or color flaking at the relevant location (hologram element 13) on the banknote 2 can be inferred. Preferably, a distinction can also be made between these defect classes based on the color of the respective pixel and / or the major axis distance D3 and / or minor axis distance D4. Preferably, the models used in the analysis or testing of the banknote pixels or the different polynomial functions characterizing the models, in particular ellipses E, are determined in advance in a training process. Preferably, it is first determined which ROI or positions, e.g. on the foil element 12 and / or hologram element 13, of a specific banknote type (e.g. denomination) are to be used to determine a model.Using the image points obtained from training banknotes, the parameters of the polynomial function(s), in particular ellipse(s), are then calculated pixel-specifically and / or statistically estimated. A preferred training method is explained in more detail below using the example of determining a "color model" ellipse. However, the described method also applies accordingly to determining other models, in particular a "bright model" and / or "contrast model" ellipse or other polynomial functions of the second and / or higher degree. For training, preferably 100 training banknotes of a specific type are used, which are classified as "FIT" or "ATM-FIT" (very good condition). Optionally, the training result can be checked using 100 additional training banknotes of the same type, which are classified as "UNFIT" (poor or unsatisfactory condition).Figure 6 shows an image captured by the capture device 3 (see Figure 1) of a viewing window provided in a training banknote with transparent sections of a foil element 12 and non-transparent sections of a hologram element 13 (see also Figure 2). Using this image, several pixels or positions on the foil element 12 and / or the hologram element 13 are defined manually and / or with the aid of a suitable image processing algorithm, which are to be used to determine the respective ellipse in the feature space. The pixels defined in the present example are highlighted in the present illustration as small rectangular or square light or gray areas and are usually located in the area of ​​contours or transitions between the transparent sections of the foil element 12 and the non-transparent sections of the hologram element 13.The “Color Model” preferably uses two color transformation features (cf. equation (1-2)) as follows:. For 100 training banknotes, the following applies to the position (x,y): ^^ ^ ^ ^^, ^^^ ൌ ^ ^^ ^ ^ ^ ^^, ^^^, ^^ ^ ଶ ^ ^^, ^^^, ^^ ^ ଷ ^ ^^, ^^^… , ^^ ^ ^^^ ^ ^^, ^^^} (2-6) ^^ ଶ ^ ^^, ^^^ ൌ ^ ^^ ଶ ^ ^ ^^, ^^^, ^^ ଶ ଶ ^ ^^, ^^^, ^^ ଶ ଷ ^ ^^, ^^^... , ^^ ଶ ^^^ ^ ^^, ^^^ ^ (2-7) In the corresponding two-dimensional feature space, these features can be described as a two-dimensional feature vector X: (2-8) where ^ ^^ ൌ ^^^ ^ ^^, ^^ ^ (2-9) ^^ଶ ൌ ^^ଶ ^ ^^, ^^ ^ (2-10)Assuming that the desired ellipse encloses most or preferably all 100 feature points of the training banknotes, preferably with minimal error, a major axis transformation can be used to determine the major axis of the ellipse. As illustrated in Figure 7, the major axis of the ellipse E is formed by the PC1 axis (first principal component). The direction of the PC1 axis is the eigenvector and its magnitude is the eigenvalue. The angle ^^ of the X1 axis to PC1 is the rotation angle used in the transformation. A line orthogonal to PC1 is calculated, which represents the second principal component PC2 and, instead of the original X2 axis, forms the new minor axis of the ellipse. The minor axis PC2 describes the largest variance not described by PC1. As illustrated in Figure 7, the greatest variation is along the PC1 axis.The broadest statistical distribution 14 of the features is found, which is encompassed by the ellipse E. In contrast, along the PC2 axis (minor axis), the variation or statistical distribution 15 of the features is smallest or narrowest. The ellipse parameters are preferably calculated or statistically estimated as follows: 1. Center of the ellipse. The center of the ellipse E is determined by the mean. and ^^ ଶ of two features ^^ ^ and ^^ ଶ derived: 2. Major / Minor Axis of the Ellipse As explained above, the eigenvalues ​​represent statistical variances of the major / minor axis of the ellipse E, and the eigenvectors indicate the direction of the major / minor axis. Eigenvalues ​​and eigenvectors are preferably determined via the covariance matrix of ^^. Preferably, a symmetric covariance matrix is ​​created first: The eigenvalues ​​^^ of the covariance matrix ^^ are determined by solving the following equation: ^^ ^^ ^^^Σ െ ^^ ^^^ ൌ 0 (2-14) Here, ^^ represents the identity matrix. This equation can be reformulated as follows: ^Σ െ ^^ ^^^ ∙ ^^ ൌ 0 (2-15) The eigenvectors are sorted in descending order by eigenvalue. The direction of the principal axis or rotation angle ( ^^ to axis ^^ ^ ^ the ellipse E can be calculated as follows: 3. Length of the major / minor axis of the ellipse The length of the major / minor axis a or b is determined using the first eigenvalue: (2-17) ^^ ൌ ^^ ଶ ∙ ^^2 (2-18) where k1 and k2 are adaptation parameters and typically range from 2 to 8. For example, the value 3 can be used as a default value (3 Sigma principle). 4. Focal points of the ellipse The focal points fp1 and fp2 are determined by calculating the linear eccentricity c (see Figure 4) as follows: ^ ^ ൌ √^^ଶ െ ^^ଶ (2-19)5. Estimation of a measure for the size of the ellipse Using formulas (2-2) and (2-3), for each training banknote, the distance of the respective feature point P to the focal points fp1( ^^ ^ , ^^ ଶ ^ and fp2^ ^^ ^ , ^^ ଶ ^ calculated. From the 100 values ​​obtained for the distances D1 and D2, a maximum value is determined as the upper threshold D: As already explained in more detail above, during the subsequent inspection of banknotes it is preferably determined based on the upper threshold D whether it is a feature point or corresponding image point that is to be classified as "FIT" or regular or "UNFIT" or irregular. Theoretically, the value D = 2a. However, it can happen that with the parameters a, b and adaptation parameters k1, k2 determined using the "FIT" training banknotes (see formulas (2-17) and (2-18)), the feature points do not lie within the ellipse E for all 100 training banknotes. The upper threshold D determined using formula (2-20) can therefore also be greater than 2a under certain circumstances. 6. Verification of the model The determined model or the determined ellipse E is preferably verified using 100 further training banknotes that are classified as "UNFIT". For this purpose, the formulas (2-2) and (2-18) are preferably used.(2-3) the distances D1 and D2 are calculated for all training banknotes and checked to see whether the following applies to all UNFIT training banknotes: D1 + D2 > D. Alternatively or additionally, it can be checked whether the statistical distribution of the threshold or distance D determined for the FIT and UNFIT training banknotes allows a clear separation between FIT and UNFIT. This is illustrated by the diagram shown in Figure 8, in which the number n of training banknotes is plotted against the distance D determined for the respective training banknote. In the example shown, the statistical distribution 16 of the distance D for FIT training banknotes is clearly separated from the statistical distribution 17 of the distance D for UNFIT training banknotes. The distance D determined in each case based on the model is therefore suitable as a threshold to reliably determine whether a feature point lies inside or outside the ellipse.In the event that the distance D does not allow a clear separation between FIT and UNFIT, it is preferably provided to increase or decrease at least one of the adaptation parameters k1, k2 (see formulas (2-17) and (2-18)). 7. Advantages of the fitness model or the fitness ellipse Compared to known models, such as the so-called Mahalanobis distance, the fitness model described above or the fitness ellipse obtained thereby or used to check banknotes offers the following advantages in particular: a) The geometry and positions of the ellipse are interpretable. For example, the major axis of the "Color Model" ellipse describes the hue of the printing ink of the film element 12, while the size of the major axis represents a measure of the saturation and the size of the minor axis reflects fluctuations in the hue. This makes it possible not only to detect an UNFIT pixel, but also to classify it or specify it further.This is illustrated by Figure 9, which shows a two-dimensional feature space with the feature coordinates f1 and f2. In this example, the feature points for a specific banknote pixel were determined from a total of nine banknotes, with the feature points P. FITof seven banknotes lie within the ellipse E specified for this pixel and are accordingly classified as "FIT" or regular, whereas two feature points P1, P2 lie outside the ellipse E and are accordingly classified as "UNFIT" or irregular. The first feature point P1 is classified as "graffiti" due to its greater distance D3 to the major axis of the ellipse E, which suggests a strong color deviation from that of a specified color. The second feature point P2 was also classified as "UNFIT" or irregular, but is given less weight in the condition classification of the banknote or may even be ignored because, due to its short distance D3 to the major axis and / or its distance D4 to the minor axis, only a slight color deviation or only slightly different color saturation can be assumed.The position of the respective feature point P1, P2 relative to the fitness ellipse E thus provides additional information or attributes on the basis of which a condition classification and, if necessary, subsequent sorting of the respective banknote can be carried out. To classify the condition of the respective banknote, the positions (distances from the ellipse E or directions relative to the ellipse E) of the feature points of a large number of irregular pixels are preferably used. b) The fitness ellipse describes a general fitness model that is independent of the statistical properties of the two features. In contrast, the use of the Mahalanobis distance, for example, requires that the two features are statistically correlated, which does not have to be the case with the fitness ellipse. Thus, while an evaluation using the Mahalanobis distance / ellipse only checks whether a feature vector orFurthermore, when determining whether a feature point lies within or outside the Mahalanobis distance / ellipse, or the distance to the center of the ellipse, the direction in the feature space in which the feature point lies outside the ellipse is not taken into account. This means that only a distinction is made as to whether a pixel is irregular or regular, but the irregular pixels are not further classified according to different types of irregular pixels or defect classes. Furthermore, the Mahalanobis distance / ellipse in a feature space can only be created with all features if the features are correlated, and only one threshold is used as the "Mahalanobis distance." This means that the application of the Mahalanobis distance / ellipse requires a statistical correlation between features, since the value of the covariances between features cannot be close to "0." In practice, for example,the intensity value of a VIS color channel “blue” and the intensity value of the IR channel of the same pixel are not statistically correlated depending on the banknote layer.

Claims

Patent claims 1. Sensor for checking value documents (2), in particular banknotes, with a detection device (3) which is designed to detect, in particular transmitted and / or remitted, electromagnetic radiation emanating from a value document (2) to be checked in a spatially resolved manner in at least two different spectral ranges, wherein a plurality of pixels is obtained, to which at least two intensity values ​​are assigned, which characterize the intensities of the radiation detected in the at least two different spectral ranges, and a processing device (7) which is designed to - a plurality of pixels ( ^^ ^ ^ ௫,௬^) to each of the feature points (P(f1, f2)) located in a feature space, which is defined by at least two feature point coordinates (f1, f2) based on intensity values ​​assigned to the respective image point and / or neighboring pixels of the respective pixel ( ^^ ^ ^ ௫,௬^ ), - for feature points (P(f1, f2)) which are outside a range for the respective pixel ( ^^ ^ ^ ௫ ,௬^) lie in the feature space through a, in particular closed, curve or surface predetermined feature area (E), to determine in each case a piece of position information (D1-D4) by which the position of the respective feature point (P(f1, f2)) relative to the feature area (E) is characterized, and - to assign the feature points (P(f1, f2)) lying outside the feature area (E) to different defect classes depending on the respectively determined position information (D1-D4), by which different types of defects on the value document (2) are characterized.

2. Sensor according to claim 1, wherein the detection device (3) is configured to detect the electromagnetic radiation transmitted by the value document (2) in the bright field.

3. Sensor according to claim 1 or 2, wherein the detection device (3) is configured to detect the electromagnetic radiation emanating from the value document (2) in different spectral ranges, which lie in the visible, in particular red (R), green (G) and / or blue (B), and / or in the infrared (IR) spectral range.

4. Sensor according to one of the preceding claims, wherein the detection device (3) is configured to detect electromagnetic radiation emanating from the background of the value document (2) to be checked in a spatially resolved manner in the at least two different spectral ranges, wherein a plurality of background pixels (^^ ௬ ^) is obtained, each of which is assigned at least two background intensity values, and the processing device (7) is arranged to assign the background intensity values ​​to a pixel ( ^^ ^ ^ ௫ ,௬^ ) respectively assigned intensity values ​​using the background intensity values ​​of the corresponding background pixels ( ^^ ௬ ^ ) to normalize.

5. Sensor according to one of the preceding claims, wherein the processing device (7) is configured to normalize one or more of the feature point coordinates (f1, f2) of the feature points (P(f1, f2)) based on, preferably normalized, intensity values ​​which are assigned to the respective pixel ( ^^ ^ ^ ௫,௬^ ) and / or neighboring pixels of the respective pixel ( ^^ ^ ^ ௫,௬^ ), in particular by forming a difference, a sum and / or a quotient from the the respective pixel ( ^^ ^ ^௫,௬^ ) and / or by means of a local operation taking into account the intensity values ​​of the neighboring pixels of the respective pixel assigned intensity values.

6. Sensor according to one of the preceding claims, wherein the curve or surface through which the feature area (E) in the feature space for the respective pixel ( ^^ ^ ^ ௫,௬^) is described by a polynomial function, in particular an equation for an ellipse or an ellipsoid or hyperellipsoid, in the feature space.

7. Sensor according to one of the preceding claims, wherein the position information (D1-D4) determined for the feature points (P(f1, f2)) lying outside the feature area (E) contains direction information and / or distance information which characterizes a direction or a distance in which the respective feature point (P(f1, f2)) lies relative to the feature area (E).

8. Sensor according to one of the preceding claims, wherein the different defect classes characterize one or more of the following types of defects on the value document (2): paint splashes, inscriptions subsequently applied to the value document, wear, color loss, creases, scratches, holes, and / or a defect classified as "unknown." 9.Sensor according to one of the preceding claims, wherein the processing device (7) is configured to provide state information concerning the assignment of the feature points (P(f1, f2)) to the. to determine different defect classes and to use the status information to check the value documents.

10. Sensor according to claim 9, wherein the status information includes one or more of the following information: - number of feature points (P(f1, f2)) lying outside the feature range (E); - number of feature points (P(f1, f2)) assigned to a defect class; - proportion of the number of feature points (P(f1, f2)) assigned to a defect class to the number of feature points (P(f1, f2)) lying outside the feature range (E).

11. Sensor according to claim 9 or 10, wherein the processing device (7) is configured to check the value document (7) to be checked with regard to its status on the basis of the status information, e.g.to be classified as “FIT” or “UNFIT”, and / or the processing device (7) is configured to provide the status information of the value document (7) / the value documents and / or information derived from the status information, in particular information relating to the status of the value document (7) / the value documents, for further use, in particular for output to an operator of the sensor and / or a value document processing device (1) equipped with the sensor and / or for storage together with other information relating to the value document (2) and / or during the processing of the value document (2).

12. Value document processing device (1) with at least one processing device (4, 5, 11) for processing, in particular. Transporting and / or checking and / or counting and / or sorting and / or destroying value documents (2), in particular banknotes, and at least one sensor (3, 7, 8) according to one of the preceding claims.

13. A method for checking value documents (2), in particular banknotes, in which - electromagnetic radiation emanating from a value document (2) to be checked, in particular transmitted and / or remitted, is detected in a spatially resolved manner in at least two different spectral ranges, wherein a plurality of pixels (^^ ^ ^ ௫,௬^ ), to which at least two intensity values ​​are assigned, which characterize the intensities of the radiation detected in the at least two different spectral ranges, - several pixels ( ^^ ^ ^ ௫ ,௬^) is assigned a feature point (P(f1, f2)) located in a feature space, which is defined by at least two feature point coordinates (f1, f2) that are based on intensity values ​​assigned to the respective image point ( ^^ ^ ^ ௫,௬^ ) and / or neighboring pixels of the respective pixel ( ^^ ^ ^ ௫ ,௬^ ), - for feature points (P(f1, f2)) which are outside a range for the respective pixel ( ^^ ^ ^ ௫,௬^) lie in the feature space through a, in particular closed, curve or surface of a predetermined feature area (E), in each case a position information (D1-D4) is determined, by which the position of the respective feature point (P(f1, f2)) relative to the feature area (E) is characterized, and - the feature points (P(f1, f2)) lying outside the feature area (E) are assigned to different defect classes depending on the respectively determined position information (D1-D4), by which different types of defects on the value document are characterized.

14. Method according to claim 13, wherein the ) by a feature area (E) specified in the feature space, in particular a closed curve or surface, in which - electromagnetic radiation emanating from several test value documents, in particular classified as "FIT", is recorded in a spatially resolved manner in the at least two different spectral ranges, wherein for each test value document, a plurality of image points are obtained, each of which is assigned at least two intensity values ​​that characterize the intensities of the radiation emanating from the respective test value document and recorded in the at least two different spectral ranges, - several image points of each test value document are each assigned a feature point located in a feature space, which is defined by at least two feature point coordinates that are based on intensity values,which are assigned to the respective pixel and / or neighboring pixels of the respective pixel, and - for several of the pixels, a curve or surface, in particular a closed one, is determined within which the, in particular all, feature points obtained from the test value documents for the respective pixel lie, and this, in particular a closed, curve or surface is specified as the feature area (E).

15. The method according to claim 14, wherein the curve or surface determined for each pixel (^^, ^ ^ ௫ ,௬^ ) given feature range (E) in the feature space by an equation for an ellipse or an ellipsoid or hyperellipsoid is described, which is determined by means of principal axis transformation from the feature points obtained from the test value documents for the respective pixel.

Images