EP2787489A1 - Method for testing printed articles - Google Patents

Abstract

The invention relates to a method for checking printed objects (1), in particular printing units such as banknotes, wherein a digital image (2) is created from the respective object to be tested (1) and assigned to the respective object (1) and the digital image ( 2) is reduced in terms of its resolution, in particular its color depth, the number of its color channels and its local resolution, and thus creates a reduced digital image (3) and the respective object (1) is assigned. According to the invention, a comparison of the reduced digital image (3) with at least one reduced reference image (5) is performed as part of a coarse examination, the reduced digital image (3) in the case of a match exceeding a predetermined first threshold as a defect-free object (1 ) and the examination of the respective item (1) is ended, the item being classified as potentially defective in the case of a match of the reduced digital image (3) with the reduced reference image (5) in the case of a first threshold not exceeding in that an object (1) has been classified as potentially defective, the object is subject to a fine test, the digital image (2) associated therewith being used for the object (1) and this digital image (2) being compared with at least one reference image (4) , wherein the article (1) in the case of a second Schwe Anything that exceeds this value is classified as "error-free", otherwise it is classified as "incorrect".

Description

The invention relates to a method for testing printed objects according to the preamble of claim 1. Furthermore, the invention relates to a device for testing printed objects according to the preamble of patent claim 8.

The invention is preferably used to test freshly printed or fed banknotes before dispensing. Of course, other objects, in particular printing units, can also be checked for conformity with a given reference object or with a reference printing unit.

A variety of different methods are known from the prior art, with which objects, in particular printing units, are compared with reference objects. In this case, a digital image is preferably created both from the respective printing unit to be tested and from the reference printing unit, and the two digital images thus created are compared with one another.

With modern recording methods it is easily possible to create a digital image with very high resolution for each of the objects to be tested. Although a check of this digital image based on a comparison with a reference image is possible in principle, it is time-consuming. In particular, it requires a tremendous computing power, as it were to examine printing units in real time after their production, if the digital image created by this printing unit is used in full resolution for the respective printing unit. A reduction in the resolution or the use of cameras with a lower resolution, conversely, that certain printing errors or low-quality printing units can not be detected at all. Even if the particular resolution used for the test is adjustable within certain limits, an optimal resolution, which allows both a resource-saving and a reliable test, can not be found. Thus, according to the state of the art, it is only possible to provide a test device with increased computing power.

Fig. 1 shows schematically the procedure of the prior art. In a production step, printing units 1 are created by a printing press 15. After production, the printing units 1 by means of a conveyor belt 12 through the Receiving area of a receiving unit 11 transported. The recording unit 11 creates a digital image 2 of the respective printing unit 1.

The digital image 2 is supplied to a computing unit 16. The arithmetic unit 16 optionally reduces the resolution of the digital image 2 of the printing unit 1, so that the arithmetic unit 16 can check the digital images 2 supplied to it at a speed which corresponds to the production rate or the clock rate of the objects 1. This avoids that the arithmetic unit 16 operates too slowly and an examination of the printing units in real time is not possible.

• " The Colour Image Processing Handbook", herausgegeben von S. J. Sangwine und R. E. N. Horne, veröffentlicht von Chapman & Hall, 1998, Seiten 8 bis 23, 66 bis 90 und 376 bis 384
• " Optical Pattern Recognition based on Color Vision Models" von M. S. Millán und M. Corbalán, erhalten am 2. Mai 1995, veröffentlicht von der Optical Society of America, 1995 , Optical Letters Vol. 20, Nr. 16, 15. August 1995, Seiten 1722 bis 1724
• " Opponent Color Space Motivated by Retinal Processing" von Silvio Borer et al, Verfahren des CGIV 2002 -- First European Conference on Color in Graphics, Imaging and Vision (CGIV), Poitiers, Frankreich, April 2002 .
• " Trichromatic opponent color classification" von E. J. Chichilnisky et al, erhalten am 4. März 1998, akzeptiert am 30. November 1998, veröffentlicht von Elsevier Science Ltd. , 1999, 0042-6989/99/$, Vision Research 39 (1999), Seiten 2444 bis 2458
• " Opponent Color Processing Based on Neural Models" von M. Bollmann et al, Advances in Structural and Syntactical Pattern Recognition, 6th International Workshop, SSPR '96, Leipzig, Deutschland, 20. bis 23. August 1996, Lecture Notes in Computer Science 1121, herausgegeben von C. Goos et al (1996 ), J. Hartmanis und J. van Leeuwen
• DE 19744999 A1 der Heidelberger Druckmaschinen AG, veröffentlicht am 23. Juli 1998.

During the test, the respective digital image 2 is compared with a reference image 4, which has the same resolution in terms of number of pixels, color depth and number of image channels as the digital image 2. The reference image 4 corresponds to the expected digital image of a defect-free object 1. The specific nature of the comparison between digital image 2 and reference image 4 is not relevant to the present process. From general knowledge, a variety of methods for examining images are known. By way of example only, reference may be made to the following publications, with which differences between images and reference images can be determined:

• " The Color Image Processing Handbook, edited by SJ Sangwine and REN Horne, published by Chapman & Hall, 1998, pp. 8-23, 66-90, and 376-384
• " Optical Pattern Recognition based on Color Vision Models "by MS Millán and M. Corbalán, received on May 2, 1995, published by the Optical Society of America, 1995 . Optical Letters Vol. 20, No. 16, 15 August 1995, pages 1722 to 1724
• " Opponent Color Space Motivated by Retinal Processing "by Silvio Borer et al, CGIV 2002 method - First European Conference on Color in Graphics, Imaging and Vision (CGIV), Poitiers, France, April 2002 ,
• " Trichromatic opponent color classification "by EJ Chichilnisky et al., Issued March 4, 1998, accepted November 30, 1998, published by Elsevier Science Ltd. . 1999, 0042-6989 / 99 / $, Vision Research 39 (1999), pages 2444 to 2458
• " Opponent Color Processing Based on Neural Models "by M. Bollmann et al, Advances in Structural and Syntactical Pattern Recognition, 6th International Workshop, SSPR '96, Leipzig, Germany, August 20-23, 1996, Lecture Notes in Computer Science 1121, edited by C. Goos et al. (1996 ), J. Hartmanis and J. van Leeuwen
• DE 19744999 A1 Heidelberger Druckmaschinen AG, published on July 23, 1998.

As a result of such a comparison, a comparison result in the form of a comparison value is obtained. This comparison value is compared with a threshold, which indicates which comparison result is just considered to be tolerable. If the determined comparison result is worse than the threshold value, the digital image 2 and the object 1 from which the digital image originates are classified as defective; otherwise the digital image 2 and the object 1 assigned to it are classified as error-free.

Depending on the result of the classification by the arithmetic unit 16, the position of a switch 13 is switched, so that those objects 1b which have been identified as defective are rejected by the objects 1a identified as faultless. Faulty objects 1b are destroyed or recycled or, if there are no serious defects, put on the market as inadequate goods. In the present example, the articles 1a, 1b are stored and made available in separate containers 17a, 17b according to their classification via separate conveyor belts 12a, 12b downstream of the switch 13.

The aim of the invention is to provide a method which, on the one hand, permits a precise examination and, on the other hand, has a lower consumption of resources than the methods mentioned above.

The invention solves this problem in a method of the type mentioned above with the characterizing features of claim 1.

1. a) von dem jeweiligen zu prüfenden Gegenstand ein Digitalbild erstellt und dem jeweiligen Gegenstand zugeordnet wird,
2. b) das Digitalbild hinsichtlich seiner Auflösung, insbesondere seiner Farbtiefe, der Anzahl seiner Farbkanäle sowie seiner örtlichen Auflösung, reduziert wird und derart ein reduziertes Digitalbild erstellt und dem jeweiligen Gegenstand zugeordnet wird,
   vor, dass
3. c) im Rahmen einer Grobprüfung ein Vergleich des reduzierten Digitalbilds mit zumindest einem reduzierten Referenzbild durchgeführt wird,
4. d) das reduzierte Digitalbild im Falle einer einen vorgegebenen ersten Schwellenwert übersteigenden Übereinstimmung als übereinstimmend erkannt wird, wodurch der Gegenstand als fehlerfrei erkannt und die Überprüfung des Gegenstands beendet wird,
5. e) der Gegenstand im Falle einer den ersten Schwellenwert nicht übersteigenden Übereinstimmung des reduzierten Digitalbilds mit dem reduzierten Referenzbild als potentiell fehlerhaft eingestuft wird, und
6. f) für den Fall, dass ein Gegenstand als potentiell fehlerhaft eingestuft wurde, der Gegenstand einer Feinprüfung unterzogen wird, wobei für den Gegenstand das ihm jeweils zugeordnete Digitalbild herangezogen wird und dieses Digitalbild mit zumindest einem Referenzbild verglichen wird, wobei der Gegenstand im Falle einer einen zweiten Schwellenwert übersteigenden Übereinstimmung als "fehlerfrei" andernfalls als "fehlerhaft" klassifiziert wird.

The invention provides a method for testing printed objects, in particular printing units such as banknotes, wherein

1. a) a digital image is created by the respective object to be tested and assigned to the respective object,
2. b) the digital image is reduced in terms of its resolution, in particular its color depth, the number of its color channels and its local resolution, and such a reduced digital image is created and assigned to the respective object,
   before that
3. c) a comparison of the reduced digital image with at least one reduced reference image is performed as part of a coarse examination,
4. d) if the reduced digital image is detected as coincident in the case of a match exceeding a predetermined first threshold value, whereby the article is recognized as free of errors and the examination of the object is terminated,
5. e) the object is classified as potentially defective in the case of a match of the reduced digital image with the reduced reference image that does not exceed the first threshold value, and
6. (f) in the case where an object has been classified as potentially defective, subjecting the object to a fine test, using the digital image associated with the object and comparing that digital image with at least one reference image, the object in the case of a second match exceeding threshold as "error free" otherwise classified as "faulty".

This procedure has the significant advantage that a fine test is only required if the rough check previously carried out has not yielded a clear result. As a result, the effort associated with the fine inspection can be reduced to a small proportion of the objects, such as printing units.

• die zu untersuchenden Gegenstände vor und/oder während ihrer Aufnahme relativ zu der die Aufnahme durchführenden Aufnahmeeinheit, insbesondere mit einer Transportvorrichtung, vorzugsweise mit einem Förderband, befördert werden,
• die Gegenstände nach ihrer Grobprüfung hinsichtlich des Prüfungsergebnisses, insbesondere mit einer ersten Weiche, separiert und die als potentiell fehlerhaft erkannten Gegenstände in einen Zwischenspeicher überführt werden, und
• die im Zwischenspeicher befindlichen Gegenstände nach Durchführung der Feinprüfung einer zweiten Weiche zugeführt und entsprechend ihrer Klassifikation separiert werden.

A particularly simple integration into an automated production process or testing process, in particular printing process, can take place by

• the objects to be examined are conveyed before and / or during their picking up relative to the picking unit performing the picking, in particular with a transport device, preferably with a conveyor belt,
• the objects after their coarse examination with respect to the test result, in particular with a first turnout, separated and transferred as potentially defective detected objects in a cache, and
• the items in the buffer are fed after performing the fine test a second switch and separated according to their classification.

In order to avoid an unnecessary consumption of resources in the case of evidently defective objects, it can be provided that in the coarse examination objects whose assigned reduced digital images have a third threshold below the matching reduced reference image are recognized and classified as "in any case erroneous".

In order to eliminate such defective items quickly from the production or inspection, it can be provided that objects which are clearly detected as faulty in the coarse test, separated or discarded and not fed to the fine test, and / or only those items whose comparison no final assignment as faulty or error-free, the fine test to be supplied.

A simple and efficient implementation provides that the coarse test and the fine test respectively identical or different test algorithms are used, which are adapted to the respective resolution of the images to be compared.

In order to preclude premature classification of defective printing units as error-free, it can be provided that the first threshold value used for the coarse test for the distinction between error-free and potentially defective reduced digital images indicates a higher match than the second threshold used in the fine test.

In order to save computing time and storage requirements during the rough check, it can be provided that
In the creation of the reduced digital image and / or the reduced reference image, the resolution of the digital image and / or the reference image is reduced by reducing the local resolution of the respective image and / or by combining individual color channels, in particular by
the respective image is converted from color to black and white or grayscale and / or
by reducing the resolution depth of the individual color channels.

In order to make it possible to adapt the reduced digital image to the reduced reference image and to exclude transport-related effects such as rotations, fluttering, etc., it can be provided that the reduced digital images are fed to a piecewise affine transformation before the coarse examination and brought into coincidence with the reduced reference image.

In order to allow adaptation of the digital image to the reference image and to exclude transport-related effects such as rotations, flutter and moisture-related curvature of the object or printing unit during recording, it can be provided that the digital images before the fine test of a non-linear equalization, and possibly one before the fine test successive rotation, distortion or piecewise affine transformation supplied and gebacht with the reference image to cover.

In order to be able to make a concrete statement about the nature of the error that has occurred during production, in particular during printing, it can be provided that in the case of the Grobprüfung a subdivision of the class of detected as potentially defective objects in the manner of the potentially existing error made in several subclasses and assigned each detected as a faulty digital image in each case one or more of these subclasses. Through this information obtained in the course of the error classification, it is possible to avoid certain errors in the production of new objects, in particular in the printing of printing units, and to adapt parameters during production.

In order to achieve a uniform utilization of the rough test and the fine test, it can be provided that in the sequential testing of a plurality of objects, the first threshold value is set such that a proportion of 1 / n, in particular 3% to 10%, of the objects is detected to be potentially erroneous, where n corresponds to the ratio between the memory size of the digital image and the memory size of the reduced digital image. By doing so, it can be avoided that the fine test is interrupted at a time during the examination procedure for lack of test jobs.

In order to avoid that objects of inferior quality are considered faultless, it can be provided that a minimum value of conformity is specified and objects whose conformity with the reduced reference image falls below this minimum value are in any case recognized as potentially defective.

In order to enable a further distribution of the utilization to a number of test stages running separately from each other, it can be provided that those objects that have been regarded as potentially defective in the course of the rough test are subjected to one or more sequential intermediate tests, in each case an intermediate classification is carried out and only such items are supplied to the respective subsequent intermediate test or the fine test, which were not clearly detected as "error-free" or "faulty".

In order to avoid an overflow of the buffer, it can be provided that, when the buffer is overflowed, at least one object located in the buffer is classified as defective and / or discarded.

The invention solves the above object in a device of the type mentioned above with the characterizing features of claim 8.

1. a) eine Transporteinheit, insbesondere ein Förderband, zum Transport der Gegenstände,
2. b) eine Aufnahmeeinheit, deren Aufnahmebereich auf die auf der Transporteinheit befindlichen Gegenstände gerichtet ist und die für den Fall, dass sich ein Gegenstand in ihrem Aufnahmebereich befindet, ein Digitalbild erstellt,
3. c) eine der Aufnahmeeinheit nachgeschaltete Reduktionseinheit, der die erstellten Digitalbilder zugeführt werden und die das jeweilige Digitalbild hinsichtlich seiner Auflösung, insbesondere seiner Farbtiefe und/oder der Anzahl seiner Farbkanäle und/oder seiner örtlichen Auflösung, reduziert und derart ein reduziertes Digitalbild erstellt und dem jeweiligen Gegenstand zuordnet,
4. d) eine erste Recheneinheit zur Durchführung einer Grobprüfung der Gegenstände durch Vergleich des reduzierten Digitalbilds mit einem reduzierten Referenzbild und anschließender Grobklassifikation des Vergleichsergebnisses als "potentiell fehlerhaft" oder "fehlerfrei",
5. e) eine der Transporteinheit nachgeschaltete und von der ersten Recheneinheit gesteuerte erste Weiche zur Separation der Gegenstände gemäß ihrer Grobklassifikation,
6. f) eine zweite Recheneinheit zur Durchführung einer Feinprüfung der in der Grobprüfung als "potentiell fehlerhaft" klassifizierten Gegenstände durch Vergleich des dem jeweiligen Gegenstand zugeordneten Digitalbilds mit einem Referenzbild und anschließender Feinklassifikation des Vergleichsergebnisses als "fehlerhaft" oder "fehlerfrei", und
7. g) eine der ersten Weiche nachgeschaltete und von der zweiten Recheneinheit gesteuerte zweite Weiche zur Separation der als "potentiell fehlerhaft" Gegenstände gemäß ihrer Feinklassifikation.

In a device according to the invention for testing printed objects, in particular printing units such as banknotes, it is provided:

1. a) a transport unit, in particular a conveyor belt, for transporting the objects,
2. b) a receiving unit whose receiving area is directed towards the objects located on the transport unit and which creates a digital image in the event that an object is in its receiving area,
3. c) a reduction unit connected downstream of the recording unit to which the digital images produced are supplied and which reduces the respective digital image with respect to its resolution, in particular its color depth and / or the number of color channels and / or its local resolution, and thus creates a reduced digital image and the respective one Assigns an object,
4. d) a first arithmetic unit for performing a coarse examination of the objects by comparing the reduced digital image with a reduced reference image and then rough classification of the comparison result as "potentially defective" or "error-free",
5. e) a first switch downstream of the transport unit and controlled by the first computer unit for separating the items according to their rough classification,
6. f) a second arithmetic unit for performing a fine check of the objects classified as "potentially defective" in the coarse test by comparing the digital image associated with the respective object with a reference image and subsequent fine classification of the comparison result as "faulty" or "faultless", and
7. g) a second switch connected downstream of the first switch and controlled by the second computer unit for the purpose of separating the articles "potentially defective" according to their fine classification.

This will only require a fine check if the previous rough check has not produced a clear result. As a result, the effort associated with the fine test can be reduced to a small proportion of the objects or printing units.

• das reduzierte Digitalbild mit einem vorab vorgegebenen reduzierten Referenzbild vergleicht und einen der Übereinstimmung zwischen diesen reduzierten Bildern entsprechenden ersten Vergleichswert ermittelt,
• diesen ersten Vergleichswert mit einem vorgegebenen ersten Schwellenwert vergleicht,
• den Gegenstand als "fehlerfrei" klassifiziert, wenn der erste Vergleichswert eine höhere Übereinstimmung indiziert als der erste Schwellenwert,
• den Gegenstand andernfalls als "potentiell fehlerhaft" klassifiziert, und
• das so ermittelte Klassifikationsergebnis an ihrem Ausgang zur Verfügung hält.

In order to achieve an advantageous separation after the coarse testing, it can be provided that
the first arithmetic unit for rough checking and rough classification of a reduced digital image present at its input

• compares the reduced digital image with a predefined reduced reference image and determines a first comparison value corresponding to the correspondence between these reduced images,
• compares this first comparison value with a predetermined first threshold value,
• classifies the item as "error free" if the first comparison value indicates a higher match than the first threshold,
• otherwise classify the item as "potentially defective", and
• the result of the classification thus obtained is available at its exit.

For excreting articles of very poor quality, provision can be made for the first arithmetic unit to classify the object as "in any case erroneous" for coarse testing and rough classification of a reduced digital image present at its input if the first comparison value indicates a lower match than a predefined third threshold value.

• das Digitalbild mit einem vorab vorgegebenen Referenzbild vergleicht und einen der Übereinstimmung zwischen diesen Bildern entsprechenden zweiten Vergleichswert ermittelt,
• diesen zweiten Vergleichswert mit einem vorgegebenen zweiten Schwellenwert vergleicht,
• den Gegenstand als "fehlerfrei" klassifiziert, wenn der zweite Vergleichswert eine höhere Übereinstimmung indiziert als der zweite Schwellenwert,
• den Gegenstand andernfalls als " fehlerhaft" klassifiziert, und
• das so ermittelte Klassifikationsergebnis an ihrem Ausgang zur Verfügung hält.

In order to achieve an advantageous separation after the fine test, it can be provided that the second arithmetic unit for fine testing and fine classification of a present at its input digital image

• compares the digital image with a predetermined reference image and determines a second comparison value corresponding to the match between these images,
• compares this second comparison value with a predetermined second threshold value,
• classifies the item as "error free" if the second comparison value indicates a higher match than the second threshold,
• otherwise classify the item as "defective", and
• the result of the classification thus obtained is available at its exit.

• dass die erste Weiche einen Eingang und zumindest zwei Ausgänge aufweist,
• dass der Eingang der ersten Weiche der Transporteinheit nachgeschaltet ist,
• dass der erste Ausgang der ersten Weiche über eine zweite Transporteinrichtung zu einem Behälter für fehlerfreie Gegenstände führt, und
• dass der zweite Ausgang der ersten Weiche über eine dritte Transporteinrichtung zur zweiten Weiche führt,
• dass die erste Weiche Gegenstände von ihrem Eingang zu ihrem ersten Ausgang transportiert, wenn ihr das Grobklassifikationsergebnis "fehlerfrei" zugeführt ist, und
• dass die erste Weiche Gegenstände von ihrem Eingang zu ihrem zweiten Ausgang transportiert, wenn ihr das Grobklassifikationsergebnis "potentiell fehlerhaft" zugeführt ist.

A preferred arrangement of the first turnout provides

• that the first switch has an input and at least two outputs,
• that the input of the first turnout of the transport unit is connected downstream,
• that the first exit of the first turnout leads via a second transport means to a container for faultless articles, and
• the second exit of the first turnout leads to the second turnout via a third transport device,
• that the first switch transports items from their entrance to their first exit when the gross classification result is "error free", and
• that the first switch transports items from their entrance to their second exit when the gross classification result "potentially defective" is fed to them.

• dass die zweite Weiche einen Eingang und zumindest zwei Ausgänge aufweist,
• dass der Eingang der zweiten Weiche dem Ausgang der ersten Weiche für "potentiell fehlerhafte" Gegenstände nachgeschaltet ist,
• dass der erste Ausgang der zweiten Weiche über eine vierte Transporteinrichtung zu einem weiteren Behälter für fehlerfreie Gegenstände führt,
• dass der zweite Ausgang der zweiten Weiche über eine fünfte Transporteinrichtung zu einem weiteren Behälter für fehlerhafte Gegenstände führt,
• dass die zweite Weiche Gegenstände von ihrem Eingang zu ihrem ersten Ausgang transportiert, wenn ihr das Feinklassifikationsergebnis "fehlerfrei" zugeführt ist, und
• dass die zweite Weiche Gegenstände von ihrem Eingang zu ihrem zweiten Ausgang transportiert, wenn ihr das Feinklassifikationsergebnis " fehlerhaft" zugeführt ist.

A preferred arrangement of the second switch provides

• that the second switch has an input and at least two outputs,
• that the input of the second switch is connected downstream of the output of the first switch for "potentially defective" objects,
• that the first exit of the second switch leads via a fourth transport device to a further container for faultless articles,
• that the second exit of the second switch leads via a fifth transport device to a further container for defective articles,
• that the second switch transports items from their entrance to their first exit when the fine classification result is "error free", and
• the second diverter transports items from their entrance to their second exit when the fine classification result is "erroneous".

A preferred arrangement of the first switch provides that for the removal of objects with very poor quality, the first switch has a third output, which leads via a transport device provided for this purpose to a container for at least faulty items
and that the first diverter transports items from their entrance to their third exit if the gross classification result is "in any case erroneous".

• von der ersten Weiche ausgehende Gegenstände zwischenlagert und
• diese Gegenstände auf Anfrage der zweiten Recheneinheit, insbesondere nach der Durchführung der Feinprüfung des betreffenden Gegenstands, über die dritte Transporteinrichtung der zweiten Weiche zuführt.

In order to buffer the irregular occurrence of potentially defective objects and to avoid overloading of the second arithmetic unit, provision can be made for a, in particular FIFO, intermediate memory to be provided in the third transport device

• intermediately stored from the first turnout objects and
• these items on request of the second processing unit, in particular after the completion of the fine examination of the object in question, via the third transport means of the second switch supplies.

A simple and efficient implementation provides that for the coarse test and the fine test respectively programs with the same test algorithms are stored in the first arithmetic unit and the second arithmetic unit, which are adapted to the respective resolution of the digital images and reference images to be compared.

In order to rule out premature classification of faulty printing units as error-free, it can be provided that the first threshold used for the coarse test for the distinction between error-free and potentially erroneous pictures indicates a higher match than the second threshold used in the fine check.

Preferred embodiments of the invention and variants thereof are illustrated in more detail with reference to the following drawing figures.

Fig. 2 schematically shows a first embodiment of a test device according to the invention. Fig. 3 shows schematically the distribution of the quality of printing units and made by the coarse and fine check separation of the printing units based on the print quality.

In Fig. 2 a first embodiment of a test device according to the invention is shown. This testing device is preceded by a production unit 25, which creates the objects to be tested 1. Alternatively, it is also possible that the objects to be tested are located in a memory and are supplied to the respective test with a time delay.

The objects 1 are transported via a transport device 22, which in the present case is designed as a conveyor belt, and arrive on their transport through the receiving area 21 'of a receiving unit 21. The receiving unit 21 creates a digital image 2 from the respective object 1. A reduction unit 29 is created Basis of the digital image 2 a digital image reduced in its resolution 3.

The reduction of the resolution can be carried out according to different criteria, wherein only the reduction of the amount of data of the reduced digital image 3 with respect to the digital image 2 is required. In particular, a reduction of the resolution can be achieved by reducing the number of pixels. Alternatively or additionally, the number of image channels can also be reduced; for example, the reduced digital image 3 can be created by weighting the individual color values associated with a pixel and assigning them to the positionally corresponding pixel of the reduced digital image 3. It can also be provided that the color depth is reduced, ie that the discretization, with which individual color values are resolved, is coarsened. For example, from a digital image 2, for each color channel and pixel each 16 bits available a reduced digital image 3 can be created for which only 4 or 8 bits are available per color channel and pixel.

It is also possible to combine individual ways of reducing the resolution of the digital image 2 with each other, i. for example, to halve the number of pixel rows and pixel columns, and in addition to reduce the number of three 16-bit color channels to an 8-bit resolved black and white color channel. In this way a reduction of the image information by a factor of 24 is achieved.

The reduction of the resolution of the digital image 2 and the concomitant creation of the reduced digital image 3 is performed in the present embodiment of the reduction unit 29, which is the recording unit 21 downstream. The reduced digital image 3 is fed to a first computing unit 26a, which carries out a rough checking and rough classification of the object 1 or of the reduced digital image 3 that it has produced.

Alternatively, however, the digital image 2 can also be fed to the first arithmetic unit 26a, the reduction unit 29 being designed as part of the first arithmetic unit 26a, which generates the reduced digital image 3 on the basis of the digital image 2 supplied to it.

Furthermore, a reference image 4 is available for testing the object 1, which in terms of its resolution, its color depth and the channels used corresponds to the digital image 2 of a defect-free object 1 which is expected during the recording. This reference image 4 can be predetermined or created by taking a good object. Alternatively, the reference image 4 can also be created by a statistical method, for example by averaging or median formation, of the values of digital images of several error-free reference objects.

Furthermore, it is also possible to set a minimum and a maximum tolerable color or brightness value for each pixel, or to specify a maximum permissible deviation value analogously for each pixel.

Based on the reference image 4, a reduced reference image 5 is created. The reduction of the image information of the reduced reference image 5 is carried out in the same way as in the creation of the reduced digital images 3 based on the captured digital images 2. The reduced reference image 5 is stored in the first processing unit 26a or otherwise kept available for retrieval. Each reduced digital image 3 arriving at the first arithmetic unit 26a or created by the first arithmetic unit 26a is compared with the reduced reference image 5. For the comparison, the types of image comparison mentioned in connection with the prior art come into question. In particular, for comparison of a reduced digital image 3 with the reduced reference image 5, comparison methods can be used, as described in the cited documents. As a result of the comparison method in the coarse examination, a first comparison result V _{1 is obtained} in the form of a digital numerical value. This first comparison result V ₁ is compared with a first threshold value S ₁ . If the first comparison result V ₁ indicates a worse match of the reduced digital image 3 with the reduced reference image 5 than a comparison result corresponding to the first threshold value S ₁ , the respective reduced digital image 3 and thus also the object 1 from which the respective reduced digital image 3 originates , recognized as potentially defective and viewed. Otherwise, the article 1 is recognized as a defect-free article 1a and viewed.

In this particular embodiment of the invention, in addition, a third threshold value S _{3 is} predetermined, which corresponds to a very poor comparison result. If the comparison result V ₁ obtained on the basis of the comparison of the reduced digital image 3 with the reduced reference image 5 indicates a worse match than the third threshold S ₃ , then the respective object 1 from which the reduced digital image 3 originates is referred to as a "faulty" object 1c viewed. Such a sorting out, at any rate, of defective articles 1c by the first arithmetic unit 26a merely constitutes a special development of the invention for relieving the fine test yet to be displayed, and basically does not need to be provided.

In the manner described above, a rough classification into one of the classes "error-free" and "potentially defective" is made by the first arithmetic unit 26a for each of the items 1, the items optionally also being assigned to further classes, such as "in any case erroneous" can be. In addition, subclassifications such as "potential misprints of the species gradient" or "potential misprints of the type doctor blade" are also possible as subclasses of the class "potentially defective". These sub-classifications can be used to initiate feedback to the production unit 25, if present, and to carry out corresponding readjustments.

Depending on the classification by the first arithmetic unit 26a, a first diverter 23a is actuated by the latter. If the respective object 1 reaches the first switch 23a, the switch 23a is set depending on its classification, so that the object 1 reaches one of the transport units 22a, 22b, 22c assigned to the respective classes specified by the classification. The first switch 23a as well as the second switch 23b, shown later, each have a point setting unit which brings the respective switch 23a, 23b respectively into the position predetermined by the respective calculating unit 26a, 26b and permits separation of the supplied articles 1, 1b.

The first switch 23a has an input and at least two outputs. The input of the first switch 23a is connected downstream of the transport unit 22. The first exit of the first switch 23a leads via a second transport device 22a to a container 27a for faultless articles 1a. The second output of the first switch 23a leads via a third transport device 22b to the second switch 23b. The first switch 23a is, as already mentioned, controlled by the first computing unit 26a. The first diverter 23a transports items 1 from their entrance to their first exit when the gross classification result is fed "error-free". If the article 1 is classified as error-free, it is conveyed via a first transport unit 22a to a container 27a for defect-free articles 1a and held therein.
The first diverter 23a transports items 1 from their entrance to their second exit, when the gross classification result "potentially erroneous" is supplied to them. If the item 1 is classified as "potentially defective", it is conveyed via a second transport unit 22b to a buffer 24 for potentially defective items 1b and kept available there. The buffer 24 has the task of buffering the irregular arrival of objects 1b classified as "potentially defective" and of making such objects 1b available on a regular basis, without an overloading of the second arithmetic unit 26b during the fine inspection of the objects 1b.

In this preferred embodiment of the invention, the first switch 23a has a third output and transports articles 1 from their input to their third output when the rough classification result is supplied to them "in any case erroneous". The objects 1c recognized as "in any case defective", provided that the classification carried out by the arithmetic unit 26a provides such a class, conveyed via a transport device 22c provided for this purpose to a container 27c for in any case defective articles 1c and stored therein, possibly discarded or destroyed. As already mentioned, these devices, in particular a container 27c for in any case defective articles 1c as well as the transport device 22c leading to it, are not a mandatory component of the invention.

The first diverter 23a thus has a number of outputs corresponding to the number of classes provided by the classification of the first arithmetic unit 26a. Depending on the assignment of the respective object 1 to one of the classes, the position of the first switch 23a is selected accordingly, so that the respectively differently classified objects are separated according to their classification.

In order to obtain a clear association between the respective object 1 and the reduced digital image 3 and the determined first comparison result V ₁ and the classification during the control of the first switch 23a by the first computer 26a, at least the respective first comparison results V ₁ stored in a FIFO memory of the first arithmetic unit 26a. If an object 1 now arrives at the first switch 23a, the oldest value written to the FIFO memory is read out of it and the first switch 23a is set in accordance with this value. The value is then deleted from the FIFO memory.

As part of a fine test, the objects 1b classified as "potentially defective" are rechecked by a second arithmetic unit 26b. As already mentioned, these potentially defective articles 1b are kept available in a buffer 24 connected downstream of the second outlet of the first switch 23b.

Furthermore, the digital images 2 associated with the objects classified as potentially defective are transferred to the second arithmetic unit 26b and in this case according to the FIFO principle, i. in the order of their arrival, worked off. The order of execution of the fine examination of the digital images 2 corresponds to the order of the objects 1b in the buffer.

Alternatively, it is also possible for the second arithmetic unit 26b to be supplied with the digital images 2 of all objects 1 to be tested. In this case, the arithmetic unit 26b discards the digital images 2 of those not classified as "potentially defective" Objects 1a, 1c and only processes the digital images 2 of the objects 1b classified as "potentially defective".

The second arithmetic unit 26b carries out the fine classification on the basis of a comparison of the digital image 2 produced by the recording unit 21 with a reference image 4. For this purpose, a reference image 4 is provided which corresponds in resolution, the channels used and the color depth to the expected digital image of a defect-free object 1.

In the present embodiment of the invention, the second arithmetic unit 26b compares the digital image 2 of the subject 1 with the reference image 4 in the same way as the first arithmetic unit 26a compares the reduced digital image 3 with the reduced reference image 5. The only difference between the comparison process carried out on the first and the second arithmetic unit 26a, 26b is that the second arithmetic unit 26b determines a second comparison value V ₂ with the full image resolution. This second comparison value V ₂ is compared with a second threshold value S ₂ and a fine classification is carried out, with the respective object ₁ b being indicated as a "defective" object ₁ c in a comparison result V ₂ which indicates a worse match than the second threshold value S ₂ . otherwise it is classified as a "faultless" item 1a '.

Basically, comparison methods can be used for the coarse and the fine test, which differ only in terms of the resolution used.

However, it is particularly advantageous if several different test methods are available for the fine test, which are adapted to different types of errors. If a subclassification for potential error types is made in the rough check, depending on the subclassification in the course of the rough classification, a fine check specially tailored to the presumed error can be carried out. For example, if an item is considered "potentially defective" in the coarse classification, with a subclassification "potential misprints of the squeegee type" being made, the fine test may use a test method that specifically investigates the presence of squeegee strips. In addition, the rough classification can be added to the estimated position of the respective error and used as a parameter of fine classification.

If the fine classification of a potentially defective object 1b has ended, the objects 1b fed to the buffer 24 are supplied on request to the second processing unit 26b to a second switch 23b, which is also controlled by the second processing unit 26b.

In the manner described above, the second arithmetic unit 26b carries out a fine classification for each of the items 1 into one of the classes "error-free" and "erroneous", wherein the items may optionally also be assigned to further classes. In addition, subclassifications such as "Color Error Fault Type Error" or "Squeegee Style Fault Type Error" are also possible as subclasses of the "Faulty" class. These sub-classifications can be used to initiate feedback to the production unit 25, if present, and to carry out corresponding readjustments.

According to the respective fine classification, the second switch 23b separates the articles 1b conveyed from the buffer 24 according to their classification into "defect-free" articles 1a 'and "defective" articles 1c' by guiding them to respective different transport units 28a, 28c.

In the present exemplary embodiment, the second switch 23b has one input and two outputs, it being additionally possible to provide a further output for each possible classification result of the fine classification. The input of the second switch 23b is connected downstream of the output of the first switch 23a for "potentially defective" articles 1 via the third transport unit 22b.

The first exit of the second switch 23b leads via a fourth transport device 28a to a further container 27a 'for faultless articles 1a'. The second output of the second switch 23b leads via a fifth transport device 28c to a further container 27c 'for defective articles 1c'. The second switch 23b transports items 1 from their entrance to their first exit when the fine classification result is supplied "error-free" by the second processing unit 26b. The second diverter 23b transports items 1 from their entrance to their second exit when the fine classification result is "erroneously" supplied from the second arithmetic unit 26b.

Articles 1a 'classified as such "free of defects" are transported via the fourth transport unit 28a to the further container 27a' for defect-free articles 1a 'and stored therein and made available. Articles 1c 'classified as "defective" in this way are transported via the fifth transport unit 28c to the container 27c' for defective articles 1c 'and stored therein and made available.

In order to obtain a clear association between the respective object 1b and the digital image 2 and the determined second comparison result V ₂ and the classification during the control of the second switch 23b by the second computing unit 26b, at least the respective second comparison results V ₂ in FIG another FIFO memory of the second arithmetic unit 26b stored. If an object 1b now arrives at the second switch 23b, the value first written to the further FIFO memory is read out of it and the second switch 23b is set according to this value. The value is then deleted from the further FIFO memory.

In Fig. 3 schematically the setting of the first and the second threshold value S ₁ , S _{2 is} shown. Fig. 3 indicates on the ordinate the probability that an item 1 has a certain match with an ideal reference item. The abscissa shows the respective agreement, the comparison value V ₁ , V ₂ = 0 indicating an ideal match. The larger the respective comparison value V ₁ , V ₂ , the worse is the correspondence between the respective digital image ₂ , 3 and the associated reference image ₄ , 5.

The abscissa represents a threshold value T which corresponds to the quality that an object 1 may have in the worst case in order to be classified as error-free. Due to the examination of the coincidence of the respective digital image 2, 3 with the reference image 4, 5 by the first and second arithmetic unit 26a, 26b, the comparison value V ₂ determined by the second arithmetic unit 26b has a greater accuracy than that of the first Arithmetic unit 26a determined first comparison value V ₁ .

The inaccuracy with which the coarse test is affected is represented by the first interval I ₁ , whereby the qualitatively worst boundary, here the upper one, of this first interval I ₁ corresponds to the limit value T ₁ . This ensures that no item 1 in the first classification is classified as error-free, the one has poorer quality than indicated by the T limit. For this purpose, the first threshold value S ₁ to a value S ₁ = T - set I _1/2, which is shifted by the respective half the interval width of the first interval _I1 with respect to the threshold T. For alternative measures of comparison, in which a high first comparison value V ₁ indicated good agreement, applies to S ₁ = T + I _1/2.

Similarly, the second threshold S ₂ used in the fine test is set to a value S ₂ = T - I _2/2 , which is shifted by the respective half-interval width of the first interval I ₂ relative to the threshold value T. For alternative comparison measures in which a high second comparison value V ₂ indicates a good match, S ₂ = T + I _2/2 applies accordingly. Since the fine test is much more accurate, the second interval I _{2 has} a smaller interval width than the first interval I ₁ . Due to the greater inaccuracy in the determination of the first comparison value V ₁ compared to the determination of the second comparison value V ₂ , the second threshold S _{2 is} closer to the threshold T than the first threshold S ₁ , whereby a more precise good / bad statement is obtained during the fine test ,

The width of the first interval I ₁ in this embodiment corresponds to the error spread of the comparison made during the coarse test. In this exemplary embodiment, the width of the second interval I ₂ corresponds to the lower error spread of the comparison made during the fine test compared to the rough test. In the present example, it was assumed that the intervals lie symmetrically about the respective threshold values S ₁ , S ₂ , S ₃ .

The obliquely shaded surface area under the curve indicates the maximum expected proportion of the objects to be tested in the course of the fine test. The closer the interval limits of the first interval I _{1 are} selected, the fewer items need to be subjected to the fine check, however, the more complex the coarse check will be.

If, according to the special embodiment of the invention presented at the beginning, a classification of the objects 1 into "error-free", "potentially defective" and "in any case erroneous" is carried out during the coarse test, then a third threshold value S ₃ can be selected as shown below. In order to avoid that reduced digital images 2, 3 of defect-free objects 1, which have been identified as potentially defective by the coarse examination or in which V ₁ > S ₁ , are discarded, the third threshold value S ₃ = T + I ₁ ' / 2 set so that it at least the Interval width I ₁ '/ 2 worse, here greater, is selected as the threshold T. For alternative comparison measures in which a high first comparison value V ₁ indicates a good match, S ₃ = T - I ₁ ' / 2 applies accordingly. The width of the third interval I ₁ 'corresponds in this embodiment, the width of the first interval I _1, that is, the error diffusion of the comparison made in the rough examination. If, during the coarse examination, a classification is also carried out for any defective objects 1c, this proportion is reduced by the vertically hatched area under the curve. In order to avoid that error-free objects 1 are detected as "in any case faulty", the interval widths of the first and third intervals are defined as follows: I ₁ '≥ I ₁ .

Particularly advantageously, the following alternative adaptive procedure for creating a uniform utilization between the first arithmetic unit 26a and the second arithmetic unit 26b can be used. For example, if it is known that the fine test of a digital image 2 lasts about n times as long as the rough test of a reduced digital image 3, the first threshold S ₁ 'can alternatively be chosen so that on average every n-th reduced digital image 2 is potentially defective is recognized. For this purpose, the adaptive first threshold value S ₁ 'may be set in an interval between an optimum quality S _max and the first threshold value S ₁ . For the comparative measure used in which a low first comparison value V ₁ indicated good agreement, applies to S _{max <1} S '<S ₁ = T - I _1/2. For alternative comparison measurements, in which a high first comparison value V ₁ indicates a good agreement, applies to S _max> S ₁ '> S ₁ = T + I _1/2. Here S ₁ corresponds to a minimum value of agreement.

If too few reduced digital images 3 identified as potentially faulty or classified, the adaptive first threshold S ₁ 'within the predetermined interval is adapted to a greater consistency indexed value, so that more reduced digital images are classified as potentially faulty 3 after this change. Conversely, if the number of reduced digital images 3 identified as potentially defective is too large, the adaptive first threshold value S ₁ 'is adapted so that less reduced digital images 3 are classified as potentially defective after this change.

In the present exemplary embodiment, the first threshold value S _{1 is} adaptively determined by specifying that only 4% of the digital images classified by the first arithmetic unit 26 a are classified as "potentially defective". If, as before If the number of objects 1b classified as "potentially defective" is approximately 4% of the total objects 1 to be inspected, then for a real-time capable fine classification with the arithmetic unit 26b for each of the objects 1b approximately 25 times the time for fine checking is available ,

In order to allow a particularly accurate comparison, in a particular embodiment of the invention, the reduced digital images 3 are supplied before the coarse examination of a piecewise affine transformation and brought into coincidence with the reduced reference image 5. Such a procedure is known from the prior art and is performed by the first computing unit 26a.

Likewise, to improve the comparison, the digital images 2 may be fed before the fine check of a non-linear equalization, and possibly a before or after rotation, distortion or piecewise affine transformation, and be covered with the reference image 4 to cover. This is likewise known from the prior art and is performed by the second arithmetic unit 26b.

Of course, it is also possible to perform several sequential intermediate tests between the rough test and the fine test. Items which have been considered potentially defective in the course of the rough examination will be successively subjected to the individual intermediate tests. Each of the intermediate tests is followed by an intermediate classification, with only items which are not clearly identified as "error-free" or "defective" being supplied to the respectively subsequent intermediate test or the fine test.

Claims (15)

Method for testing printed objects (1), in particular printing units such as banknotes, wherein a) a digital image (2) is created by the respective object to be tested (1) and assigned to the respective object (1), b) the digital image (2) with respect to its resolution, in particular its color depth, the number of its color channels and its local resolution, is reduced and such a reduced digital image (3) created and the respective object (1) is assigned, characterized in that c) a comparison of the reduced digital image (3) with at least one reduced reference image (5) is carried out as part of a coarse examination, d) recognizing the reduced digital image (3) as belonging to a defect-free object (1) in the event of a match exceeding a predetermined first threshold, and terminating the examination of the respective object (1), e) the object is classified as potentially defective in the case of a match of the reduced digital image (3) with the reduced reference image (5) which does not exceed the first threshold value, and f) in the event that an object (1) has been classified as potentially defective, the object is subjected to a fine test, using the digital image (2) assigned to it for the object (1) and including this digital image (2) with at least one Reference image (4) is compared, wherein the object (1) is classified in the case of a second threshold exceeding agreement as "error-free", otherwise as "faulty". Method according to claim 1, characterized in that - That the objects to be examined (1) before and / or during their recording relative to the receiving unit performing receiving unit (11), in particular with a transport device (22), preferably conveyed with a conveyor belt, - that the objects (1) after their coarse examination with respect to the test result, in particular with a first switch (23a), separated, and the potentially erroneously recognized objects (1) are transferred to a buffer (24), and - That the objects (1) located in the intermediate memory (24) are fed to a second switch (23b) after the fine test has been carried out and separated according to their classification. Method according to one of the preceding claims, characterized in that a) in the coarse examination objects (1) whose associated reduced digital images (3) have a third threshold (S ₃ ) below matching with the reduced reference image (5) are recognized as "at least faulty" and classified, and in particular b) objects that are clearly detected as faulty in the coarse test, separated or separated and not passed to the fine test, and / or only those items (1) whose comparison does not allow final assignment as faulty or error-free, the fine test are supplied. Method according to one of the preceding claims, characterized in that a) identical or different test algorithms are used in the coarse test and the fine test, which are adapted to the respective resolution of the images to be compared (2, 3, 4, 5), and / or b) the first threshold value (S ₁ ) used for the rough examination for the distinction between error-free and potentially defective reduced digital images (4) indicates a higher match than the second threshold value (S ₂ ) used in the fine test and / or c) in the creation of the reduced digital image (3) and / or the reduced reference image (5) the resolution of the digital image (2) and / or the reference image (4) is reduced by the local resolution of the respective image is reduced and / or by combining individual color channels, in particular by c1) the respective image is converted from color to black and white or grayscale and / or c2) by reducing the resolution depth of the individual color channels. Method according to one of the preceding claims, characterized in that a) the reduced digital images (3) are fed before the coarse examination of a piecewise affine transformation and brought into coincidence with the reduced reference image (5), and / or b) the digital images (2) are fed before the fine test of a non-linear equalization, and if necessary, a rotation, distortion or piecewise affine transformation before the fine test and covered with the reference image (4), and / or c) in the coarse examination a subdivision of the class of potentially detected as defective objects (1) made in a plurality of subclasses according to the type of potentially existing error and each recognized as faulty digital image (2) each one or more of these subclasses is assigned, in particular d) in the case of the fine classification for the objects identified as potentially defective, depending on the respective subclass to which the object has been assigned, a test method assigned to the respective subclass and adapted to the respective potentially existing error is carried out. Method according to one of the preceding claims, characterized in that a) in the sequential testing of a plurality of objects, the first threshold value is set such that a proportion of 1 / n, in particular 3% to 10%, of the objects is recognized as potentially defective, where n is the ratio between the memory size of the digital image ( 2) corresponds to the memory size of the reduced digital image (3), in particular b) a minimum value of conformity is specified and objects whose conformity with the reduced reference image falls below this minimum value are in any case recognized as potentially defective. Method according to one of the preceding claims, characterized in that (a) subjecting those items considered potentially defective in the course of the rough examination to one or more intermediate sequential tests each carrying out an intermediate classification and only those items which are not clearly assigned to the subsequent intermediate or fine inspection; "error-free" or "faulty" were detected and / or b) at overflow of the buffer (24) at least one in the buffer (24) befindlicher object classified as defective and / or discarded. Device for testing printed objects (1), in particular of printing units such as banknotes, for example a) a transport unit (22), in particular a conveyor belt, for transporting the objects (1), b) a receiving unit (21) whose receiving area (21 ') on the transport unit (22) located objects (1) is directed and in the event that an object (1) in its receiving area (21') is located , created a digital image (2), c) one of the recording unit (21) downstream reduction unit (29) to which the created digital images (2) are supplied and the respective digital image (2) in terms of its resolution, in particular its color depth and / or the number of its color channels and / or its local resolution, reduced, and thus creates a reduced digital image (3) and assigns it to the respective object (1), d) a first arithmetic unit (26a) for performing a coarse examination of the objects (1) by comparison of the reduced digital image (3) with a reduced reference image (5) and subsequent rough classification of the comparison result (V ₁ ) as "potentially defective" or "error-free" . e) a first switch (23a) connected downstream of the transport unit and controlled by the first computer unit (26a) for separating the articles (1) according to their rough classification, f) a second arithmetic unit (26b) for performing a fine check of the objects (1) classified as "potentially defective" by comparison of the digital image (2) associated with the respective object with a reference image (4) and subsequent fine classification of the comparison result (V ₂ ) as "faulty" or "error free", and g) a second switch (23b), which is connected downstream of the first switch and controlled by the second computer unit (26a), for the purpose of separating the articles (1b) which are classified as "potentially defective" according to their fine classification. Apparatus according to claim 8, characterized in that a) the first arithmetic unit (26a) for rough checking and rough classification of a reduced digital image (3) present at its input comparing the reduced digital image (3) with a predefined reduced reference image (5) and determining a first comparison value (V ₁ ) corresponding to the correspondence between these reduced images (3, 5), comparing this first comparison value (V ₁ ) with a predetermined first threshold value (S ₁ ), the item (1) is classified as "error-free" if the first comparison value (V ₁ ) indicates a higher match than the first threshold value (S ₁ ), otherwise classify the item (1) as "potentially defective", and - the thus determined classification result is available at its output, and / or b) the first arithmetic unit (26a) for rough checking and rough classification of a reduced digital image (3) present at its input classifies the object (1) as "in any case erroneous" if the first comparison value (V ₁ ) indicates a lower match than a predetermined third one Threshold value (S ₃ ). Device according to one of claims 8 or 9, characterized in that the second arithmetic unit (26b) for fine checking and the fine classification of a present at its input digital image (2) comparing the digital image (2) with a predetermined reference image (4) and determining a second comparison value (V ₂ ) corresponding to the agreement between these images (2, 4), comparing this second comparison value (V ₂ ) with a predetermined second threshold value (S ₂ ), the item (1) is classified as "error-free" if the second comparison value (V ₂ ) indicates a higher match than the second threshold value (S ₂ ), - classify the item (1) otherwise as "defective", and - the thus determined classification result is available at its output. Device according to one of claims 8 to 10, characterized in that the first switch (23a) has an input and at least two outputs, - That the input of the first switch (23a) of the transport unit (22) is connected downstream, - That the first output of the first switch (23a) via a second transport means (22a) leads to a container (27a) for error-free articles (1a), - That the second output of the first switch (23a) via a third transport means (22b) leads to the second switch (23b), - That the first switch (23a) transports items (1) from their entrance to their first exit, when the gross classification result is fed "error-free", and - That the first switch (23a) transports items (1) from their entrance to their second exit when the gross classification result "potentially defective" is supplied to them. Device according to one of claims 8 to 11, characterized in that the second switch (23b) has an input and at least two outputs, - That the input of the second switch (23b) is connected downstream of the output of the first switch (23a) for "potentially defective" objects (1), - That the first output of the second switch (23b) via a fourth transport means (28a) to another container (27a ') for error-free articles (1a') leads, and - That the second output of the second switch (23b) via a fifth transport means (28c) to another container (27c ') for defective articles (1c') leads, and - That the second switch (23b) transports articles (1) from their entrance to their first exit, when the fine classification result is fed "error-free", and - That the second switch (23b) transports articles (1) from their entrance to their second exit, when the fine classification result is fed "incorrectly". Device according to one of claims 8 to 12, characterized in that the first switch (23a) has a third output which leads via a transport device (22c) provided for this purpose to a container (27c) for in any case faulty articles
and that the first switch (23a) transports items (1) from their entrance to their third exit when the gross classification result is fed to them "in any case erroneous". Device according to one of claims 8 to 13, characterized in that in the third transport means (22b) a, in particular FIFO, temporary store (24) is provided which - Intermediate from the first switch (23a) outgoing objects (1b) and - This objects (1b) on request of the second processing unit (26b), in particular after performing the fine examination of the object in question (1b), via the third transport means (22b) of the second switch (23b) supplies. Device according to one of claims 8 to 14, characterized in that a) for the coarse test and the fine test in each case programs with the same test algorithms in the first arithmetic unit (26 a) and the second arithmetic unit (26 b) are stored, the respective resolution of the digital images to be compared (2, 3) and reference images (4, 5 ), and / or b) the first threshold value (S ₁ ) used for the coarse examination for the distinction between error-free and potentially faulty images indicates a higher match than the second threshold value (S ₂ ) used in the fine test, and / or c) the first arithmetic unit (26a) carries out a subclassification of the class "potentially defective" depending on the type of the assumed error and transmits the respectively determined subclass to the second arithmetic unit (26b), and the second arithmetic unit (26b) passes a fine test depending on the subclass communicated to it, it selects a fine check method adapted to the suspected error associated with the respective subclass.

Images