EP3680106B1 - Mn-detection in a printed image - Google Patents

Description

The present invention relates to a method for checking the print quality of an inkjet printing machine using a camera and a computer.

The invention lies in the technical field of digital printing.

When operating inkjet printing machines, there are specific printing errors that are specific to this type of printing machine. The most common are so-called white line errors, which occur when individual print nozzles of the inkjet print heads used deviate from their desired standard behavior. If this deviation exceeds certain limits, the printing nozzles in question are generally deactivated since they disturb the printed image. However, a printing nozzle that is deactivated in this way then generates a corresponding whiteline error. It is so named because it is most prominent in the case of a monochrome printed surface, since in this case the underlying print substrate, which is usually white, is exposed. If, on the other hand, a light printing color (e.g. opaque white) is printed on a dark-looking background, the errors appear as so-called dark line errors. But even in multicolored image areas where several print nozzles from different print heads print the individual color separations one on top of the other, the failure or deactivation of a print nozzle involved leads to corresponding color distortions in the print image to be produced. Since the print nozzles emit ink in a line in the printing direction, the print defect that is produced is also linear, which is where the term white/darkline print defect comes from.

The reasons for the occurrence of these deviations in the operation of printing nozzles can be of a diverse nature. A main problem here is drying of the ink in the event that the corresponding print head is not used for a long time and is not properly stored in the idle state. The drying ink clogs the nozzle outlet opening and thus leads to a deviating pressure point of the print nozzle in question or, in extreme cases, even to complete failure. In each In this case, the print nozzle no longer prints exactly where the actual print point should be and the print strength also deviates from the actually desired standard values. In addition to dried ink, the ingress of dust particles and similar dirt can also cause white line errors.

A number of approaches are known from the prior art for finding these white line errors. The most common is certainly the printing of test patterns and the detection of the whitelines via automated recording and evaluation of these test patterns. However, this approach has the disadvantage that the printing of the test pattern causes waste depending on the position and size on the printing substrate. For this reason, there are methods which examine the print image produced itself and then detect white line errors that occur from this print image. This also has the advantage that only those whitelines, or the printing nozzles that cause them, are detected that are really disruptive to the print image that is currently being generated.

The German patent application DE 2017 220361 A1 discloses such a method for detecting and compensating for failed printing nozzles in an inkjet printing machine using a computer, which involves printing a current print image, recording the printed print image using an image sensor and digitizing the recorded print image using the computer, adding digitized color values of the recorded print image of each column over the entire print image height and dividing the summed up color values by the number of pixels to obtain a column profile, subtracting an optimized column profile without failed printing nozzles from the original column profile to obtain a difference column profile, setting a threshold for maximum values exceeding which one out, applying this threshold for maximum values to the difference column profile, whereby in the resulting column profile each maximum marks a out of print nozzle, and compensating for the marked out print nozzles in the subsequent printing process as steps of the method.

The disadvantage of this method, however, is that it cannot be implemented robustly in practice. The method is based on the fact that there are very little differences between one reference image and a camera image. But that is not always the case in practice. Reasons for this are, for example, incorrectly calibrated cameras, a non-optimal or outdated white balance, different types of paper or suboptimal colors in the printing unit. In addition, areas in the printed image that are as monochromatic as possible are used for the detection of the whitelines, which is why the method can only be used to a limited extent in the case of printed images that do not have such areas.

From the US patent US 9,944,104 B2 a whiteline inspection system is known. A simple limit value comparison is suggested here for detecting the whitelines, which assumes that the motif to be checked is homogeneous at this point. In the case of an image to which this condition does not apply, it is proposed that the signal be generated by subtracting a locally aligned reference image generated from prepress print data. However, a difference image calculation is still necessary.

The patent application EP3300907A1 On the other hand, describes how a whiteline detection system can achieve an increased quality by using different methods, depending on the printing situation, in particular to avoid weak and therefore uncritical whitelines being detected, or poorly compensated whitelines that are invisible to the human eye as a defect be recognized. As in US 9,944,104 B2 a reference image generation step is also necessary here in order to generate reference data for finding whitelines, which one would like to do without.

The US patent application U.S. 2012/092409 A1 also discloses a system and method according to the preamble of claim 1 for detecting missing inkjets in an inkjet imaging system. The system and method detect missing inkjets in an inkjet imaging system. The system creates digital images of printed documents that do not contain any test pattern data. The digital images are processed to detect light stripes, and the positions of the light stripes are correlated to the ink jet positions in the printheads. The identification of the ink color associated with the correlated ink jet positions is then obtained through an analysis of color separated images and/or color errors.

The object of the present invention is therefore to find a method for determining printing errors in a printing process of an inkjet printing machine which is more efficient and determines printing errors, in particular white line errors, better and more reliably than the methods known from the prior art.

The object is solved by a method for determining printing errors in a printing process, carried out in an inkjet printing machine for processing a print job by a computer, with printed products produced during the printing process being recorded and digitized using a camera system that generated camera image is fed to a detection algorithm on the computer, if printing errors are detected, a message is sent to a machine controller, which then, if necessary, ejects the printed product via a waste gate, which is characterized in that the detection algorithm separates color sections of the camera images and detects the printing errors in the color sections , links images of the individual color separations to a candidate image, filters the candidate image and finally enters the remaining detected printing errors in a list and sends this to the printing press. The core of the method according to the invention is therefore to determine the printing errors directly from the generated camera image of the recorded and digitized printed product. The printing errors are detected directly in the separated color separations, since the printing errors can be detected more easily here than in the composite camera image. However, it is important that the printing errors in the generated camera image can be recognized at all. For example, if the resolution of the generated camera image is too low, the information about the corresponding printing errors is lost and the entire detection algorithm comes to nothing. It is also important to note that the camera usually delivers RGB images, so that a separation of the individual color separations of the camera image generated naturally delivers individual RGB color separations and not CMYK color separations, which correspond to the color space of the inkjet printing machine used. However, this does not pose a problem for the method according to the invention, since what matters above all is the exact position of the corresponding printing errors, or that printing errors that impair the printing quality are reliably detected at all. Which color separation in the machine color space, i.e. which ink and thus which print head, is affected by the nozzle failure can be determined by the computer through appropriate color space transformations. To improve the detection algorithm, following the detection in the color separations, the images of the individual color separations are linked again to form a common candidate image and this image is then subjected to further filtering to ensure that only printing errors that lead to waste are detected accordingly. All columns in the candidate image that contain a detected printing error are marked in order to be able to detect the printing nozzles causing the printing errors later.

Advantageous and therefore preferred developments of the method result from the associated subclaims and from the description with the associated drawings.

A preferred development of the method according to the invention is that the printing errors are white line or dark line errors caused by defective printing nozzles of the inkjet printing machine. The algorithm is therefore primarily intended to recognize the whiteline errors already described, since these printing errors in particular reduce the print quality of the printing process to such an extent that waste occurs.

A further preferred development of the method according to the invention is that in a further method step, before sending to the printing machine, the computer filters out pseudo-whiteline or darkline errors from the list of whiteline or darkline errors by using a specific test method . It is important that no false positive errors are output by the detection algorithm. In particular, thin, light lines in the print image to be generated, such as barcodes, are very susceptible to being marked as pseudo-whitelines. Therefore, in a further process step, the detection algorithm should use specific tests to check whether the detected white line is really a real white line in order to rule out the possibility that desired print image components are incorrectly detected as white line errors and thus cause additional waste unintentionally.

A further preferred development of the method according to the invention is that the computer determines the causative defective printing nozzles from the list of the remaining detected white line or dark line errors and, depending on this, compensates the white line or dark line errors using suitable compensation methods. Although the actual goal of the method according to the invention is to precisely identify printed products in the form of printed sheets that have such quality-reducing whiteline errors that make the printed sheet produced a waste sheet, the information about the whiteline errors determined by the detection algorithm can of course be used also be used to to determine the causal defective printing nozzles and to have them compensated by a suitable compensation method. Compensation for the defective print nozzles finally allows the relevant inkjet printing machine to continue to be used to process the current print job without changing the print head.

A further preferred development of the method according to the invention is that the computer for the specific test method creates a reference image from preliminary data of the print job, applies the detection algorithm to this reference image and thus gains knowledge about either received candidates for pseudo white line or dark line errors and removes them from the list of whiteline or darkline defects or areas in the camera image with probable pseudo whiteline or darkline defects to which the detection algorithm is then not applied. The easiest way to detect pseudo-white lines is to create a reference image from good data, e.g. the preliminary stage data, and then check whether the structure found that was detected as a white line is also located in this reference image. If this is the case, it is logically a pseudo white line. With the realization that you have found one, you can deal with it in two ways. One can easily remove the found pseudo whiteline error from the list, which is certainly the easiest way to do it. However, if you want to avoid the same pseudo-whiteline error being found again by the detection algorithm in the ongoing method according to the invention, it is best to exclude the area in the camera image in which this pseudo-whiteline error occurred from the detection according to the invention .

A further preferred development of the method according to the invention is that the computer creates the reference image in several sizes and/or resolutions, applies the detection algorithm several times to the various reference images and summarizes and uses the knowledge gained therefrom. This procedure increases the reliability of both the detection algorithm for the specific marking of whitelines and for the determination of pseudo-whiteline errors.

A further preferred development of the method according to the invention is that the algorithm is not applied to such areas, or results from such areas are excluded, which are characterized by a high variation of the gray values in a limited, local environment in the reference image. Such areas, such as barcodes, are particularly susceptible to the detection of pseudo-whiteline or darkline errors and must therefore be excluded from the algorithm's check.

A further preferred development of the method according to the invention is that the list of white line or dark line errors is created via column totals in the filtered candidate image by applying a limit value to the respectively determined column total in the candidate image. True spurious whiteline/darkline errors typically span a larger area of the captured camera image. In order to prevent even very small, short misfires of the individual printing nozzles from leading to a detection of a printing error, although these may not be disturbing at all or they are even pseudo-whiteline/darkline errors, which is the case with very short whiteline errors is very probable, only those print columns in the candidate image are marked in which the determined print error exceeds a certain limit value.

A further preferred development of the method according to the invention is that the candidate image of the individual color separations is linked by the computer using a mathematical OR operation. This type of composition of the individual color extracts to form the candidate image has proven to be the most suitable from a computational point of view.

A further preferred development of the method according to the invention is that the filtering of the candidate image is carried out by the computer using morphological operations. In particular, this allows very short printing errors or white lines to be filtered out, which are usually pseudo-white lines anyway or do not affect the print quality of the printed product or printed sheet so much that a waste decision has to be made here.

A further preferred development of the method according to the invention is that the detection algorithm is applied multiple times by the computer to the generated camera image, with the method being parameterized differently in each case in order to detect differently pronounced dark line or white line errors and the results of all color separations of all applications of the procedure can be logically linked with one another. In addition to the multiple application of the detection algorithm to multiple reference images, which is an optional method step of the method according to the invention, the detection algorithm can also be applied multiple times to the generated camera image. In particular, this increases the accuracy of the detection algorithm when filtering out pseudo whiteline or darkline errors, but is also positive for the hit accuracy when finding real whiteline or darkline errors.

A further preferred development of the method according to the invention is that for each of the differently parameterized applications of the method, the camera image is previously limited to a maximum gray value in each pixel. The advantage of this is that bright outliers in paper white areas that could falsify the mean are filtered out.

A further preferred development of the method according to the invention is that the candidate image of a color channel is generated by dividing the image into horizontal strips, with each strip being reduced to a line signal by suitable averaging of each of its columns, in which whitelines or darklines are then a special search method can be sought and each line evaluated in this way results in a line of the whiteline candidate image. This is an important feature of the method according to the invention, since the white/dark line detection using the detection algorithm in these strips is more efficient than if the algorithm had to work with the overall image.

A further preferred development of the method according to the invention is that the computer uses the white or dark line search method to search for a dark or

Whiteline at a position by evaluating a constrained neighborhood around the pixel of interest in the line signal. The classification of whether a found error is really a real white line or dark line error is done by evaluating the immediately adjacent pixels. Only then can a pseudo-whiteline or darkline error be ruled out.

A further preferred development of the method according to the invention is that the search method first convolves the line signal with different filter kernels and converts the results into logical signals by comparing them with possibly different limit values, which are then converted into a white or dark line signal with the aid of a logical link. candidate line signal is converted.

The invention as such and constructively and/or functionally advantageous developments of the invention are described in more detail below with reference to the associated drawings using at least one preferred exemplary embodiment. In the drawings, corresponding elements are given the same reference numbers.

Figur 1:

ein Beispiel des Aufbaus einer Bogen-Inkjet-Druckmaschine

Figur 2:

ein Beispiel eines zur Druckinspektion verwendeten Bilderfassungssystems

Figur 3:

ein Beispiel eines erfassten Kamerabildes

Figur 4:

ein Streifen des erfassten Kamerabildes

Figur 5:

ein Streifen des erfassten Kamerabildes mit markierten Whitelines

Figur 6:

ein vergrößerter Ausschnitt mit markierten Whitelines aus dem Streifen des erfassten Kamerabildes

Figur 7

aus Bildstreifen zusammengesetztes Bild mit markierten Whiteline-Kandidaten

Figur 8:

markierte Whiteline-Bereiche in einem Kamerabild

Figur 9:

Störung des Spaltenmittelwerts durch helle Papierweiß-Bereiche oder einzelne helle Pixel

Figur 10:

der schematische Ablauf des erfindungsgemäßen Verfahrens

The drawings show:

Figure 1:

an example of the structure of a sheet-fed inkjet printing press

Figure 2:

an example of an image acquisition system used for print inspection

Figure 3:

an example of a captured camera image

Figure 4:

a strip of the captured camera image

Figure 5:

a strip of the captured camera image with marked whitelines

Figure 6:

an enlarged section with marked whitelines from the strip of the captured camera image

figure 7

Image composed of image strips with marked whiteline candidates

Figure 8:

Marked whiteline areas in a camera image

Figure 9:

Disturbance of the column mean value due to light paper white areas or individual light pixels

Figure 10:

the schematic sequence of the method according to the invention

The area of application of the preferred embodiment variant is an inkjet printing machine 7. An example of the basic structure of such a machine 7, consisting of feeder 1 for feeding the printing substrate 2, usually a printed sheet 2, into the printing unit 4, where it is printed by the print heads 5 is printed, up to delivery 3, is in figure 1 shown. This is a sheet-fed inkjet printing machine 7 which is controlled by a control computer 6 . As already described, when this printing press 7 is in operation, failures of individual printing nozzles in the print heads 5 in the printing unit 4 can occur. The result is white/dark lines or, in the case of a multicolored print, distorted color values. An example of such a white/dark line 9 in a captured camera image 8 is shown in figure 3 shown.

1. 1. Nach dem Druck wird der bedruckte Bogen mit Hilfe eines Kamerasystems 10, welches Teil eines Inline- Bilderfassungssystems 12 sein kann, digitalisiert. Figur 2 zeigt ein Beispiel für ein solches Bilderfassungssystem 12, welches das erfindungsgemäße Verfahren einsetzt. Es besteht aus mindestens einem Bildsensor 10, üblicherweise einer Kamera 10, welche in die Inkjet-Druckmaschine 7 integriert ist. Die mindestens eine Kamera 10 nimmt die von der Druckmaschine 7 erzeugten Druckbilder 13 auf und sendet die Daten an einen Rechner 6, 9 zur Auswertung. Dieser Rechner 6, 9 kann ein eigener separater Rechner 9 sein, z.B. ein oder mehrere spezialisierte Bildverarbeitungsrechner 9, oder auch mit dem Steuerungsrechner 6 der Druckmaschine 7 identisch sein. Mindestens der Steuerungsrechner 7 der Druckmaschine 7 besitzt ein Display 11, auf welchem die Ergebnisse der Bildinspektion dem Anwender 8 angezeigt werden. Bevorzugt wird für das im Folgenden beschriebene Verfahren der Bildverarbeitungsrechner 9 eingesetzt, auf welchem dann der Bildverarbeitungsalgorithmus läuft, welcher das erfindungsgemäße Verfahren ausführt. Das so erzeugte Kamerabild 13 hat dabei eine geringere Auflösung als der Druck. Üblich ist eine Kameraauflösung von 670dpi bei einer Druckauflösung von 1200dpi. Die Auflösung und Optik muss so gewählt sein, dass White-/Darklines 14 als 1-2 Kamerapixel breite, vertikale, aufgehellte Streifen erscheinen. Ist die Auflösung zu hoch, so kann das Bild 13 zunächst mit bekannten Verfahren der Bildverarbeitung auf passende Auflösung reduziert werden, insbesondere pyramidale Bildrepräsentationen können sich hier als nützlich erweisen.
2. 2. Das Kamerabild 13 wird einem im Weiteren näher beschriebenen White-/Darkline-Detektionsalgorithmus zugeführt. Zusätzlich kann es parallel noch weiteren Auswertungen zugeführt werden.
3. 3. Werden White-/Darklines 13 vom Detektionsalgorithmus erkannt, so werden diese vom Bildverarbeitungsrechner 9 an den Steuerungsrechner 6 der Druckmaschine 7 gemeldet, welche dann, zusammen mit anderen Daten der Druckmaschine 7, eine Makulaturentscheidung trifft und den bedruckten Bogen 2 eventuell über eine Makulaturschleuse ausschleust.
4. 4. Optional können die gefundenen White-/Darklines 14 einer genaueren Auswertung unterzogen werden, um die defekte Düse zu identifizieren und diese Information zur Kompensation der defekten Düse zu nutzen.

In contrast to methods known from the prior art, the embedding of the detection method for the white/dark lines 14 in the overall flow of the printing process is different in the method according to the invention and also no longer requires interaction of the printer 8 . The overall sequence of the method in a first, preferred embodiment is shown schematically in figure 10 shown:

1. 1. After printing, the printed sheet is digitized using a camera system 10, which can be part of an inline image acquisition system 12. figure 2 shows an example of such an image acquisition system 12, which uses the method according to the invention. It consists of at least one image sensor 10, usually a camera 10, which is integrated into the inkjet printing machine 7. The at least one camera 10 records the printed images 13 generated by the printing machine 7 and sends the data to a computer 6, 9 for evaluation. This computer 6, 9 can be its own separate computer 9, for example one or more specialized image processing computers 9, or it can be identical to the control computer 6 of the printing press 7. At least the control computer 7 of the printing machine 7 has a display 11 on which the results of the image inspection are shown to the user 8 . The image processing computer 9 is preferably used for the method described below, on which the image processing algorithm then runs, which executes the method according to the invention. The camera image 13 generated in this way has a lower resolution than the pressure. A camera resolution of 670dpi with a print resolution of 1200dpi is usual. The resolution and optics must be selected in such a way that white/dark lines 14 appear as 1-2 camera pixel wide, vertical, brightened stripes. If the resolution is too high, the image 13 can first be reduced to a suitable resolution using known image processing methods; pyramidal image representations in particular can prove useful here.
2. 2. The camera image 13 is supplied to a white/darkline detection algorithm, which is described in more detail below. In addition, it can be fed to other evaluations in parallel.
3. 3. If white/dark lines 13 are recognized by the detection algorithm, these are reported by the image processing computer 9 to the control computer 6 of the printing press 7, which then, together with other data from the printing press 7, makes a waste decision and possibly transfers the printed sheet 2 via a waste lock ejects.
4. 4. Optionally, the white/dark lines 14 found can be subjected to a more precise evaluation in order to identify the defective nozzle and to use this information to compensate for the defective nozzle.

This process makes it clear that continuous processing of the camera images 13 is essential for the overall system 12 to function.

In contrast to the prior art, the already mentioned algorithm for white/dark line detection is only applied to the camera image 13 here. figure 3 shows an example of a printed sheet 2 with captured camera images 13, one of which has a white/darkline error 14. Optionally, in a further embodiment variant, however, additional, downstream filtering can also be carried out with the aid of a reference image. This will be explained in more detail later in the description.

1. 1. Separiere die RGB-Farbauszüge und für jeden Farbauszug C separat:
   • 1.1. Teile das Kamerabild 13 in ca. 1-10mm hohe Streifen 15, sh. Figur 4
   • 1.2. Jeder Streifen 15 wird in Laufrichtung des Bogens 2, also in y-Richtung, gemittelt. Es ergibt sich ein Signal I_s(x) für den s-ten Streifen 15
   • 1.3. In jedem Streifen 15 werden White-/Darklines 14 separat detektiert, indem für jede x-Position ein Wahrheitswert berechnet wird:
     • 1.3.1. Als optionaler Schritt werden Pixel mit Grauwerten I_C,s(x) > G_max nicht berücksichtigt, da White-/Darklines 14 in hellen Bildbereichen nicht sichtbar werden.
     • 1.3.2. WLC(x,s) = (I_C,s (x)- I_C,s(x-1) > L) and ((I_C,s(x)- I_C,s(x+1)>L) or (I_C,s(x) - IC,s (x+2) > L)) or (I_C,s(x) - I_C,s(x-2) > L) and ((I_C,s(x) - I_C,s(x+1) > L)) Dieser Ausdruck prüft, ob eine 1 oder 2 Pixel breite White-/Darkline 14 mit einer Aufhellung um mehr als L Graustufen besteht und schließt Kanten im Bild auf effektive Weise aus. Figur 5 und 6 zeigen jeweils einen Bildstreifen 15 mit erkannten White-/Darklines 14, welche mit einer entsprechenden Markierung 16 sind. Figur 6 zeigt dabei einen vergrößerten Ausschnitt 17 aus dem Streifen 15 mit Markierung 16 und White-/Darkline 14.
   • 1.4. So ergibt sich ein schwarz-weiß-Bild WLCC (x,y), in dem alle White-/Darkline-Kandidaten 14 vermerkt sind.
2. 2. Die Bilder WLC_C(x,y) der einzelnen Farbauszüge werden ODER-Verknüpft und ergeben dann ein einziges Kandidatenbild WLC(x,y) 21, welches beispielhaft in Figur 7 dargestellt wird. Es zeigt ein aus Bildstreifen 15 zusammengesetztes Bild 21 mit markierten White-/Darkline-Kandidaten 14.
3. 3. Das Bild WLC(x,y) 21 kann nun noch mit morphologischen Operationen gefiltert werden. So erlaubt ein Erodieren mit einem Strukturelement SE der Form:
   • 0 1 0
   • 0 1 0
   • sehr kurze White-/Darklines 14 auszufiltern. Die Höhe von SE kann variabel eingestellt werden und erlaubt es so eine Mindestlänge der zu detektierenden White-/Darklines 14 voreinzustellen.
4. 4. Die gleiche Auswertung, wie in den Schritten 1-3 beschrieben kann in einem weiteren Ausführungsbeispiel parallel auf ein eventuell vorhandenes Referenzbild, welches als RGB-Bild, direkt vom RIP erzeugt wird, angewendet werden. Die sich daraus ergebenden White-/Darkline-Kandidaten WLC_REF(x,y) 14 markieren Bereiche des Druckbildes, in denen falsch-positive White-/Darkline-Detektionen durch das Kundenmotiv wahrscheinlich sind. Diese Bereiche sollten aus WLC(x,y) 21 aus dem Kamerabild 13 entfernt werden. Dazu werden zunächst die Bereiche in WLC_REF(x,y) mit Hilfe einer morphologischen Dilatation ausgeweitet. Dies entspricht einer Glättung von WLC_REF(x,y). Danach wird WLC(x,y) 21 durch WLC_REF(x,y) gefiltert:
   WLC(x,y) ← WLC(x,y) and (not WLC_REF(x,y))
5. 5. Abschließend werden alle Spalten C_WL in WLC(x,y) 21 detektiert, die eine White-/Darkline 14 enthalten. Dies kann über einen Grenzwert minWLPerColumn an die Spaltensumme erfolgen und zwar als Kodierung: keine White-/Darkline = 0, White-/Darkline = 1 in WLC(x,y) 21, d.h. zählen der als White-/Darkline 14 markierten Einträge in WLC(x,y) 21: $$\mathit{CWL}=\left\{x|{\displaystyle {\sum }_{\mathrm{y}}\mathrm{WLC}\left(x,y\right)>\mathrm{minWLPerColumn}}\right\}$$
   

The basis of the detection algorithm presented here is the subdivision of the captured camera image 13 into horizontal strips 15, 15a, 15b. The algorithm comprises the following steps:

1. 1. Separate the RGB separations and for each separation C separately:
   • 1.1. Divide the camera image 13 into strips 15 approx. 1-10mm high, sh. figure 4
   • 1.2. Each strip 15 is averaged in the running direction of sheet 2, ie in the y-direction. A signal I _s (x) results for the sth strip 15
   • 1.3. In each strip 15, white/dark lines 14 are detected separately by calculating a truth value for each x-position:
     • 1.3.1. As an optional step, pixels with gray values I _C,s (x) > G _max are not taken into account since white/dark lines 14 are not visible in bright image areas.
     • 1.3.2. WLC( x , s ) = (I _C,s (x)- I _C,s (x-1) > L) and (( I _C,s (x)- I _C,s (x + 1)>L ) or (I _C,s (x) - IC,s (x + 2) > L)) or (I _C,s (x) - I _C,s (x-2) > L) and ((I _{C ,s} (x) - I _C,s (x + 1) > L )) This expression checks whether there is a 1 or 2 pixel wide white/dark line 14 with a lightening by more than L gray levels and reveals edges in the image effective way. Figure 5 and 6 each show an image strip 15 with recognized white/dark lines 14, which have a corresponding marking 16. figure 6 shows an enlarged section 17 from the strip 15 with marking 16 and white/dark line 14.
   • 1.4. This results in a black-and-white image WLCC( x,y ) in which all white/darkline candidates 14 are noted.
2. 2. The images WLC _C ( x , y ) of the individual color separations are OR-linked and then result in a single candidate image WLC(x,y) 21, which is shown as an example in figure 7 is pictured. It shows an image 21 composed of image strips 15 with marked white/darkline candidates 14.
3. 3. The image WLC(x,y) 21 can now be filtered with morphological operations. This allows eroding with a structural element SE of the form:
   • 0 1 0
   • 0 1 0
   • to filter out very short white/dark lines 14 . The height of SE can be adjusted variably and thus allows a minimum length of the white/dark lines 14 to be detected to be preset.
4. 4. In a further exemplary embodiment, the same evaluation as described in steps 1-3 can be applied in parallel to a reference image which may be present and which is generated directly by the RIP as an RGB image. The resulting white/darkline candidates WLC _REF ( x , y ) 14 mark areas of the printed image in which false-positive white/darkline detections by the customer motif are likely. These areas should be removed from WLC(x,y) 21 from the camera image 13. For this purpose, the areas in WLC _REF ( x , y ) are first expanded using a morphological dilatation. This corresponds to a smoothing of WLC _REF ( x , y ). After that, WLC(x,y) 21 is filtered by WLC _REF ( x , y ):
   WLC( x , y ) ← WLC( x , y ) and (not WLC _REF ( x , y ))
5. 5. Finally, all columns C _WL in WLC(x,y) 21 that contain a white/dark line 14 are detected. This can be done using a minWLPerColumn limit value for the column sum, specifically as a coding: no white/darkline = 0, white/darkline = 1 in WLC( x , y ) 21, ie count the entries marked as white/darkline 14 in WLC( x , y ) 21: $$\mathit{CWL}=\left\{x|{\displaystyle {\sum }_{\mathrm{y}}\mathrm{WLC}\left(x,y\right)>\mathrm{minWLPerColumn}}\right\}$$
   

• So können die nachgelagerten Filter variiert werden.
• Die Anzahl der der White-/Darkline-Kandidaten 14 muss eine Mindestanzahl pro Spalte erreichen um als White-/Darkline 14 gemeldet zu werden
• Für den Helligkeitswert der Pixel wird ein Maximalwert definiert, damit sehr helle Pixel nicht den Mittelwert verfälschen. Denn eine White-/Darkline 14 hat keine sehr hellen Pixel in einem 670 dpi Kamerabild 13. D.h. alle Grauwerte im Bild > 50 werden auf 50 begrenzt;
• Es wird geprüft, ob an den relevanten Stellen im Referenzbild starke Strukturen vorhanden sind, die schon im Referenzbild zu einer White-/Darkline-ähnlichen Struktur führen und deshalb im Kamerabild 13 ausgeblendet werden müssen. Dafür muss das Referenzbild nicht in der vollen Auflösung vorliegen, da nur eine grobe Abschätzung benötigt wird ob der Teilbereich des Referenzbildes strukturiert oder homogen ist; sh. Schritt 4.
• Das oben beschriebene Verfahren kann zur schritthaltenden Anwendung auf einer Grafikkarte (GPU) als Rechenbeschleuniger implementiert werden.
• Der beschriebene Detektionsalgorithmus kann als Teilkomponente des Bilderfassungssystems 12, welches die Bildinspektion durchführt, implementiert werden. Aus dem WLC(x,y)-Bild 21 können dann noch Daten für einen Report an den Bediener 8, bzw. Kunden abgeleitet werden, indem zusammenhängende Flächen (Blobs) im Bild 13 erkannt werden und in einem Übersichtsbild für den Bediener/-in 8 einer späteren Auswertung markiert werden. Figur 8 zeigt ein Beispiel für ein Kamerabild 13 mit markierten White-/Darkline-Bereichen 20 als Teil eines solchen Reports.

The method according to the invention can also be adapted in further preferred embodiments:

• In this way, the downstream filters can be varied.
• The number of White/Darkline 14 candidates must meet a minimum number per column to be reported as White/Darkline 14
• A maximum value is defined for the brightness value of the pixels so that very bright pixels do not falsify the mean value. Because a white/dark line 14 does not have any very bright pixels in a 670 dpi camera image 13. This means that all gray values in the image > 50 are limited to 50;
• It is checked whether strong structures are present at the relevant points in the reference image, which already lead to a white/darkline-like structure in the reference image and therefore have to be masked out in the camera image 13 . The reference image does not have to be available in full resolution for this, since only a rough estimate is required as to whether the partial area of the reference image is structured or homogeneous; sh. step 4
• The method described above can be implemented on a graphics card (GPU) as a computational accelerator for real-time application.
• The detection algorithm described can be implemented as a sub-component of the image acquisition system 12 that performs the image inspection. Data for a report to the operator 8 or customer can then be derived from the WLC( x , y ) image 21 by recognizing connected areas (blobs) in the image 13 and in an overview image for the operator 8 can be marked for later evaluation. figure 8 shows an example of a camera image 13 with marked white/darkline areas 20 as part of such a report.

However, in these other preferred embodiments, a reference image is usually required, which, in addition to the disadvantages already mentioned, impairs the processing speed. However, the use of a reference image can further improve the quality of the method by avoiding false positive detected white/dark lines 14 .

The method according to the invention thus has many advantages over the prior art. In the case of large color deviations between the target image and the camera image 13, for example in the case of an incorrectly calibrated workflow with regard to cameras 10, white balance, type of paper, short white/dark lines 14, they are often lost in the image/signal noise. This disadvantage is eliminated with the method according to the invention. Also must at from the Methods known in the art allow the reference image to be transported into the computer 9 in full resolution, for example with 670 dpi. With the technical means currently available, this is only possible at great expense. Since the algorithm presented here does not require a reference image or at least a high-resolution reference image, these costs can be saved. Finally, no reference image is fundamentally required for detection, even if it can be used to avoid false-positive white/darkline detections due to structures contained in the customer's motif. In particular for the detection of white/dark line candidates 14, a direct comparison of the reference and camera images 13 no longer has to take place.

1. 1. Das Kamerabild 13 wird in seinen Grauwerten/Farbkanalwerten komprimiert. Die Kompression wird so ausgeführt, dass Helligkeitswerte oberhalb einer Schwelle S_max auf den Schwellwert S_max begrenzt werden. Dies unterdrückt effektiv alle Strukturen im Bild 13, die heller als S_max sind. Damit werden in diesem Schritt White-/Darklines 14 in dunklen Flächen in homogenen und inhomogenen Bereichen sehr gut gefunden. Die Kompression wird vor dem ersten Schritt des vorherigen Ausführungsbeispiels ausgeführt.
2. 2. Auch hier wird das Kamerabild 13 in seinen Grauwerten, bzw. Farbkanalwerten auf komprimiert. Die Kompression wird hier jedoch so ausgeführt, dass Helligkeitswerte oberhalb einer Schwelle K_max (K_max > S_max) auf den Schwellwert K_max begrenzt werden. Die Kompression wird vor dem dritten Schritt des vorherigen Ausführungsbeispiels ausgeführt. Zusätzlich wird die lokale Homogenität des Bildes 13 berechnet, indem bei der Mittelung im zweiten Schritt des vorherigen Ausführungsbeispiels zum Mittelwert noch die Standardabweichung des Spaltensegments berechnet wird. Nur White-/Darklines 14 in relativ homogenen Flächen, d.h. bei einer Standardabweichung < σ_max, werden in die Kandidatenliste aufgenommen. Diese Filterung kann im dritten Schritt des vorherigen Ausführungsbeispiels erfolgen. In diesem Ansatz werden White-/Darklines 14 in hellen, homogenen Flächen sehr gut gefunden. White-/Darklines 14 in hellen inhomogenen Bereichen sind für das menschliche Auge ohnehin nur schwer zu sehen und werden deshalb nicht berücksichtigt.

There is also a further, particularly preferred exemplary embodiment for the method according to the invention, which further improves the method. The following two-stage algorithm, based on the previous embodiment variant, is proposed for this purpose:
In stage one, a targeted search is made for white/darkline candidates 14:
For this purpose, the algorithm presented in the previous exemplary embodiments is called up several times, with different parameters. The results of the algorithm runs are then logically linked. In addition, the algorithm will be further improved in its processes. This is done as follows:
The algorithm is applied to camera image 13 of sheet 2 multiple times. The parameters are adjusted as follows for the different applications:

1. 1. The camera image 13 is compressed in terms of its gray values/color channel values. The compression is carried out in such a way that brightness values above a threshold S _max are limited to the threshold value S _max . This effectively suppresses all structures in image 13 that are brighter than S _max . In this step, white/dark lines 14 in dark areas in homogeneous and inhomogeneous areas are found very well. The compression is performed before the first step of the previous embodiment.
2. 2. Here, too, the camera image 13 is compressed in terms of its gray values or color channel values. However, the compression is carried out here in such a way that brightness values above a threshold K _max (K _max >S _max ) are limited to the threshold value K _max . The compression is performed before the third step of the previous embodiment. In addition, the local homogeneity of the image 13 is calculated by also calculating the standard deviation of the column segment from the mean value in the second step of the previous exemplary embodiment. Only white/dark lines 14 in relatively homogeneous areas, ie with a standard deviation <σ _max , are included in the candidate list. This filtering can be done in the third step of the previous embodiment. In this approach, white/dark lines 14 are found very well in bright, homogeneous areas. White/dark lines 14 in bright inhomogeneous areas are difficult for the human eye to see anyway and are therefore not taken into account.

Both results are combined with a logical OR and combined into a white/darkline candidate list. More complex links with further information are also optionally conceivable.

• Median statt Mittelwert; Vorteil ist hier, dass das Verfahren sehr robust auf Ausreißer reagiert
• Mittelwert nur über Pixel, die einen maximalen Helligkeitswert G_max,mean nicht überschreiten. Vorteil ist hier das Ausfiltern von hellen Ausreißern oder Papierweiß-Bereichen, die den Mittelwert verfälschen können. Dies ist beispielhaft in Figur 9 dargestellt. Gut ist hier die Störung des Spaltenmittelwerts im oberen und unteren Teil von Figur 9 durch helle Papierweiß-Bereich oder einzelne helle Pixel zu erkennen. Ein Problem stellt hier allerdings das Auftreten von Pseudo-White-/Darklines 14b, sowie ein mangelnder Kontrast des erfassten Druckbildes 13 dar. Im mittleren Teil ist dabei das von der Kamera 10 erfasste Druckbild 13 mit einem White-/Darkline-Fehler 14 abgebildet. Aus diesem Druckbild 13 wird jeweils ein Streifen mit Text 15a und ein Streifen am Bildrand 15b ausgeschnitten, aus denen dann jeweils Bildsignale 18, 19 erzeugt werden. Im Bildsignal 18 des Streifens mit Text 15a erkennt man gut die genannte Auswirkung des White-/Darkline-Fehlers 14 im Signal in Form eines entsprechenden Peaks 14a im Signal 18. Zudem ist ein Peak eines Pseudo-White-/Darkline-Fehler 14b dargestellt, der sich aus der Textdarstellung ergibt. Gut zu erkennen ist, dass eine Unterscheidung im Bildsignal 18 zwischen echtem Peak eines White-/Darkline-Fehlers 14a und dem Peak 14b eines Pseudo-White-/Darkline-Fehlers 14b schwer zu unterscheiden ist, da die Mindestdetektionshöhe 19 von beiden Peaks 14a, 14b überschritten wird. Im unteren Teil sind zwei Bildsignale 18a, 18b für den Fall des erzeugten Signals für den Bildrand dargestellt. Hier wird die Mindestdetektionshöhe 19 nur im Signal mit verbessertem Kontrast 18a überschritten und die White-/Darkline 14 sicher erkannt. Im zweiten Signal mit niedrigen Kontrast 18b wird die Mindestdetektionshöhe 19 nicht überschritten und die White-/Darkline 14 entsprechend nicht erkannt.

In the second step of the previous exemplary embodiment, various averaging methods, in addition to simple averaging, can also be applied to a generated image signal, which have advantageous properties. These are for example:

• Median instead of mean; The advantage here is that the method reacts very robustly to outliers
• Mean value only over pixels that do not exceed a maximum brightness value G _max,mean . The advantage here is the filtering out of bright outliers or paper white areas that can falsify the mean value. This is an example in figure 9 shown. Good here is the perturbation of the column mean in the upper and lower parts of figure 9 recognizable by bright paper white areas or individual bright pixels. A problem here, however, is the appearance of pseudo-white/dark lines 14b and a lack of contrast in the recorded print image 13. The print image 13 captured by the camera 10 is shown with a white/dark line error 14 in the middle part. A strip with text 15a and a strip at the edge of the image 15b are cut out of this printed image 13, from which image signals 18, 19 are then generated. The image signal 18 of the strip with text 15a clearly shows the aforementioned effect of the white/darkline error 14 in the signal in the form of a corresponding peak 14a in the signal 18. A peak of a pseudo white/darkline error 14b is also shown, resulting from the text representation. It is easy to see that it is difficult to differentiate in the image signal 18 between the real peak of a white/darkline error 14a and the peak 14b of a pseudo white/darkline error 14b, since the minimum detection height 19 of both peaks 14a, 14b is exceeded. In the lower part, two image signals 18a, 18b are shown for the case of the generated signal for the edge of the image. Here the minimum detection level 19 is only exceeded in the signal with improved contrast 18a and the white/dark line 14 is reliably detected. In the second signal with low contrast 18b, the minimum detection height 19 is not exceeded and the white/dark line 14 is accordingly not recognized.

1. 1. Es werden zwei Schwellen verwendet, je nachdem ob eine 1-Pixel-breite, oder eine 2-Pixel-breite White-/Darkline 14 detektiert werden soll. Je nach Kamerauflösung kann es zudem auch sinnvoll sein, auch 3-, 4-, ... N-Pixel breite White-/Darklines 14 zu finden. Es müssen dann entsprechend mehr Schwellen angewendet werden. Der Detektions-Ausdruck aus dem dritten Schritt lautet dann mit den zwei Schwellen L1 und L2: $$\mathrm{WLC}\left(x,s\right)=\left(\left(I_{C,s}\left(x\right)-I_{C,s}\left(x-1\right)>\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">L</figure-callout>}1\right)\mathrm{and}\left(I_{C,s}\left(x\right)-I_{C,s}\left(x+1\right)>\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">L</figure-callout>}1\right)\right)\mathrm{or}\left(\left(\begin{array}{l}{I}_{C,s}\left(x\right)-\\ I_{C,s}\left(x-1\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\end{array}\right)\mathrm{and}\left(I_{C,s}\left(x\right)-I_{C,s}\left(x+2\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\right)\right)\mathrm{or}\left(\left(I_{C,s}\left(x\right)-I_{C,s}\left(x-2\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\right)\mathrm{and}\left(\begin{array}{l}{I}_{C,s}\left(x\right)\\ -I_{C,s}\left(x+1\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\end{array}\right)\right)$$
   
2. 2. Die Schwelle kann von der lokalen Umgebung jedes Pixel x abhängig gemacht werden, sodass für White-/Darklines 14 in hellen Bildbereichen höhere Schwellen angewendet werden, als in Bildbereichen mit niedriger Helligkeit. Als Maß für die lokale Helligkeit können die Grauwerte in einer engen Umgebung um die Position x unter Ausschluss einer evtl. vorhandenen White-/Darkline 14 gemittelt werden. Alternativ kann ein gleitender Medianfilter auf I_C,s(x) angewendet werden.

In the third step of the previous exemplary embodiment, white/dark lines 14 are detected via a threshold L. In this further exemplary embodiment, two further improvements are also found for this threshold:

1. 1. Two thresholds are used, depending on whether a 1 pixel wide or a 2 pixel wide white/dark line 14 is to be detected. Depending on the camera resolution, it can also make sense to find white/dark lines 14 that are 3, 4, ... N pixels wide. Correspondingly more thresholds must then be applied. The detection expression from the third step then reads with the two thresholds L1 and L2: $$\mathrm{WLC}\left(x,s\right)=\left(\left(I_{C,s}\left(x\right)-I_{C,s}\left(x-1\right)>\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">L</figure-callout>}1\right)\mathrm{other}\left(I_{C,s}\left(x\right)-I_{C,s}\left(x+1\right)>\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="imgf0002.png,imgf0006.png"\; state="\{\{state\}\}">L</figure-callout>}1\right)\right)\mathrm{or}\left(\left(\begin{array}{l}{I}_{C,s}\left(x\right)-\\ I_{C,s}\left(x-1\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\end{array}\right)\mathrm{other}\left(I_{C,s}\left(x\right)-I_{C,s}\left(x+2\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\right)\right)\mathrm{or}\left(\left(I_{C,s}\left(x\right)-I_{C,s}\left(x-2\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\right)\mathrm{other}\left(\begin{array}{l}{I}_{C,s}\left(x\right)\\ -I_{C,s}\left(x+1\right)>\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="imgf0001.png,imgf0003.png"\; state="\{\{state\}\}">L</figure-callout>}2\end{array}\right)\right)$$
   
2. 2. The threshold can be made dependent on the local environment of each pixel x, so that higher thresholds are used for white/dark lines 14 in bright image areas than in image areas with low brightness. As a measure of the local brightness, the gray values can be averaged in a narrow area around the position x, excluding any white/dark line 14 that may be present. Alternatively, a moving median filter can be applied to _IC,s (x).

• Berechnung des Luminanz-Kanals aus dem Lab-Farbraum
• Berechnung des Helligkeitswertes oder Sättigungswertes aus dem HSB-Farbraum
• Mittelung der geeignet gewichteten RGB-Farbkanäle, auf das menschliche Auge angepasst

As a further advantageous improvement of the previous exemplary embodiment, the algorithm cannot be applied to an RGB image 13, but instead the RGB image 13 is previously converted into a gray scale image using a suitable method, which has the best possible contrast for white/dark lines 14. Suitable transformation operations for this are:

• Calculation of the luminance channel from the Lab color space
• Calculation of the lightness value or saturation value from the HSB color space
• Averaging the appropriately weighted RGB color channels, adjusted to the human eye

In stage 2, pseudo-white/darklines 14b are now filtered out of the white/darkline candidates 14 determined in stage one by using one or more filters. There are the following improvements compared to the previous embodiment:
By applying a column filter to the white/darkline candidate list, all white/darkline candidates 14 that do not have at least the number N _col,min of other white/darkline candidates 14 in one and the same image column are removed from the white/darkline candidate list. Removed darkline candidate list. The idea behind this filter is to rule out very brief or isolated false detections. Because a white/dark line 14 will affect several areas of a column in most realistic print motifs, while false detections only occur locally.

The filter described in step four of the previous exemplary embodiment using the reference image is also carried out here with all the modifications described above. As an improvement, the size of the reference image is adjusted beforehand. It can also make sense to process the reference image several times in different resolutions and to combine the results of these stages before filtering. This simulates the loss of quality on the "perfect" reference image by the camera system 10 and thus makes it possible to effectively detect different structures that can lead to white/darkline-like structures in the camera image 13 .

The further, particularly preferred exemplary embodiment has the additional advantage over the previous exemplary embodiment that it has a higher detection performance for white/dark lines 14 with a simultaneously smaller number of pseudo white/dark lines 14b. However, this requires a reference image evaluation, the additional processing steps of which can lead to longer computing times for the computer 6, 9 used. The decision as to which preferred embodiment is used should therefore be based on the requirements of the specific application. Print jobs in which the white/dark line detection is time- or performance-critical should rather use the first exemplary embodiment presented, while print jobs in which print images require a thorough detection of white/dark lines 14 and/or in which an increased There is a risk of pseudo-white/dark lines 14b appearing, should rather resort to the second exemplary embodiment presented.

Reference List

1

investor

2

print substrate

3

boom

4

inkjet printing unit

5

Inkjet print head

6

Control computer of the inkjet printing machine

7

inkjet printing machine

8th

user

9

image processing computer

10

image sensor / camera

11

screen

12

image capture system

13

captured print image

14

Whiteline/Darkline misprints

14a

Peak of a whiteline/darkline in the generated image signal

14b

Peak of a pseudo whiteline/darkline in the generated image signal

15

Streaks of the captured print image

15a

Stripes of a captured print image with text content

15b

Streaks of a captured print image at the edge of the image

16

recognized and marked white/darklines

17

Enlarged section from the strip of the recorded print image

18

generated image signal from the strip of the captured print image with text content

18a

generated image signal from the strip of the captured print image at the edge of the image

18b

generated image signal from the strip of the captured print image at the edge of the image

19

Minimum detection level of a white/dark line in the generated image signal

20

marked white/darkline areas#

21

candidate image composed of stripes

Claims (15)

1. Method of determining print defects by means of a computer (9) in a printing operation carried out in an inkjet printing machine (7) to process a print job, wherein printed products (2) created during the printing operation are recorded and digitized by means of a camera system (10), the camera image (13) generated in this way is fed to a detection algorithm on the computer (9), and, when print defects (14) are detected, a message is sent to a machine control unit (6), which then potentially removes the printed product (2) via a waste deflector,
   characterized
   in that the detection algorithm separates colour separations of the camera images (13), detects the print defects (14) in the colour separations, links images of the individual color separations to a candidate image (21), filters the candidate image (21), wherein the candidate image (21) is filtered to ensure that only print defects (14) that lead to unusable prints are detected, and subsequently the remaining detected print defects (14) are entered into a list which is then sent to the machine control unit (6) of the printing machine (7).
2. Method according to claim 1,
   characterized
   in that the print defects (14) are white line or dark line defects (14) caused by defective printing nozzles of the inkjet printing machine (7).
3. Method according to claim 2,
   characterized
   in that in a further step of the method implemented before sending to the machine control (6) of the printing machine (7), the computer (9) filters out pseudo-white line or pseudo-dark line defects (14) from the list of white line or dark line defects (14) by applying a specific testing method.
4. Method according to any one of claims 2 to 3,
   characterized
   in that based on the list of the remaining detected white line or dark line defects (14), the computer (9) determines the defective printing nozzles that cause the defects and compensates for the white line or dark line defects (14) as a function of the determination and using compensation methods suitable for each case.
5. Method according to claim 4,
   characterized
   in that for the specific testing method, the computer (9) creates a reference image based on print preparation data of the print job, applies the detection algorithm to said reference image, thus either gaining information on candidates for pseudo-white line or pseudo-dark line defects (14b) it has received and removing the latter from the list of white line or dark line defects (14) or gaining information on regions in the camera image (13) that probably include pseudo-white line or pseudo-dark line defects (14) and thus not applying the detection algorithm to these areas in the camera image.
6. Method according to claim 5,
   characterized
   in that the computer (9) creates the reference image in different sizes and/or resolutions, correspondingly applies the detection algorithm multiple times to the different reference images, and sums up and applies the information obtained in this way.
7. Method according to claim 6,
   characterized
   in that the algorithm is not applied to regions which are characterized by a high variation of grey values in a limited local environment in the reference image or in that results from such regions are excluded.
8. Method according to any one of the preceding claims,
   characterized
   in that the computer (9) creates the list of white line or dark line defects (14) by column totals in the filtered candidate image (21) by applying a limiting value to the respective calculated column total in the candidate image (21).
9. Method according to any one of the preceding claims,
   characterized
   in that the computer (9) links the candidate image (21) of the individual colour separations (9) by means of a mathematical OR operation.
10. Method according to any one of the preceding claims,
    characterized
    in that the filtering of the candidate image (21) is done by the computer (9) by means of morphological operations.
11. Method according to any one of the preceding claims,
    characterized
    in that the detection algorithm is applied to the generated camera image (13) multiple times by the computer (9), wherein each time the method is parameterized in a different way to detect dark line or white line defects (14) of different degrees and in that the results of all colour separations of all applications of the method are logically linked to one another.
12. Method according to claim 11,
    characterized
    in that for a respective one of the applications of the method that have been parameterized in different ways, every pixel in the generated camera image (13) is limited to a maximum grey value.
13. Method according to any one of the preceding claims,
    characterized
    in that the creation of the candidate image (21) of a colour channel is done by dividing the generated camera image (13) into horizontal stripes (15), every stripe (15) being reduced by suitably averaging every one of its columns to an image signal (18) in which white or dark lines (14) are then searched for by a specific search process, every line that has been analysed in this way resulting in a line of the white line candidate image (21).
14. Method according to claim 13,
    characterized
    in that the white or dark line search process recognizes a dark or white line (14) at a position by observing a limited neighbourhood around the examined pixel in the image signal (18).
15. Method according to claim 14,
    characterized
    in that the search process firstly folds the image signal (18) with different filter kernels and converts the results to logical signals by a comparison with potentially different limiting values, and in that the logical signals are then converted to a white or dark line candidate image signal with the aid of a logical operation.

Images