Image inspection device, printing device equipped with same, and image inspection method

Abstract

To enable a user to easily set suitable inspection conditions with reference to an image inspection device for inspecting a printed image.SOLUTION: An image inspection device comprises: an inspection condition setting unit 62 for setting a determination criterion as an inspection condition for detecting a defect image; an imaging unit 61 for taking an image of a printed image to output the captured image obtained by the imaging; and a defect image detection unit 64 for detecting the defect image from the captured image based on the inspection condition. The inspection condition setting unit 62 is configured to be able to set two different determination criteria, a first parameter value PA1 and a second parameter value PA2, as the inspection condition. The defect image detection unit 64 performs first detection processing for detecting a defect image from the captured image based on the first parameter value PA1, and second detection processing for detecting a defect image from the captured image based on the second parameter value PA2.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to an image inspection device that inspects printed images. [Background technology]

[0002] In recent years, there has been a strong demand for improved quality in printed materials, and as a result, the importance of inspecting printed materials has increased. Accordingly, inkjet printing devices equipped with image inspection devices that inspect printed images (images obtained by printing on a substrate such as printing paper) are becoming increasingly common. Image inspection devices, for example, compare captured images of printed images obtained by capturing images with a camera with original images corresponding to the original printing data to detect defects (defective images) such as ink stains and white streaks (streaky images caused by ink omissions during printing) in real time. Sheets containing defects are then discarded as waste and are subject to reprinting, for example.

[0003] Among the defects detected by image inspection equipment, some are critical enough to be unacceptable in the final product, while others are non-critical enough to be ignored in some cases. Critical defects include stains caused by ink dripping, multiple missing nozzles, and stains on white paper. Non-critical defects include scumming on the printing paper, a single missing nozzle, and stains on high-density images. However, the type of defect that is considered critical varies depending on the print quality requirements of the customer of the printed product.

[0004] To perform an inspection using an image inspection system, the user (operator) must set the inspection conditions to identify defects (defect images). Typically, the inspection conditions are set by the user specifying the values ​​of parameters (size of the defect, gradation difference, etc.). The parameter values ​​(values ​​specified by the user) become thresholds that the image inspection system uses to determine whether or not to regard the image being inspected as a defective image.

[0005] Here, with reference to FIG. 34, the relationship between the strictness of the inspection conditions and the defects detected will be described. FIG. 34 focuses on five levels (L1 to L5) of the strictness of the inspection conditions (hereinafter abbreviated as "inspection levels"). The portion marked with reference numeral 91 represents the range in which no defects are detected, the portion marked with reference numeral 92 represents the range in which high-criticality defects are detected, and the portion marked with reference numeral 93 represents the range in which low-criticality defects are detected. If the inspection level is lower than L1, no defects are detected. If the inspection level is L2, half of the high-criticality defects are detected. If the inspection level is L3, all of the high-criticality defects are detected. If the inspection level is L4, all of the high-criticality defects and half of the low-criticality defects are detected. If the inspection level is L5, all of the high-criticality defects and all of the low-criticality defects are detected. As described above, the stricter the inspection conditions, the more defects are detected by inspection. The parts denoted by reference numeral 94 in FIG. 34 are used in the other drawings (FIGS. 9 to 12, 14, 15, 30 to 33, and 35 to 37) in the same sense.

[0006] In relation to the present invention, Japanese Patent Application Laid-Open No. 2016-103815 discloses an invention of an inspection device equipped with a modification unit that modifies threshold information (inspection conditions) to resolve any discrepancies that may arise between the inspection results obtained by the inspection unit and the inspection results obtained visually by the user. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-103815 Summary of the Invention [Problem to be solved by the invention]

[0008] According to conventional image inspection devices, as described above, defects are detected according to the inspection conditions (parameter values) set by the user. In order to detect defects as desired by the user, the inspection conditions must be set appropriately. However, setting the inspection conditions generally requires the user to adjust the values ​​of many parameters. Therefore, it is difficult for the user to set the inspection conditions appropriately, which can result in printing accidents (delivery of defective printed matter to the client) or wasteful disposal of printed matter. This will be explained with reference to Figures 35 to 37.

[0009] Assume here that all critical defects are unacceptable in the final product, and all low-critical defects are acceptable. If the inspection conditions are set so that the range indicated by reference numeral 94 in FIG. 35 is the inspection range, the inspection will detect all critical defects but not any low-critical defects. This is an ideal inspection result. If the inspection level is lower than that shown in FIG. 35, i.e., if the inspection conditions are more lenient than those shown in FIG. 35, the inspection range will be, for example, the range indicated by reference numeral 95 in FIG. 36. In this case, the critical defects corresponding to the range indicated by reference numeral 96 in FIG. 36 will not be detected by the inspection. As a result, a printing accident will occur. On the other hand, if the inspection level is higher than that shown in FIG. 35, i.e., if the inspection conditions are more strict than those shown in FIG. 35, the inspection range will be, for example, the range indicated by reference numeral 97 in FIG. 37. In this case, the inspection will detect all critical defects as well as low-critical defects corresponding to the range indicated by reference numeral 98 in FIG. 37. As a result, printed materials are wasted (if the practice is to discard sheets containing defects detected by inspection without the user having to visually check them).

[0010] Furthermore, according to the invention disclosed in JP 2016-103815 A, the user needs to specify areas (gap areas) where there is a discrepancy between the inspection results by the inspection unit and the user's visual inspection results. In this regard, if there are many gap areas, the user needs to specify all of the gap areas one by one, which is a significant burden for the user. Furthermore, the task of visually detecting areas that should be treated as defective areas from areas that are non-defective in the inspection results by the inspection unit is also a significant burden for the user.

[0011] In view of the above circumstances, the present invention relates to an image inspection device that inspects printed images, and has an object to enable a user to easily set suitable inspection conditions. [Means for solving the problem]

[0012] A first invention is an image inspection device that performs an inspection to detect defective images contained in a printed image, an inspection condition setting unit for setting a judgment criterion for detecting a defect image as an inspection condition; an imaging unit that captures an image of the print image and outputs the captured image; a defect image detection unit that detects a defect image from the captured image based on the inspection conditions; Equipped with The inspection condition setting unit sets the same inspection conditions as inspection The system is configured to allow setting of two different judgment criteria, a first parameter value and a second parameter value, for each item; The first parameter value is a plurality of Regarding each test item A combination of parameter values, The second parameter value is Regarding each test item A combination of parameter values, the inspection condition setting unit displays an inspection setting screen for setting the first parameter value and the second parameter value, and receives input of the first parameter value and the second parameter value from an external device; The defect image detection unit is characterized by performing a first detection process to detect a defect image from the captured image based on the first parameter value, and a second detection process to detect a defect image from the captured image based on the second parameter value.

[0013] The second invention is the first invention, The image inspection device further comprises a result display unit that displays defect images detected by the defect image detection unit as inspection results.

[0014] The third invention is the first or second invention, The defect image detection unit a defect detection processing unit that performs the first detection processing and the second detection processing; a filter unit including a trained artificial intelligence model that determines whether a given image is a good image or a bad image, and that extracts some defect images from the defect images detected by the defect detection processing unit; It consists of The artificial intelligence model is provided with a defect image detected by the defect detection processing unit, The filter unit extracts defective images that are determined to be defective images by the artificial intelligence model.

[0015] A fourth aspect of the present invention is the third aspect of the present invention, The artificial intelligence model is provided with both the defect image detected in the first detection process and the defect image detected in the second detection process.

[0016] A fifth aspect of the present invention is the third aspect of the present invention, The artificial intelligence model is provided with only one of the defect images detected in the first detection process and the defect images detected in the second detection process.

[0017] The sixth invention is any one of the third to fifth inventions, The image inspection device is characterized by further comprising a re-learning execution unit that causes the artificial intelligence model to perform re-learning by providing the artificial intelligence model with a defect image selected from the defect images detected by the defect image detection unit and a judgment result indicating whether the selected defect image is a good image or a bad image.

[0018] The seventh invention is any one of the first to sixth inventions, the image inspection device further includes an inspection condition storage unit that stores the first parameter value and the second parameter value set by the inspection condition setting unit; The inspection condition setting unit is characterized in that it is configured to be able to read out the first parameter value and the second parameter value stored in the inspection condition storage unit after the first detection process and the second detection process are completed, and to reset the first parameter value and the second parameter value.

[0019] The eighth invention is any one of the first to seventh inventions, The image inspection device further comprises a notification unit that notifies the user of content corresponding to the defect image detected by the defect image detection unit.

[0020] A ninth aspect of the present invention is the eighth aspect of the present invention, The notification unit is characterized in that it notifies an external device that executes a pre-process or post-process of the printing process that is performed by the printing device that printed the print image.

[0021] A tenth aspect of the present invention is the eighth or ninth aspect of the present invention, The notification unit performs notification by displaying a message on a predetermined display unit.

[0022] The eleventh invention is any one of the eighth to tenth inventions, The notification unit is characterized in that it notifies by voice.

[0023] A twelfth invention is any one of the eighth to eleventh inventions, The image inspection device is configured to allow selection of one of a plurality of pre-prepared notification levels, and is characterized by further comprising a notification content change unit that changes the content notified by the notification unit according to the selected notification level.

[0024] A thirteenth aspect of the present invention is the twelfth aspect of the present invention, The notification content change unit is characterized in that it changes the content notified by the notification unit only for either the notification based on the defect image detected by the first detection process or the notification based on the defect image detected by the second detection process.

[0025] The fourteenth invention is any one of the first to thirteenth inventions, the criticality corresponding to the second parameter value is lower than the criticality corresponding to the first parameter value; a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; The second detection process is characterized in that a defect image having a fatality level higher than the fatality level corresponding to the second parameter value and lower than the fatality level corresponding to the first parameter value is detected.

[0026] A fifteenth aspect of the present invention is the fourteenth aspect of the present invention, A minimum limit value, which is a value corresponding to the lowest criticality among the values ​​that can be set as the second parameter value, is predetermined; The inspection condition setting unit is configured to input the first parameter value from an external source and to detect a defect image by the second detection process. The range of fatality and sets the second parameter value to a value between the first parameter value and the minimum limit value in accordance with the range level input from the outside.

[0027] The sixteenth invention is any one of the first to thirteenth inventions, the criticality corresponding to the second parameter value is higher than the criticality corresponding to the first parameter value; a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; The second detection process is characterized in that a defect image having a higher fatality level than the fatality level corresponding to the second parameter value is detected.

[0028] A seventeenth aspect of the present invention is the sixteenth aspect of the present invention, A maximum limit value, which is a value corresponding to the highest criticality among the values ​​that can be set as the second parameter value, is predetermined; The inspection condition setting unit is configured to input the first parameter value from an external source and to detect a defect image by the second detection process. The range of fatality and sets the second parameter value to a value between the first parameter value and the maximum limit value in accordance with the range level input from the outside.

[0029] The eighteenth invention is any one of the first to thirteenth inventions, a minimum limit value, which is a value corresponding to the lowest fatality level among the values ​​that can be set as the second parameter value, and a maximum limit value, which is a value corresponding to the highest fatality level among the values ​​that can be set as the second parameter value, are determined in advance; the inspection condition setting unit receives an external input of the first parameter value and an external input of a magnitude relationship between the criticality corresponding to the first parameter value and the criticality corresponding to the second parameter value, and sets the second parameter value based on the external input content; If the fatality level corresponding to the second parameter value set by the inspection condition setting unit is lower than the fatality level corresponding to the first parameter value input from the outside, a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; a defect image having a criticality higher than the criticality corresponding to the second parameter value and lower than the criticality corresponding to the first parameter value is detected by the second detection process; If the fatality level corresponding to the second parameter value set by the inspection condition setting unit is higher than the fatality level corresponding to the first parameter value input from the outside, a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; The second detection process is characterized in that a defect image having a higher fatality level than the fatality level corresponding to the second parameter value is detected.

[0030] A nineteenth aspect of the present invention is a printing device, a conveying unit that conveys the substrate; a printing unit that performs printing based on print data on the base material transported by the transport unit; an image inspection device according to any one of the first to eighteenth aspects of the present invention, having the imaging unit arranged so as to be able to capture the printed image obtained by printing by the printing unit; The present invention is characterized by the following features.

[0031] A twentieth invention is an image inspection method for inspecting a printed image to detect a defective image, comprising: an inspection condition setting step in which an operator sets criteria for detecting defect images as inspection conditions; an imaging step in which an imaging device captures the print image; a defect image detection step in which a computer detects a defect image from the captured image obtained in the imaging step based on the inspection conditions; Including, In the inspection condition setting step, the same inspection a first parameter value and a second parameter value, which are two different judgment criteria for the item, are set by an operator; The first parameter value is a plurality of Regarding each test item A combination of parameter values, The second parameter value is Regarding each test item A combination of parameter values, In the inspection condition setting step, an inspection setting screen for setting the first parameter value and the second parameter value is displayed, and input of the first parameter value and the second parameter value by the operator is accepted; The defect image detection step is characterized in that a first detection process is performed to detect a defect image from the captured image based on the first parameter value, and a second detection process is performed to detect a defect image from the captured image based on the second parameter value. [Effects of the Invention]

[0032] According to the first aspect of the present invention, a user can set two inspection conditions (a first parameter value and a second parameter value) for an inspection to detect defective images contained in a printed image. The defective image detection unit then performs a first detection process to detect defective images based on the first parameter value and a second detection process to detect defective images based on the second parameter value. This makes it possible to obtain inspection results under two different inspection conditions. Therefore, for example, it is possible to determine the validity of the results of the inspection treated as the main inspection, and it is also easy to adjust the inspection conditions. As a result, a user can easily set suitable inspection conditions for an image inspection device that inspects printed images.

[0033] According to the second aspect of the present invention, the user can easily determine the validity of the test results on the screen.

[0034] According to the third aspect of the present invention, a filter unit including an artificial intelligence model performs a process (filtering) to extract some defect images from the defect images detected by the defect detection processing unit. This reduces the number of images that the user needs to visually check after the inspection is completed, thereby reducing the burden on the user. Furthermore, it becomes easier to adjust the inspection conditions appropriately.

[0035] According to the fourth aspect of the present invention, it is possible to perform filtering using an artificial intelligence model on both the results of an examination treated as a main examination and the results of an examination treated as a sub-examination.

[0036] According to the fifth invention, for example, it is possible to prevent filtering using an artificial intelligence model from being performed on the results of an inspection that is treated as the main inspection, thereby reducing the occurrence of printing accidents caused by incorrect judgments by the artificial intelligence model.

[0037] According to the sixth aspect of the present invention, the artificial intelligence model is retrained, so that the results of an inspection that is used as the main inspection, for example, can be brought closer to the ideal inspection result with fewer operations.

[0038] According to the seventh aspect of the present invention, the user can reset the inspection conditions (reset the first parameter value and the second parameter value) while taking into account the previous input contents.

[0039] According to the eighth aspect of the present invention, it is possible to prevent printing accidents caused by overlooking defects.

[0040] According to the ninth aspect of the present invention, it is possible to change the operation of the external device in accordance with the type of defect detected in the inspection of the printed image.

[0041] According to the tenth aspect of the present invention, for example, when a defect is detected, it is possible to alert the user.

[0042] According to the eleventh aspect of the present invention, for example, when a defect is detected, it is possible to give a stronger warning to the user.

[0043] According to the twelfth aspect of the present invention, the user can select from a plurality of levels the content of the notification to be given by the notification section when a defect is detected.

[0044] According to the thirteenth aspect of the present invention, the user can change the content notified by the notification unit for only one of the tests treated as the main test and the test treated as the sub-test, taking into account the importance of the test.

[0045] According to the fourteenth aspect of the present invention, the second detection process detects defect images that are less lethal than the defect images detected by the first detection process, thereby enabling the user to grasp defect images that would be additionally detected if, for example, the inspection conditions for the main inspection were made stricter.

[0046] According to the fifteenth aspect of the present invention, the burden on the user in setting the inspection conditions is reduced.

[0047] According to the sixteenth aspect of the present invention, defect images having a higher fatality level than the fatality level corresponding to the second parameter value are detected by the second detection process from among the defect images detected by the first detection process, thereby enabling the user to grasp the inspection results that would have been obtained if the inspection conditions for the main inspection had been relaxed, for example.

[0048] According to the seventeenth aspect of the present invention, the burden on the user in setting the inspection conditions is reduced.

[0049] According to the eighteenth aspect of the present invention, the burden on the user in setting the inspection conditions is significantly reduced.

[0050] According to the nineteenth aspect of the present invention, a printing device equipped with an image inspection device that achieves the effects of the first aspect of the present invention is realized.

[0051] According to the twentieth aspect of the present invention, the same effects as those of the first aspect of the present invention can be obtained. [Brief explanation of the drawings]

[0052] [Figure 1] 1 is a diagram illustrating the overall configuration of a printing system according to a first embodiment of the present invention. [Figure 2] FIG. 2 is a schematic diagram showing an example of the configuration of an inkjet printing apparatus according to the first embodiment. [Figure 3] FIG. 2 is a hardware configuration diagram of an image inspection computer in the first embodiment. [Figure 4] FIG. 2 is a block diagram showing a functional configuration of the image inspection device in the first embodiment. [Figure 5] FIG. 10 is a diagram showing an example of an image inspection setting screen in the first embodiment. [Figure 6] FIG. 10 is a diagram showing an example of input for a first inspection condition in the first embodiment. [Figure 7] FIG. 10 is a diagram showing an example of input for a second inspection condition in the first embodiment. [Figure 8] FIG. 10 is a diagram showing an example of a test result screen displayed by a result display unit in the first embodiment. [Figure 9] FIG. 10 is a diagram for explaining a notification by a notification unit in the first embodiment. [Figure 10] FIG. 10 is a diagram for explaining an image detected as a defective image in a sub-inspection when the second inspection level is higher than the first inspection level in the first embodiment. [Figure 11] FIG. 10 is a diagram for explaining an image detected as a defective image in a sub-inspection when the second inspection level is lower than the first inspection level in the first embodiment. [Figure 12] FIG. 10 is a diagram for explaining a first example of an inspection operation example in the first embodiment. [Figure 13] FIG. 10 is a diagram for explaining an example of adjustment of a first parameter value in the first embodiment. [Figure 14] FIG. 10 is a diagram for explaining a change in the first inspection level due to adjustment of a first parameter value in the first embodiment. [Figure 15] FIG. 10 is a diagram for explaining a second example of an example of inspection operation in the first embodiment. [Figure 16]FIG. 10 is a hardware configuration diagram of an image inspection computer according to a second embodiment of the present invention. [Figure 17] FIG. 10 is a block diagram showing a functional configuration of an image inspection device in the second embodiment. [Figure 18] 10 is a flowchart showing the overall processing procedure for constructing an image discrimination model in the second embodiment. [Figure 19] FIG. 11 is a diagram for explaining comparison and collation of captured image data and print data in the second embodiment. [Figure 20] FIG. 10 is a schematic configuration diagram of an image discrimination model in the second embodiment. [Figure 21] FIG. 11 is a diagram for explaining a process during learning in the second embodiment. [Figure 22] FIG. 11 is a diagram showing an example of a detailed configuration of a neural network section in an image discrimination model in the second embodiment. [Figure 23] 10 is a flowchart showing the procedure of a model construction process in the second embodiment. [Figure 24] FIG. 11 is a diagram for illustrating compositing white streak data with captured image data in the second embodiment. [Figure 25] FIG. 11 is a diagram for explaining the combination of dirt data with captured image data in the second embodiment. [Figure 26] FIG. 11 is a diagram for explaining addition of original image data to training data in the second embodiment. [Figure 27] FIG. 11 is a diagram for explaining input data (input data to an image discrimination model) in which original image data is added to imaging base data in the second embodiment. [Figure 28] FIG. 10 is a block diagram showing the functional configuration of an image inspection device according to a second modified example. [Figure 29] FIG. 11 is a diagram showing an example of a notification level setting screen in the second modified example. [Figure 30]FIG. 10 is a diagram for explaining a first case (a case in which the criticality corresponding to the second parameter value is lower than the criticality corresponding to the first parameter value) in the third modified example. [Figure 31] FIG. 10 is a diagram for explaining setting of a second parameter value in a first case in the second modified example. [Figure 32] FIG. 10 is a diagram for explaining a second case (a case in which the criticality corresponding to the second parameter value is higher than the criticality corresponding to the first parameter value) in the third modified example. [Figure 33] FIG. 10 is a diagram for explaining setting of a second parameter value in a second case in the second modified example. [Figure 34] 1A and 1B are diagrams for explaining the relationship between the severity of inspection conditions and detected defects, relating to the prior art. [Figure 35] FIG. 10 is a diagram for explaining a case in which ideal inspection results are obtained in the prior art. [Figure 36] FIG. 1 is a diagram for explaining cases in which printing accidents may occur in the prior art. [Figure 37] 10A and 10B are diagrams illustrating a case in which printed matter is wasted in the prior art. DETAILED DESCRIPTION OF THE INVENTION

[0053] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

[0054] <1. First embodiment> <1.1 Overall configuration of the printing system> FIG. 1 is a diagram illustrating the overall configuration of a printing system according to a first embodiment of the present invention. This printing system includes an inkjet printing device 10 and a print data generating device 40. The inkjet printing device 10 and the print data generating device 40 are connected to each other via a communication line 5. The print data generating device 40 generates print data by performing RIP processing and the like on input data such as a PDF file. The print data generated by the print data generating device 40 is transmitted to the inkjet printing device 10 via the communication line 5. The inkjet printing device 10 outputs a print image (i.e., performs printing) by ejecting ink onto a printing paper substrate based on the print data transmitted from the print data generating device 40 without using a printing plate. The inkjet printing device 10 includes a printing press main body 200, a print control device 100 that controls the operation of the printing press main body 200, and an image inspection device 300 that inspects the print state. In other words, the inkjet printing device 10 is a printing device with an inspection function. Some components of the image inspection device 300 are built into the printing press main body 200.

[0055] 1, the image inspection device 300 is one component of the inkjet printing device 10 (i.e., the image inspection device 300 is included in the inkjet printing device 10), but this is not limiting. The image inspection device 300 may also be a device independent of the inkjet printing device 10.

[0056] <1.2 Configuration of inkjet printing device> 2 is a schematic diagram showing an example of the configuration of the inkjet printing apparatus 10. As described above, the inkjet printing apparatus 10 is made up of the print control device 100, the printing press main body 200, and the image inspection device 300.

[0057] The printing press main body 200 is equipped with a paper feed section 21 that supplies printing paper (e.g., roll paper) PA, a printing mechanism 20 that prints on the printing paper PA, and a paper take-up section 27 that takes up the printing paper PA after printing. The printing mechanism 20 is equipped with a first drive roller 22 that transports the printing paper PA inside the printing mechanism 20, a plurality of support rollers 23 that transport the printing paper PA inside the printing mechanism 20, a printing section 24 that ejects ink onto the printing paper PA to print, a drying section 25 that dries the printing paper PA after printing, a camera (imaging device) 302 that captures an image of the print image (printing paper PA after printing), and a second drive roller 26 that outputs the printing paper PA from inside the printing mechanism 20. The camera 302 is a component of the image inspection device 300 and is configured using an image sensor such as a CCD or CMOS. The printing unit 24 includes, for example, a C inkjet head, an M inkjet head, a Y inkjet head, and a K inkjet head that eject C (cyan), M (magenta), Y (yellow), and K (black) ink, respectively. The first drive roller 22, the plurality of support rollers 23, and the second drive roller 26 form a conveying unit.

[0058] The print control device 100 controls the operation of the printing press main body 200 configured as described above. When a print output instruction command is given to the print control device 100, the print control device 100 controls the operation of the printing press main body 200 so that the print paper PA is transported from the paper feed unit 21 to the paper take-up unit 27. Then, first, the printing unit 24 prints on the print paper PA, then the drying unit 25 dries the print paper PA, and finally the camera 302 captures an image of the print image.

[0059] The image inspection device 300 is composed of an image inspection computer 301 and a camera 302. The camera 302 is positioned so that it can capture a print image obtained by printing by the printing unit 24. Captured image data Di obtained by capturing the print image with the camera 302 is sent to the image inspection computer 301. The image inspection computer 301 performs an inspection to detect defects (defect images) based on the captured image data Di. Defect detection is performed by image analysis based on the captured image data Di, or by comparing the captured image data Di with print data Dp (however, data that has been converted from CMYK format to RGB format) sent from the print data generation device 40. The inspection result Dr obtained by the image inspection computer 301 is sent to the print control device 100.

[0060] Although the configuration of the inkjet printing device 10 that performs color printing has been exemplified here, the present invention can also be applied to inkjet printing devices that perform monochrome printing. Also, although the configuration of the inkjet printing device 10 that uses aqueous ink has been exemplified here, the present invention can also be applied to inkjet printing devices that use UV ink (ultraviolet-curable ink), such as inkjet printing devices for label printing.

[0061] <1.3 Configuration of image inspection equipment> <1.3.1 Hardware configuration of image inspection computer> FIG. 3 is a hardware configuration diagram of the image inspection computer 301 constituting the image inspection device 300. As shown in FIG. 3, the image inspection computer 301 has a CPU 31, a ROM 32, a RAM 33, an auxiliary storage device 34, an input operation unit 35 such as a keyboard, a display unit 36, an optical disk drive 37, and a network interface unit 38. The image inspection computer 301 is also connected to a camera 302 provided inside the printing press main body 200. Data transmitted via the communication line 5 (e.g., print data Dp transmitted from the print data generating device 40) is input into the image inspection computer 301 via the network interface unit 38. The inspection result Dr obtained by the image inspection computer 301 is transmitted to the print control device 100 via the network interface unit 38 and the communication line 5. Note that the image inspection computer 301 itself may not have the input operation unit 35 or the display unit 36, and data input by a user (operator) using an input operation unit and a display unit provided in the print control device 100 may be transmitted to the image inspection computer 301.

[0062] The auxiliary storage device 34 stores an image inspection program 341. The image inspection program 341 is provided by being stored on a computer-readable recording medium (non-transitory recording medium) such as a CD-ROM or DVD-ROM. That is, a user purchases, for example, an optical disk (CD-ROM, DVD-ROM, etc.) 370 as a recording medium for the image inspection program 341, inserts it into the optical disk drive 37, reads the image inspection program 341 from the optical disk 370, and installs it in the auxiliary storage device 34. Alternatively, the image inspection program 341 transmitted via the communication line 5 may be received by the network interface unit 38 and installed in the auxiliary storage device 34. When inspecting a printed image, the image inspection program 341 stored in the auxiliary storage device 34 is read into the RAM 33, and the CPU 31 executes the image inspection program 341 read into the RAM 33.

[0063] <1.3.2 Functional configuration> 4 is a block diagram showing the functional configuration of an image inspection device 300 according to this embodiment. Functionally, this image inspection device 300 includes an imaging unit 61, an inspection condition setting unit 62, an inspection condition storage unit 63, a defect image detection unit 64, a result display unit 65, and a notification unit 66.

[0064] The imaging unit 61 is a component realized by the camera 302 as hardware, and captures a print image Pi (an image obtained by printing on print paper PA by the printing unit 24) and outputs captured image data Di obtained by the capture. Note that the imaging step is realized by the operation of this imaging unit 61.

[0065] The inspection condition setting unit 62 displays a screen 7 (hereinafter referred to as the "image inspection setting screen") on the display unit 36, which allows the user to set inspection conditions, etc., that serve as criteria for detecting defects (defect images), and accepts input from the user. FIG. 5 is a diagram showing an example of the image inspection setting screen 7. The image inspection setting screen 7 includes a setting value input area 71, a setting content list area 72, and a close button 73. The setting value input area 71 displays a screen on which the user can set (input) parameter values ​​corresponding to the items selected in the setting content list area 72. The setting content list area 72 displays a list of items related to inspection that the user can set. The user can select one of the multiple items displayed in the setting content list area 72. The close button 73 is a button that the user presses when finishing input on the image inspection setting screen 7. The inspection condition setting step is realized by the user setting parameter values ​​on the image inspection setting screen 7.

[0066] In this embodiment, two inspection conditions (first and second inspection conditions) can be set on the image inspection setting screen 7. This allows for inspections to be performed to detect defects based on each of the two inspection conditions. In this embodiment, it is assumed that the inspection based on the first inspection conditions is treated as the main inspection, and the inspection based on the second inspection conditions is treated as the sub-inspection. FIG. 5 shows the image inspection setting screen 7 when the first inspection conditions are selected from among the multiple items (mask, first inspection conditions, second inspection conditions, temporary mask, and paper shrinkage correction) displayed in the setting content list area 72. As can be seen from the portion labeled with reference numeral 74 in FIG. 5, the first inspection conditions are highlighted from among the multiple items. By selecting the first inspection conditions, a screen for the user to set (input) values ​​for four parameters (stain detection, missing part detection, defect width, and defect height) is displayed in the setting value input area 71. A similar screen is also displayed in the setting value input area 71 when the second inspection conditions are selected.

[0067] The setting value input area 71 when the first or second inspection condition is selected in the setting content list area 72 will be described in more detail below (see FIG. 5). As described above, in this example, the user can set values ​​for four parameters (stain detection, loss detection, defect width, and defect height). The field labeled 75 (hereinafter referred to as the "parameter value input field") is a field where the user inputs the value of each parameter. Buttons 76 and 77 are buttons for increasing or decreasing the input value (the value input in the parameter value input field 75). When the user presses button 76, the corresponding input value decreases by a predetermined value. When the user presses button 77, the corresponding input value increases by a predetermined value. For example, when the user presses button 76 corresponding to the input value for defect width, the input value decreases by 0.1, and when the user presses button 77 corresponding to the input value for defect width, the input value increases by 0.1.

[0068] As described above, in this embodiment, the user can set two inspection conditions (first inspection conditions and second inspection conditions) using the image inspection setting screen 7. FIG. 6 shows an example of input for the first inspection conditions, and FIG. 7 shows an example of input for the second inspection conditions. For the first inspection conditions, the defect width is set to 0.8 mm and the defect height is set to 1.5 mm. For the second inspection conditions, the defect width is set to 0.4 mm and the defect height is set to 1.0 mm. Hereinafter, the parameter values ​​set as the first inspection conditions will be referred to as "first parameter values," and the parameter values ​​set as the second inspection conditions will be referred to as "second parameter values." That is, in this embodiment, two different judgment criteria, a first parameter value and a second parameter value, are prepared as inspection conditions. Note that the first parameter value is denoted by the symbol PA1, and the second parameter value is denoted by the symbol PA2. Furthermore, hereinafter, the inspection level corresponding to the first inspection condition (i.e., the inspection level corresponding to the first parameter value PA1) will be referred to as the "first inspection level," and the inspection level corresponding to the second inspection condition (i.e., the inspection level corresponding to the second parameter value PA2) will be referred to as the "second inspection level."

[0069] When the user has completed setting the inspection conditions (first inspection conditions and second inspection conditions) using the image inspection setting screen 7, the first parameter value PA1 and the second parameter value PA2 are stored in the inspection condition storage unit 63. In other words, the inspection condition setting unit 62 accepts the user's input of the first parameter value PA1 and the second parameter value PA2, and stores the accepted first parameter value PA1 and second parameter value PA2 in the inspection condition storage unit 63. When the inspection condition setting unit 62 displays the image inspection setting screen 7 on the display unit 36 ​​after the inspection conditions have already been set (i.e., when the inspection conditions are reset), the first parameter value PA1 and the second parameter value PA2 stored in the inspection condition storage unit 63 are read out. When the first inspection condition in the setting content list area 72 is selected, the read-out first parameter value PA1 is displayed in the parameter value input field 75, and when the second inspection condition in the setting content list area 72 is selected, the read-out second parameter value PA2 is displayed in the parameter value input field 75.

[0070] The defect image detection unit 64 detects defect images from the captured image based on the inspection conditions stored in the inspection condition storage unit 63. In this regard, the defect images are detected by, for example, image analysis based on the captured image data Di and comparison of the captured image data Di with the print data Dp transmitted from the print data generation device 40. Furthermore, the defect image detection process includes a process based on the first parameter value PA1 (hereinafter referred to as the "first detection process") and a process based on the second parameter value PA2 (hereinafter referred to as the "second detection process"). That is, the defect image detection unit 64 performs the first detection process to detect defect images from the captured image based on the first parameter value PA1 stored in the inspection condition storage unit 63, and the second detection process to detect defect images from the captured image based on the second parameter value PA2 stored in the inspection condition storage unit 63. As described above, the defect image detection unit 64 performs an inspection based on the first inspection conditions (main inspection) and an inspection based on the second inspection conditions (sub-inspection). Hereinafter, an image detected as a defect image by the first detection process will be referred to as a "first detected image," and an image detected as a defect image by the second detection process will be referred to as a "second detected image." The first detected image will be denoted by the symbol E1, and the second detected image will be denoted by the symbol E2. Note that the defect image detection step is realized by the operation of this defect image detection unit 64.

[0071] The result display unit 65 displays the test result screen on the display unit 36 ​​in response to a selection input IN by the user. The selection input IN refers to an input (input by the user) for selecting whether the test results to be displayed are the results of the main test or the results of the sub-test.

[0072] FIG. 8 is a diagram showing an example of an inspection result screen 390 displayed by the result display unit 65. As shown in FIG. 8, the inspection result screen 390 includes a thumbnail area 391, a sheet display area 392, an enlarged display area 393, a defect content display area 394, and a close button 395. The thumbnail area 391 displays a list of defect images, which are the first detected images E1 or the second detected images E2, in thumbnail format in response to a selection input IN by the user. Note that a portion marked with reference numeral 396 displays a sheet number identifying the sheet containing the corresponding defect image. The user can select one of the defect images displayed in the thumbnail area 391. The sheet display area 392 displays an image showing the entire sheet containing the defect image selected by the user. In the example shown in FIG. 8, the corresponding defect image is located at the intersection of a horizontal line 397 and a vertical line 398. That is, the horizontal line 397 and the vertical line 398 indicate the position of the defect image. The enlarged display area 393 displays an enlarged image of the defect image. The defect details are displayed in text in a defect details display area 394. A close button 395 is a button that the user presses to hide the inspection result screen 390. By referring to the inspection result screen 390, the user can confirm whether or not the defects detected by the inspection are acceptable.

[0073] The notification unit 66 notifies the user of the defect detection during or after the processing of the defect image detection unit 64, based on the defect images (first detected image E1, second detected image E2) detected by the defect image detection unit 64. An example of the notification by the notification unit 66 will be described below. The notification by the notification unit 66 is performed by displaying a message on the display unit 36. Different notifications are provided for the main inspection and the sub-inspection. For example, for the first detected image E1 detected in the main inspection, a notification is provided that the sheet containing the defect image will be treated as waste, and for the second detected image E2 detected in the sub-inspection, information about the defect image is provided as reference information. In this example, it is assumed that the inspection range of the main inspection is the range indicated by reference numeral 801 in FIG. 9, and the inspection range of the sub-inspection is the range indicated by reference numeral 802 in FIG. 9. In this case, a notification is issued that defects of high fatality corresponding to the range marked with reference numeral 801 will be treated as waste, and information on defects of high fatality corresponding to the range marked with reference numeral 802 and defects of low fatality corresponding to the range marked with reference numeral 802 is notified as reference information. Whether or not to treat defects notified as reference information as waste is determined, for example, by the user visually. The provision of the notification unit 66 as described above makes it possible, for example, to alert the user when a defect is detected, thereby reducing the occurrence of printing accidents due to overlooking defects.

[0074] In the above description, an example has been described in which the notification unit 66 notifies the user by displaying a message on the display unit 36. However, this is not limiting. Instead of or in addition to displaying a message on the display unit 36, the notification unit 66 may notify an external device, such as a post-processing machine, that performs a pre- or post-processing step in the printing process performed by the inkjet printing apparatus 10, by sounding a tower lamp provided on the inkjet printing apparatus 10 or the image inspection device 300 (i.e., audible notification). In this regard, the operation of the tower lamp may be controlled so that, for example, if a defect is detected in the main inspection, the tower lamp sounds, but if a defect is detected in the sub-inspection, the tower lamp does not sound. Furthermore, the operation of the post-processing machine may be controlled so that, for example, if a defect is detected in the main inspection, sheets containing the defect are discarded by the post-processing machine, but if a defect is detected in the sub-inspection, the sheet containing the defect is not discarded by the post-processing machine.

[0075] <1.4 Test results> Next, the inspection results obtained by the processing of the defect image detection unit 64 will be described. As described above, in this embodiment, it is possible to perform inspection to detect defects based on each of two inspection conditions (first inspection condition and second inspection condition). In this regard, the user can set the second inspection level higher or lower than the first inspection level. The images detected as defective images in the main inspection (first detected images E1) are the same whether the second inspection level is higher than the first inspection level or lower than the first inspection level. More specifically, regardless of whether the second inspection level is higher or lower than the first inspection level, in the main inspection, all images determined to be defective when the judgment criterion (criterion for determining whether an image to be inspected is a defective image) is the first inspection level are detected as defective images. In contrast, the images detected as defective images in the sub-inspection (second detected images E2) are different whether the second inspection level is higher than the first inspection level or lower than the first inspection level. This will be explained below.

[0076] Fig. 10 is a diagram for explaining a case where the second inspection level is higher than the first inspection level. In Fig. 10, L11 represents the first inspection level, and L12 represents the second inspection level. In this case, the range marked with reference numeral 811 is the inspection range during the main inspection, and the range marked with reference numeral 812 is the inspection range during the sub-inspection. Therefore, images that are determined to be defective when the above-mentioned judgment criterion is the second inspection level, excluding images that are determined to be defective when the above-mentioned judgment criterion is the first inspection level, are detected as defective images in the sub-inspection.

[0077] FIG. 11 is a diagram for explaining a case where the second inspection level is lower than the first inspection level. In FIG. 11, L13 represents the first inspection level, and L14 represents the second inspection level. In this case, the range marked with reference numeral 813 is the inspection range during the main inspection, and the range marked with reference numeral 814 is the inspection range during the sub-inspection. Therefore, all images that are determined to be defective when the above judgment criterion is the second inspection level are detected as defective images in the sub-inspection. However, in this case, as can be seen from FIG. 11, all images that are detected as defective images in the sub-inspection are also detected as defective images in the main inspection.

[0078] <1.5 Operation example> Next, an example of how the inspection is carried out in this embodiment will be described.

[0079] <1.5.1 First example> In a conventional image inspection device, if the inspection level is L20 in part A of Fig. 12, defects corresponding to the range marked with reference numeral 820 will be detected by inspection, but defects corresponding to the range marked with reference numeral 829 will not be detected by inspection. In other words, some defects with high fatality will not be detected by inspection.

[0080] Therefore, in the first example, the second inspection level is set to a level higher than the first inspection level so that inspection results can be obtained when inspection conditions are made stricter based on the main inspection. For example, the inspection conditions (the first parameter value PA1 and the second parameter value PA2) are set so that the first inspection level is L21 in part B of FIG. 12 and the second inspection level is L22 in part B of FIG. 12. As a result, in addition to the defects corresponding to the range marked with reference numeral 821 being detected in the main inspection, defects corresponding to the range marked with reference numeral 822 are detected in the sub-inspection. This allows the user to understand the defects that would be additionally detected if the inspection conditions of the main inspection were made stricter, and enables the user to determine the validity of the results of the main inspection. Then, the user refers to the results of the sub-inspection and adjusts the first parameter value PA1, for example, as shown in FIG. 13 (here, it is assumed that the second parameter value PA2 is maintained). As a result, for example, as shown in FIG. 14, the first inspection level, which was L21 before the adjustment, becomes L23 after the adjustment. Note that the second inspection level is maintained at L22. In the example shown in FIG. 14, after adjusting the first parameter value PA1, defects corresponding to the range indicated by the symbol 823 are detected by the main inspection. In other words, most of the critical defects are detected by the main inspection. In this way, by appropriately adjusting the first parameter value PA1, the results of the main inspection can be made closer to the ideal inspection results. More specifically, by adjusting the first parameter value PA1 as needed while continuing to perform inspections using the image inspection device 300, the results of the main inspection can be made gradually closer to the ideal inspection results.

[0081] <1.5.2 Second example> In a conventional image inspection device, if the inspection level is L30 in part A of Fig. 15, defects corresponding to the range marked with the reference numeral 830 will be detected by inspection. If an operation is adopted in which sheets containing defects are automatically discarded as waste, then sheets containing defects that are not critical will also be automatically discarded.

[0082] Therefore, in the second example, the second inspection level is set to a level lower than the first inspection level so that inspection results can be obtained when the inspection conditions are relaxed relative to the main inspection. For example, the inspection conditions (the first parameter value PA1 and the second parameter value PA2) are set so that the first inspection level is L31 in part B of FIG. 15 and the second inspection level is L32 in part B of FIG. 15 . In this case, in addition to detecting defects corresponding to the range marked with reference numeral 831 in the main inspection, defects corresponding to the range marked with reference numeral 832 in the sub-inspection are detected. This allows the user to understand the inspection results assuming that the inspection conditions of the main inspection are relaxed, and to determine the validity of the main inspection results. As in the first example, by adjusting the first parameter value PA1 as needed while continuing to perform inspections using the image inspection device 300, the main inspection results can gradually approach ideal inspection results.

[0083] <1.6 Effects> According to this embodiment, the user can set two inspection conditions (a first parameter value PA1 and a second parameter value PA2) for the inspection to detect defective images contained in printed images. The defective image detection unit 64 then performs a first detection process to detect defective images based on the first parameter value PA1 and a second detection process to detect defective images based on the second parameter value PA2. This makes it possible to obtain inspection results under two different inspection conditions. Therefore, for example, it is possible to determine the validity of the main inspection results, and the inspection conditions can be easily adjusted. As described above, the user can easily set suitable inspection conditions for the image inspection device that inspects printed images.

[0084] 2. Second embodiment A second embodiment of the present invention will be described below, focusing mainly on the differences from the first embodiment.

[0085] <2.1 Overview> In the first embodiment, after obtaining inspection results based on two inspection conditions (the main inspection result and the sub-inspection result), the user must adjust the inspection conditions by referring to the sub-inspection result, for example, to bring the main inspection result closer to the ideal inspection result. In this case, the user must visually confirm whether the defects presented as the sub-inspection result are acceptable, which requires costly confirmation work. Furthermore, because the defects detected vary depending on the print content, it is not always possible to set the first parameter value PA1 to an appropriate value within a short period of operation. Therefore, in this embodiment, instead of presenting all images detected by the defect image detection unit 64 as defect images, a process (hereinafter referred to as "filtering") is performed using artificial intelligence (AI) technology to narrow down (extract) images to be presented as defect images from among the images detected by the defect image detection unit 64.

[0086] The images detected as defective images by the defect image detection unit 64 include images that are not actually acceptable as a final product (hereinafter, for convenience, such an image state will be referred to as a "true defect") and images that are actually acceptable as a final product (hereinafter, for convenience, such an image state will be referred to as a "false defect"). Note that, hereinafter, a true defect image will be simply referred to as a "failure image," and a false defect image and an image not detected as a defect image by the defect image detection unit 64 will be referred to as a "good image." In the filtering process, an image detected as a defective image by the defect image detection unit 64 is determined to be a good image or a defective image. Then, only the defective images are finally presented as defect images.

[0087] 2.2 Configuration of image inspection equipment <2.2.1 Hardware configuration of image inspection computer> 16 is a hardware configuration diagram of the image inspection computer 301 in this embodiment. In this embodiment, an image discrimination model construction program 342 for constructing an image discrimination model that discriminates whether an image detected as a defective image by the first and second detection processes described above is a good image or a bad image is stored in the auxiliary storage device 34. Other points are the same as those in the first embodiment.

[0088] <2.2.2 Functional configuration> FIG. 17 is a block diagram showing the functional configuration of the image inspection device 300 in this embodiment. In this embodiment, as shown in FIG. 17, the defect image detection unit 64 includes a defect detection processing unit 641 and a filter unit 642. The defect detection processing unit 641 performs first detection processing and second detection processing. The filter unit 642 performs the above-mentioned filter processing. More specifically, the filter unit 642 extracts an image to be finally presented as a defect image from the images detected as defect images by the defect detection processing unit 641. In order to extract the image to be finally presented as a defect image, the filter unit 642 includes the above-mentioned image discrimination model. The image discrimination model discriminates whether each image detected as a defect image by the defect detection processing unit 641 is a good image or a bad image. Then, an image determined to be a bad image is finally treated as a defect image. Therefore, in this embodiment, among the images detected as defective images in the first detection process, an image determined to be a defective image by the image discrimination model becomes the first detected image E1, and among the images detected as defective images in the second detection process, an image determined to be a defective image by the image discrimination model becomes the second detected image E2. A detailed explanation of the image discrimination model will be given later. Note that, for convenience of explanation, the images detected as defective images by the first detection process and the second detection process will be referred to as "defect candidate images" below.

[0089] Furthermore, in addition to the components (see FIG. 4) provided in the first embodiment, the image inspection device 300 of this embodiment is provided with a re-learning execution unit 67. The image discrimination model included in the filter unit 642 is a trained artificial intelligence model, as will be described later, and the re-learning execution unit 67 causes the image discrimination model to re-learn. Re-learning will be described in detail later.

[0090] The components other than the "defect image detection unit 64 and the re-learning execution unit 67" are the same as those in the first embodiment.

[0091] <2.3 Image discrimination model> <2.3.1 Overview> As described above, the filter unit 642 (see FIG. 17) includes an image discrimination model that discriminates whether a defect candidate image is a good image or a bad image. A defect candidate image that is determined to be a bad image by the image discrimination model is ultimately treated as a defective image. Learning in the image discrimination model is performed using training data corresponding to bad images and training data corresponding to good images. For ease of explanation, the training data corresponding to bad images will be referred to as "first training data" and the training data corresponding to good images will be referred to as "second training data."

[0092] Generally, inkjet printing devices rarely produce printing defects such as ink stains and white streaks, making it difficult to secure a sufficient number of learning data sets for the first teacher data. Therefore, in this embodiment, pseudo-failure data representing ink stains, white streaks, and the like is prepared, and a large number of first teacher data sets are created by combining the pseudo-failure data in various ways with captured image data obtained by capturing a good printed image. Note that high-density pseudo-failure data and low-density pseudo-failure data are prepared as the pseudo-failure data. For example, pseudo-failure data representing ink stains corresponds to high-density pseudo-failure data, and pseudo-failure data representing white streaks corresponds to low-density pseudo-failure data.

[0093] Furthermore, in this embodiment, in order to ensure that defective parts (defective parts) in an image are focused on during discrimination by the image discrimination model, not only data corresponding to the discrimination target image (a defect candidate image detected by the first detection process or the second detection process) but also data corresponding to the original image of the discrimination target image are provided to the image discrimination model as input data. Note that in this specification, the term "discrimination target image" is used to refer not only to the image provided to the image discrimination model during discrimination but also to the image provided to the image discrimination model during learning.

[0094] <2.3.2 Overall process for building an image classification model> An image discrimination model must be constructed in advance to implement the filtering process by the filter unit 642. Therefore, the overall processing procedure for constructing an image discrimination model will be described with reference to FIG.

[0095] First, print data is prepared for printing by the inkjet printing device 10 (step S10). In step S10, for example, the print data generating device 40 performs RIP processing on input data such as a PDF file, thereby generating print data in bitmap format.

[0096] Next, based on the print data prepared in step S10, the inkjet printing device 10 prints out (step S20). Then, the image inspection device 300 inspects the print image, which is the result of the printout, based on predetermined inspection conditions (steps S30 and S40). In step S30, the camera (imaging device) 302 included in the image inspection device 300 captures (reads) the print image. This results in captured image data Di, which is data in RGB format. In step S40, as shown in FIG. 19, the image inspection computer 301 compares the captured image data Di with print data Dp converted from CMYK format to RGB format (hereinafter, the CMYK format print data and the print data converted from that to RGB format are collectively referred to as "original image data"). This results in defect information F, which is the inspection result. This defect information F includes data on images that are true defects (true defect data) F1 and data on images that are false defects (false defect data) F2. The defect information F and the original image data are sent from the image inspection device 300 to the print data generation device 40 as needed.

[0097] The processes of steps S10 to S40 are typically performed for a plurality of print data. That is, by repeating the processes of steps S10 to S40, defect information F based on a large number of print images is obtained.

[0098] After the defect information F is obtained, the image data to be combined is collected (step S50) as the data into which the pseudo-defective data is combined in step S70, which will be described later. In step S50, data of the printed image that was not determined to be a print defect (defective image) in the inspection by the image inspection device 300 and the false-defective data F2 included in the defect information F are collected as the image data to be combined. Whether the data included in the defect information F is true-defective data F1 or false-defective data F2 is determined by visual inspection. The data collected as the image data to be combined is image data of a portion of the entire printed image (entire page).

[0099] Next, pseudo-failure data is collected based on the true failure data F1 included in the failure information F (step S60). The pseudo-failure data is collected by, for example, extracting only the ink stained portion from the image of the true failure data F1 using image editing software in the image inspection computer 301 or the print data generation device 40. The color of the stain in the extracted portion is a mixture of the color of the ink causing the stain and the color of the background (original image). The image from which the true failure data F1 is generated is a printed image output from the inkjet printing device 10. In other words, in this embodiment, the pseudo-failure data (stain data) representing the ink stain is image data extracted from a printed image previously output from the inkjet printing device 10. Image data randomly generated using a graphic generation means (such as a program for automatically generating graphics) may also be used as the pseudo-failure data.

[0100] Thereafter, a process (model construction process) is performed to construct an image discrimination model using the to-be-combined image data collected in step S50 and the pseudo-failure data collected in step S60 (step S70). Details of step S70 will be described later.

[0101] <2.3.3 Structure of image discrimination model> Before describing the details of the model construction process (step S70 in FIG. 18), the structure of the image discrimination model will be described. Note that the structure described here is just an example, and the present invention is not limited to this.

[0102] Fig. 20 is a schematic diagram of an image discrimination model 500. As shown in Fig. 20, the image discrimination model 500 comprises a neural network unit 50 that performs machine learning, and a result output unit 58 that outputs a discrimination result 59 indicating whether the discrimination target image is a good image or a bad image. In this embodiment, the neural network unit 50 is realized by a convolutional neural network.

[0103] The neural network unit 50 is provided with input data such as "data combining captured image data and original image data" or "data combining original image data and data obtained by combining captured image data with pseudo-failure data." Hereinafter, "captured image data" and "data obtained by combining captured image data with pseudo-failure data" will be collectively referred to as "captured base data." The input data is in RGB format, and the input data for each color consists of captured base data and original image data. Therefore, as shown in FIG. 20 , the neural network unit 50 is provided with input data such as red captured base data 6_R1, red original image data 6_R2, green captured base data 6_G1, green original image data 6_G2, blue captured base data 6_B1, and blue original image data 6_B2.

[0104] As described above, six channels of data are input to the neural network unit 50. Each channel of data is made up of n (n is plural) pieces of pixel value data. For example, the red imaging base data 6_R1 is made up of n pieces of pixel value data 6_R1(1) to 6_R1(n) as shown in FIG. 20. As a result, (6×n) pieces of pixel value data are input to the neural network unit 50.

[0105] The neural network unit 50 outputs discrimination data Do for discriminating whether an image of the imaging base data is a good image or a bad image. The discrimination data Do is numerical data between 0 and 1. During learning, the values ​​of the parameters used in the neural network unit 50 (convolutional neural network) are updated by processing of an error backpropagation method based on the difference (typically, squared error) (see FIG. 21 ) between the value of the discrimination data Do and the value of the correct data Dc (for example, the value of correct data corresponding to a bad image is set to 1, and the value of correct data corresponding to a good image is set to 0). During discrimination, for example, if the value of the discrimination data Do is 0.5 or more, the result output unit 58 outputs a discrimination result 59 indicating that the discrimination target image is a bad image, and if the value of the discrimination data Do is less than 0.5, the result output unit 58 outputs a discrimination result 59 indicating that the discrimination target image is a good image.

[0106] FIG. 22 is a diagram illustrating an example of a detailed configuration of the neural network unit 50 in the image discrimination model 500. When the six-channel data (red imaging base data 6_R1, red original image data 6_R2, green imaging base data 6_G1, green original image data 6_G2, blue imaging base data 6_B1, and blue original image data 6_B2) described above is input to the neural network unit 50, the input data is subjected to a convolution operation using one or more sets of convolution filters 51. Note that one set of convolution filters 51 includes six filters, and one feature map 52 is obtained by the convolution operation using one set of convolution filters 51. For example, when three sets of convolution filters 51 are used, three feature maps 52 are obtained by the convolution operation. Then, a pooling operation is performed on each feature map 52, thereby obtaining pooled data 53 with reduced dimensions. The pooled data 53 obtained in this manner is provided to a fully connected layer 54, which outputs the discrimination data Do described above. Based on the value of this discrimination data Do, as described above, parameters are updated during learning, and discrimination results 59 for the discrimination target images are output during discrimination. Finally, as discrimination results 59, images determined to be defective are displayed on result display unit 65. When displaying images determined to be defective on result display unit 65, defective images and good images can be displayed in different display formats. Specifically, images determined to be defective may be displayed before images determined to be good. Images determined to be defective may be displayed with a mark and displayed in a different manner from images determined to be good, making it easier for the user to distinguish between defective images. Alternatively, only images determined to be defective may be displayed on result display unit 65, while images determined to be good may not be displayed on result display unit 65, thereby allowing defective images and good images to be displayed in different display formats. Using such a display format on result display unit 65 reduces the number of images that the user needs to visually check after the inspection is completed, thereby reducing the burden on the user.Furthermore, this allows the user to set the inspection conditions appropriately.

[0107] <2.3.4 Model construction process> Next, the procedure of the model construction process (step S70 in FIG. 18) will be described with reference to the flowchart shown in FIG. 23. The process described here is performed by executing the image discrimination model construction program 342 on the image inspection computer 301. However, this process may also be performed by the print data generation device 40, for example.

[0108] As described above, in this embodiment, a process is performed to create training data (first training data) corresponding to a defective image. This process is performed by combining pseudo-defective data with captured image data as the image data to be combined. In this regard, it is desirable that the first training data be data that is close to the defective image (defective image) that should actually be detected by the image inspection device 300. For example, a portion where ink stains adhere to a lightly inked background or a portion where ink is missing from a heavily inked background should be detected as a defective image (defective image).

[0109] Therefore, in order to vary the pseudo-failure data to be combined with the captured image data depending on the density of the background ink, first, it is determined whether the partial image (discrimination target image) to be combined with the pseudo-failure data in the print image used for learning is a high-density image or a low-density image (step S710). In step S710, the determination of whether the discrimination target image is a high-density image or a low-density image is made based on the captured image data corresponding to the discrimination target image. However, this determination may also be made based on the original image data corresponding to the discrimination target image. Since the captured image data is data in RGB format, the average values ​​of the pixels constituting the captured image data are calculated for each RGB color. That is, three average values ​​(average value of red, average value of green, and average value of blue) are calculated. Each of these three average values ​​is then compared with a predetermined threshold value, and whether the discrimination target image is a high-density image or a low-density image is determined depending on whether a predetermined condition is met.

[0110] If it is determined in step S710 that the discrimination target image is a high-density image, the process proceeds to step S721. On the other hand, if it is determined in step S710 that the discrimination target image is a low-density image, the process proceeds to step S722.

[0111] In step S721, first training data is generated by combining the white streak data as pseudo defect data with the captured image data. In step S721, for example, as shown in Fig. 24, the white streak data is combined with the captured image data 611 to generate first training data 612 including white streak data 613.

[0112] In step S722, first training data is generated by combining the dirt data as pseudo-failure data with the captured image data. The dirt data is prepared in advance in step S60 of Fig. 18. In step S722, for example, as shown in Fig. 25, the dirt data is combined with the captured image data 621 to generate first training data 622 including dirt data 623.

[0113] The print image from which the first training data is to be created is selected by the user. Based on the selected print image, the processes of steps S710, S721, and S722 are performed without user intervention. In other words, the first training data is created automatically.

[0114] In step S730, the original image data is added to the training data (first training data and second training data) used for learning in step S750 (described later). The reason for this processing is to focus attention on defective parts (defective parts) in the image during discrimination by the image discrimination model 500, as described above.

[0115] Next, a process called data augmentation is performed on the training data (step S740). More specifically, the number of training data is increased by performing conversion processes such as inversion, enlargement, and reduction on the training data images present at the start of step S740. Note that the process of step S740 is not necessarily required, but increasing the number of training data in this way has the effect of improving robustness.

[0116] Typically, the processing of steps S710 to S740 is repeated the same number of times as the number of pieces of image data to be combined collected in step S50 of Fig. 18. In this way, a large number of pieces of first teacher data to be used for learning are created.

[0117] Thereafter, training data (first training data and second training data) are sequentially provided to the neural network unit 50, thereby performing learning (machine learning) (step S750). As a result, the parameters of the neural network unit 50 are optimized, and the image discrimination model 500 is constructed.

[0118] <2.3.5 Adding original image data to training data> Next, the process of adding original image data to the training data (first training data and second training data) (the process of step S730 in FIG. 23) will be described in detail.

[0119] When ink stains occur on multiple sheets of paper when the data printed by the inkjet printing device 10 is variable data, the background typically differs from one sheet to another. Furthermore, the ink stains also vary in shape and color. For these reasons, even when an artificial intelligence model is trained on an image, the training may not focus on the ink stains. For example, the training may focus on a specific design in the background and result in a determination that an image is truly defective even when it is not. With conventional methods, the background significantly affects the training results, preventing accurate discrimination between ink stains and white streaks, which are truly defective, and images that are not truly defective.

[0120] Therefore, in this embodiment, as described above, the image discrimination model 500 is provided with not only data corresponding to the discrimination target image as input data but also raw image data for the discrimination target image so that the image discrimination model 500 can perform discrimination while focusing on defective portions (defective portions) in the image. That is, prior to learning, a process of adding the raw image to the discrimination target image is performed. This process is performed for both the first training data and the second training data. In this embodiment, the first training data is created by combining captured image data with pseudo-defective data. Therefore, the addition of the raw image data to the first training data is performed by adding raw image data 652 to imaging base data 651, which is data obtained by combining captured image data with pseudo-defective data 659, as shown in FIG. 26. In this manner, input data 653 to the image discrimination model 500 is obtained. The addition of the raw image data to the second training data is performed by adding the raw image data to imaging base data, which is the captured image data itself. In this manner, the background is prevented from affecting the learning results.

[0121] The input data 663 obtained by adding the original image data 662 to the imaging base data 661 is provided to the neural network section 50 in the image discrimination model 500 as six-channel data consisting of red imaging base data 6_R1, red original image data 6_R2, green imaging base data 6_G1, green original image data 6_G2, blue imaging base data 6_B1, and blue original image data 6_B2, as shown in Figure 27.

[0122] As described above, by adding original image data corresponding to the actual printed image to be obtained to the training data, learning (learning in the image discrimination model 500) is performed while focusing on the difference between the image to be discriminated and the original image.

[0123] <2.3.6 Re-learning> This image inspection device 300 obtains inspection results based on two inspection conditions (main inspection result and sub-inspection result), and for example, defects presented as sub-inspection results are visually confirmed to be acceptable or unacceptable. Depending on the operation, defects presented as main inspection results are also visually confirmed to be acceptable or unacceptable. In this embodiment, the image inspection device 300 is provided with a re-learning execution unit 67 so that the image discrimination model 500 can be re-trained based on the results obtained by such visual confirmation (see FIG. 17 ).

[0124] In the image inspection device 300 of this embodiment, in order to cause the image discrimination model 500, which is an artificial intelligence model, to perform re-learning, the user can select any defect image displayed on the inspection result screen 390 and input a judgment result RD indicating whether the selected defect image SE is a good image or a bad image. Then, the defect image SE selected by the user and the judgment result RD are provided to the image discrimination model 500 included in the filter unit 642.

[0125] More specifically, during relearning, a combination of data on the defect image selected by the user and the corresponding original image data is provided as input data to the neural network unit 50 constituting the image discrimination model 500. As a result, discrimination data Do is output from the neural network unit 50. A value corresponding to the above-mentioned judgment result RD obtained by visual confirmation is provided to the image discrimination model 500 as correct answer data Dc. Then, the values ​​of the parameters used in the neural network unit 50 are updated by processing using an error backpropagation method based on the difference (typically, squared error) between the value of the discrimination data Do and the value of the correct answer data Dc. Note that, as the value corresponding to the above-mentioned judgment result RD, for example, 1 is adopted if the corresponding defect image is determined to be a bad image, and 0 is adopted if the corresponding defect image is determined to be a good image.

[0126] <2.4 Effects> According to this embodiment, a filter unit 642 including an image discrimination model 500, which is an artificial intelligence model, performs a filtering process to narrow down the images detected by the defect detection processing unit 641 to those that will ultimately be presented as defect images. This reduces the number of images that the user needs to visually check after the print image inspection is completed compared to the first embodiment, thereby reducing the burden on the user. Furthermore, since a filtering process is performed to narrow down the images detected by the defect detection processing unit 641 to those that will ultimately be presented as defect images, it becomes easier to appropriately adjust the first parameter value PA1. Furthermore, by re-learning the image discrimination model 500, it becomes possible to bring the results of the main inspection closer to the ideal inspection results with fewer operations.

[0127] Note that the first parameter value PA1 may be automatically adjusted by the image inspection program 341 using data provided to the image discrimination model 500 for re-learning, for example.

[0128] <3. Modifications> The following describes modified examples.

[0129] <3.1 First modified example> In the second embodiment, both the image detected as a defect image by the first detection process and the image detected as a defect image by the second detection process were filtered by the filter unit 642 (i.e., filtered using the image discrimination model 500, which is an artificial intelligence model). However, discrimination using an artificial intelligence model increases the risk of erroneous judgment, especially when learning is insufficient. Furthermore, since the results obtained using an artificial intelligence model cannot be theoretically explained, many users do not prefer to use discrimination using an artificial intelligence model.

[0130] Therefore, in this modification, filter unit 642 (see FIG. 17) does not filter images detected as defective images by the first detection process, but filters only images detected as defective images by the second detection process. Therefore, in this modification, all images detected as defective images by the first detection process become first detected images E1, and images detected as defective images by the second detection process that are determined to be defective images by image discrimination model 500 become second detected images E2.

[0131] According to this modification, the results of the main inspection are not filtered using an artificial intelligence model, thereby reducing the occurrence of printing accidents due to erroneous judgments by the artificial intelligence model. Furthermore, by adjusting the first parameter value PA1 as needed while continuing to operate the inspection using the image inspection device 300, the inspection range in which the artificial intelligence model performs judgment can be narrowed. By adjusting the first parameter value PA1 as needed in this way, the occurrence of printing accidents due to erroneous judgments by the artificial intelligence model can also be reduced.

[0132] As in the second embodiment described above, in this modified example, by re-learning the image discrimination model 500, it is possible to bring the results of the main inspection closer to the ideal inspection results with fewer operations.

[0133] <3.2 Second modified example> In each of the above embodiments, it was assumed that the notification unit 66 would issue a certain notification depending on the defect image detection results by the defect image detection unit 64. However, even when similar defects are detected, the notification content desired by different users varies. Some users also desire different notification content depending on the task. Furthermore, some users desire notifications for the results of sub-inspections that are similar to the results of the main inspection. Therefore, in this modified example, the image inspection device 300 is provided with a function that allows the user to select from multiple levels the content of the notification to be issued by the notification unit 66.

[0134] Fig. 28 is a block diagram showing the functional configuration of an image inspection device 300 in this modified example. In addition to the components provided in the first embodiment (see Fig. 4), the image inspection device 300 in this modified example is provided with a notification content change unit 68. Note that the notification content change unit 68 may be provided in addition to the components provided in the second embodiment (see Fig. 17).

[0135] The notification content change unit 68 displays a screen 670 (hereinafter referred to as the "notification level setting screen") on the display unit 36, which allows the user to set a notification level that determines the notification content, and accepts input from the user. FIG. 29 is a diagram showing an example of the notification level setting screen 670. This notification level setting screen 670 includes a notification level selection area 671 and a close button 672. The notification level selection area 671 has three check boxes corresponding to three notification levels ("weak," "medium," and "strong"). The user selects a notification level by clicking one of the check boxes. Note that FIG. 29 shows an example in which "weak" has been selected. The close button 672 is a button that the user presses when finishing selecting a notification level using this notification level setting screen 670.

[0136] As described above, the notification content change unit 68 receives the user's selection of the notification level LV. Then, the notification content change unit 68 provides the received notification level LV to the notification unit 66. The notification unit 66 changes the notification content according to the notification level LV provided by the notification content change unit 68.

[0137] Below, we will explain an example of how to change the notification content (change the notification level). Here, three notification modes are available from the notification unit 66: "notifying the post-processing machine of information," "displaying a message on the display unit 36," and "ringing the tower lamp." Also, by changing the notification level LV, it is possible to change the notification content for the results of the sub-inspection.

[0138] When the notification level LV is set to "weak," the notification unit 66 issues the following notification: Even if a defective image is detected in the sub-inspection, the sheet containing the defective image is not treated as waste by the post-processing machine (i.e., the sheet containing the defective image is not discarded). When a defective image is detected in the sub-inspection, information about the defective image is displayed on the display unit 36 ​​as reference information. Even if a defective image is detected in the sub-inspection, the tower lamp does not sound.

[0139] When the notification level LV is set to "medium," the notification unit 66 issues the following notifications: Even if a defective image is detected during the sub-inspection, the sheet containing the defective image will not be treated as waste by the post-processing machine (i.e., the sheet containing the defective image will not be discarded). When a defective image is detected during the sub-inspection, information about the defective image is displayed on the display unit 36 ​​as reference information. When a defective image is detected during the sub-inspection, the tower lamp will sound.

[0140] When the notification level LV is set to "strong," the notification unit 66 issues the following notification: If a defective image is detected during the sub-inspection, the sheet containing the defective image is treated as waste by the post-processing machine (i.e., the sheet containing the defective image is discarded). If a defective image is detected during the sub-inspection, a message indicating that the sheet containing the defective image will be treated as waste is displayed on the display unit 36. If a defective image is detected during the sub-inspection, the tower lamp sounds.

[0141] As described above, according to this modification, the user can select from a plurality of levels the content of the notification to be sent by the notification unit 66 when a defect is detected.

[0142] <3.3 Third modified example> In the above-described embodiments, both the first parameter value PA1 and the second parameter value PA2 are set by the user. In other words, the user is required to set the inspection conditions for the main inspection and the inspection conditions for the sub-inspection. However, setting multiple inspection conditions places a heavy burden on the user. Therefore, in this modified example, the second parameter value PA2 is set by the inspection condition setting unit 62. Note that, while the above-described embodiments associated the first parameter value PA1 and the second parameter value PA2 with the inspection level, this modified example will be described by associating the first parameter value PA1 and the second parameter value PA2 with the criticality. The higher the inspection level, the lower the corresponding criticality, and the lower the inspection level, the higher the corresponding criticality.

[0143] First, a case where the fatality level corresponding to the second parameter value PA2 is lower than the fatality level corresponding to the first parameter value PA1 (hereinafter referred to as the "first case") will be described. The case shown in Fig. 10 corresponds to the first case. That is, in the first case, the fatality level corresponding to the second parameter value PA2 is lower than the fatality level corresponding to the first parameter value PA1, the first detection process detects a defective image having a fatality level higher than the fatality level corresponding to the first parameter value PA1, and the second detection process detects a defective image having a fatality level higher than the fatality level corresponding to the second parameter value PA2 and lower than the fatality level corresponding to the first parameter value PA1.

[0144] In the first case, a value corresponding to the lowest criticality among the values ​​that can be set as the second parameter value PA2 (hereinafter referred to as the "minimum limit value") is internally predetermined. The first parameter value PA1 is set by the user. Here, it is assumed that L41 in FIG. 30 is the criticality corresponding to the first parameter value PA1 set by the user. It is also assumed that L49 in FIG. 30 is the criticality corresponding to the predetermined minimum limit value. Instead of setting the second parameter value PA2, the user must specify a range level that indicates the width of the inspection range of the sub-inspection (in other words, the target range in which defect images are detected by the second detection process). Therefore, in this modified example, the image inspection setting screen 7 is configured to allow the user to specify (input) the range level. It is assumed here that three range levels ("narrow," "medium," and "wide") are available.

[0145] Under the above assumptions, the second parameter value PA2 is set by the inspection condition setting unit 62 as follows, depending on the range level specified by the user. Note that, regardless of the range level specified by the user, the range marked with reference symbol 841 in FIG. 31 becomes the inspection range for the main inspection. When "narrow" is specified by the user, the second parameter value PA2 is set so that the criticality L42a, which corresponds to a position one-third of the way from the criticality L41 corresponding to the first parameter value PA1 to the criticality L49 corresponding to the minimum limit value, becomes the criticality corresponding to the second parameter value PA2 (see part A of FIG. 31). In this case, the range marked with reference symbol 842a becomes the inspection range for the sub-inspection. When "medium" is specified by the user, the second parameter value PA2 is set so that the criticality L42b, which corresponds to a position two-thirds of the way from the criticality L41 corresponding to the first parameter value PA1 to the criticality L49 corresponding to the minimum limit value, becomes the criticality corresponding to the second parameter value PA2 (see part B of FIG. 31). In this case, the range marked with symbol 842b becomes the inspection range of the sub-inspection. If the user specifies "wide," the second parameter value PA2 is set so that the criticality L49 corresponding to the minimum limit value matches the criticality L42c corresponding to the second parameter value PA2 (see part C of FIG. 31). In this case, the range marked with symbol 842c becomes the inspection range of the sub-inspection.

[0146] The first case will be further described using a specific example. While multiple parameter values ​​are actually set as the second parameter value PA, for convenience of explanation, only the defect width value will be focused on here. For example, assume that the minimum limit value is preset to 0.3 mm and the defect width (first parameter value PA1) under the first inspection conditions is set to 0.9 mm by the user. In this case, the defect width (second parameter value PA2) under the second inspection conditions is set by the inspection condition setting unit 62 as follows, depending on the range level specified by the user: If the user specifies "narrow," the defect width under the second inspection conditions is set to 0.7 mm. In this case, the range of defect width for the sub-inspection is set to 0.7 mm to 0.9 mm. If the user specifies "medium," the defect width under the second inspection conditions is set to 0.5 mm. In this case, the range of defect width for the sub-inspection is set to 0.5 mm to 0.9 mm. If the user specifies "wide," the defect width under the second inspection conditions is set to 0.3 mm. In this case, the range of defect widths from 0.3 mm to 0.9 mm is the inspection range for the sub-inspection.

[0147] As described above, in the first case, a minimum limit value, which is a value corresponding to the lowest criticality among the values ​​that can be set as the second parameter value PA2, is predetermined. Then, the inspection condition setting unit 62 receives an external input of the first parameter value PA1 and an external input of a range level that indicates the width of the target range in which defect images are to be detected by the second detection process, and sets the second parameter value PA2 to a value between the first parameter value PA1 and the minimum limit value in accordance with the externally input range level.

[0148] Next, a case where the fatality level corresponding to the second parameter value PA2 is higher than the fatality level corresponding to the first parameter value PA1 (hereinafter referred to as the "second case") will be described. The case shown in Fig. 11 corresponds to the second case. That is, in the second case, the fatality level corresponding to the second parameter value PA2 is higher than the fatality level corresponding to the first parameter value PA1, and the first detection process detects a defective image having a fatality level higher than the fatality level corresponding to the first parameter value PA1, and the second detection process detects a defective image having a fatality level higher than the fatality level corresponding to the second parameter value PA2.

[0149] In the second case, a value corresponding to the highest criticality among the values ​​that can be set as the second parameter value PA2 (hereinafter referred to as the "maximum limit value") is internally predetermined. The first parameter value PA1 is set by the user. Here, it is assumed that L51 in FIG. 32 is the criticality corresponding to the first parameter value PA1 set by the user. It is also assumed that L59 in FIG. 32 is the criticality corresponding to the predetermined maximum limit value. As in the first case, instead of setting the second parameter value PA2, the user must specify a range level that indicates the width of the inspection range of the sub-inspection. Here too, it is assumed that three range levels ("narrow," "medium," and "wide") are provided.

[0150] Under the above assumptions, the second parameter value PA2 is set by the inspection condition setting unit 62 as follows, depending on the range level specified by the user. Note that, regardless of the range level specified by the user, the range marked with reference numeral 851 in FIG. 33 becomes the inspection range for the main inspection. When the user specifies "narrow," the second parameter value PA2 is set so that the criticality L59 corresponding to the maximum limit value and the criticality L52a corresponding to the second parameter value PA2 coincide (see part A of FIG. 33). At this time, the range marked with reference numeral 852a becomes the inspection range for the sub-inspection. When the user specifies "medium," the second parameter value PA2 is set so that the criticality L52b corresponding to one-third of the way from the criticality L59 corresponding to the maximum limit value to the criticality L51 corresponding to the first parameter value PA1 becomes the criticality corresponding to the second parameter value PA2 (see part B of FIG. 33). At this time, the range marked with reference numeral 852b becomes the inspection range for the sub-inspection. If the user specifies "wide," the second parameter value PA2 is set so that the criticality L52c, which corresponds to a position two-thirds of the way from the criticality L59 corresponding to the maximum limit value to the criticality L51 corresponding to the first parameter value PA1, becomes the criticality corresponding to the second parameter value PA2 (see part C of FIG. 33). In this case, the range marked with symbol 852c becomes the inspection range of the sub-inspection.

[0151] The second case will be further explained using a specific example. Note that, again, we will focus only on the defect width value as a parameter value. For example, assume that the maximum limit value is preset to 1.5 mm and the defect width under the first inspection condition (first parameter value PA1) is set to 0.3 mm by the user. In this case, the defect width under the second inspection condition (second parameter value PA2) is set by the inspection condition setting unit 62 as follows, depending on the range level specified by the user: If the user specifies "narrow," the defect width under the second inspection condition is set to 1.5 mm. In this case, the range specified for the defect width as "1.5 mm or more" becomes the inspection range for the sub-inspection. If the user specifies "medium," the defect width under the second inspection condition is set to 1.1 mm. In this case, the range specified for the defect width as "1.1 mm or more" becomes the inspection range for the sub-inspection. If the user specifies "wide," the defect width under the second inspection condition is set to 0.7 mm. In this case, the range specified as "0.7 mm or more" for the defect width becomes the inspection range for the sub-inspection.

[0152] As described above, in the second case, a maximum limit value, which is a value corresponding to the highest criticality among the values ​​that can be set as the second parameter value PA2, is predetermined. Then, the inspection condition setting unit 62 receives an external input of the first parameter value PA1 and an external input of a range level that indicates the width of the target range in which defect images are to be detected by the second detection process, and sets the second parameter value PA2 to a value between the first parameter value PA1 and the maximum limit value in accordance with the externally input range level.

[0153] Next, a third case will be described in which the user specifies the magnitude relationship between the criticality corresponding to the first parameter value PA1 and the criticality corresponding to the second parameter value PA2. In the third case, the minimum limit value and the maximum limit value are internally determined in advance. The first parameter value PA1 is set by the user. The range level may be specified by the user as in the first and second cases, or may be determined internally.

[0154] Under the above assumptions, the second parameter value PA2 is set by the inspection condition setting unit 62 as follows, depending on the magnitude relationship specified by the user. Note that the inspection range of the main inspection is the same as in the first and second cases. If the user specifies that the criticality corresponding to the second parameter value PA2 is lower than the criticality corresponding to the first parameter value PA1, the second parameter value PA2 is set in the same manner as in the first case. If the user specifies that the criticality corresponding to the second parameter value PA2 is higher than the criticality corresponding to the first parameter value PA1, the second parameter value PA2 is set in the same manner as in the second case.

[0155] As described above, in the third case, a minimum limit value, which is a value corresponding to the lowest criticality among the values ​​that can be set as the second parameter value PA2, and a maximum limit value, which is a value corresponding to the highest criticality among the values ​​that can be set as the second parameter value PA2, are predetermined. The inspection condition setting unit 62 then receives an external input of the first parameter value PA1 and an external input of the magnitude relationship between the criticality corresponding to the first parameter value PA1 and the criticality corresponding to the second parameter value PA2, and sets the second parameter value PA2 based on the external input. [Explanation of symbols]

[0156] 7...Image inspection setting screen 10...Inkjet printing device 40...print data generating device 50...Neural Network Department 61...imaging unit 62...Inspection condition setting section 63...Inspection condition memory section 64...Defect image detection unit 65…Result display area 66…Notification section 67...Relearning execution unit 68...Notification content change section 100...printing control device 200...printing machine body 300...Image inspection device 301...Image inspection computer 302...Camera (imaging device) 390...Test result screen 500...Image discrimination model 641...defect detection processing section 642...Filter section 670...Notification level setting screen Di...Captured image data Dp...print data PA1, PA2...First parameter value, second parameter value

Claims

1. An image inspection device that performs an inspection to detect defective images contained in a printed image, an inspection condition setting unit for setting a judgment criterion for detecting a defect image as an inspection condition; an imaging unit that captures an image of the print image and outputs the captured image; a defect image detection unit that detects a defect image from the captured image based on the inspection conditions; Equipped with the inspection condition setting unit is configured to be able to set, as the inspection conditions, a first parameter value and a second parameter value, which are two different judgment criteria for the same inspection item; The first parameter value is a combination of parameter values ​​for each of a plurality of test items, the second parameter value is a combination of parameter values ​​for each of the plurality of test items, the inspection condition setting unit displays an inspection setting screen for setting the first parameter value and the second parameter value, and accepts input of the first parameter value and the second parameter value from an external device; The image inspection device is characterized in that the defect image detection unit performs a first detection process to detect a defect image from the captured image based on the first parameter value and a second detection process to detect a defect image from the captured image based on the second parameter value.

2. 2. The image inspection device according to claim 1, further comprising a result display unit that displays the defect image detected by said defect image detection unit as a result of the inspection.

3. The defect image detection unit a defect detection processing unit that performs the first detection processing and the second detection processing; a filter unit including a trained artificial intelligence model that determines whether a given image is a good image or a bad image, and that extracts some defect images from the defect images detected by the defect detection processing unit; It consists of The artificial intelligence model is provided with a defect image detected by the defect detection processing unit, 3. The image inspection device according to claim 1, wherein the filter unit extracts defective images determined to be defective images by the artificial intelligence model.

4. 4. The image inspection device according to claim 3, wherein both the defect image detected in the first detection process and the defect image detected in the second detection process are provided to the artificial intelligence model.

5. 4. The image inspection device according to claim 3, wherein the artificial intelligence model is provided with only one of the defect images detected in the first detection process and the defect images detected in the second detection process.

6. 6. The image inspection device according to claim 3, further comprising a re-learning execution unit that causes the artificial intelligence model to perform re-learning by providing the artificial intelligence model with a defect image selected from the defect images detected by the defect image detection unit and a determination result indicating whether the selected defect image is a good image or a bad image.

7. an inspection condition storage unit that stores the first parameter value and the second parameter value set by the inspection condition setting unit; The image inspection device described in any one of claims 1 to 6, characterized in that the inspection condition setting unit is configured to read out the first parameter value and the second parameter value stored in the inspection condition memory unit after the first detection process and the second detection process are completed, and to be able to reset the first parameter value and the second parameter value.

8. 8. The image inspection device according to claim 1, further comprising a notification section that notifies the user of content corresponding to the defect image detected by the defect image detection section.

9. 9. The image inspection device according to claim 8, wherein the notification unit notifies an external device that executes a process before or after the printing process performed by the printing device that printed the print image.

10. 10. The image inspection device according to claim 8, wherein the notification unit performs notification by displaying a message on a predetermined display unit.

11. 11. The image inspection device according to claim 8, wherein the notification unit issues a notification by voice.

12. 12. An image inspection device as described in any one of claims 8 to 11, characterized in that it is configured to be able to select one of a plurality of pre-prepared notification levels, and further comprises a notification content change unit that changes the content of notification by the notification unit in accordance with the selected notification level.

13. 13. The image inspection device according to claim 12, wherein the notification content change unit changes the content of the notification by the notification unit only for one of the notification based on the defect image detected by the first detection process and the notification based on the defect image detected by the second detection process.

14. the criticality corresponding to the second parameter value is lower than the criticality corresponding to the first parameter value; a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; 14. The image inspection device according to claim 1, wherein the second detection process detects a defect image having a fatality level higher than the fatality level corresponding to the second parameter value and lower than the fatality level corresponding to the first parameter value.

15. a minimum limit value, which is a value corresponding to the lowest criticality among the values ​​that can be set as the second parameter value, is predetermined; 15. The image inspection device according to claim 14, wherein the inspection condition setting unit receives an external input of the first parameter value and an external input of a range level indicating a range of criticality for detecting defect images by the second detection process, and sets the second parameter value to a value between the first parameter value and the minimum limit value in accordance with the externally input range level.

16. the criticality corresponding to the second parameter value is higher than the criticality corresponding to the first parameter value; a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; 14. The image inspection device according to claim 1, wherein the second detection process detects a defect image having a higher fatality level than the fatality level corresponding to the second parameter value.

17. a maximum limit value, which is a value corresponding to the highest criticality among the values ​​that can be set as the second parameter value, is predetermined; 17. The image inspection device according to claim 16, wherein the inspection condition setting unit receives an external input of the first parameter value and an external input of a range level indicating a range of criticality for detecting defect images by the second detection process, and sets the second parameter value to a value between the first parameter value and the maximum limit value in accordance with the externally input range level.

18. a minimum limit value, which is a value corresponding to the lowest fatality level among the values ​​that can be set as the second parameter value, and a maximum limit value, which is a value corresponding to the highest fatality level among the values ​​that can be set as the second parameter value, are determined in advance; the inspection condition setting unit receives an external input of the first parameter value and an external input of a magnitude relationship between the criticality corresponding to the first parameter value and the criticality corresponding to the second parameter value, and sets the second parameter value based on the contents input from the outside; If the criticality corresponding to the second parameter value set by the inspection condition setting unit is lower than the criticality corresponding to the first parameter value input from the outside, a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; a defect image having a criticality higher than the criticality corresponding to the second parameter value and lower than the criticality corresponding to the first parameter value is detected by the second detection process; If the fatality level corresponding to the second parameter value set by the inspection condition setting unit is higher than the fatality level corresponding to the first parameter value input from the outside, a defect image having a higher fatality level than the fatality level corresponding to the first parameter value is detected by the first detection process; 14. The image inspection device according to claim 1, wherein the second detection process detects a defect image having a higher fatality level than the fatality level corresponding to the second parameter value.

19. a conveying unit that conveys the substrate; a printing unit that performs printing based on print data on the base material transported by the transport unit; The image inspection device according to any one of claims 1 to 18, further comprising the imaging unit arranged to be able to capture the printed image obtained by printing by the printing unit; A printing device comprising:

20. An image inspection method for detecting defective images contained in a printed image, comprising: an inspection condition setting step in which an operator sets criteria for detecting defect images as inspection conditions; an imaging step in which an imaging device captures the print image; a defect image detection step in which a computer detects a defect image from the captured image obtained in the imaging step based on the inspection conditions; Including, In the inspection condition setting step, a first parameter value and a second parameter value, which are two different judgment criteria for the same inspection item, are set by an operator as the inspection conditions; The first parameter value is a combination of parameter values ​​for each of a plurality of test items, the second parameter value is a combination of parameter values ​​for each of the plurality of test items, In the inspection condition setting step, an inspection setting screen for setting the first parameter value and the second parameter value is displayed, and input of the first parameter value and the second parameter value by the operator is accepted; An image inspection method characterized in that the defect image detection step includes performing a first detection process to detect a defect image from the captured image based on the first parameter value and a second detection process to detect a defect image from the captured image based on the second parameter value.

Images