JP2023125203A - Method and device for generating learned model for printed matter inspection system - Google Patents

Abstract

To improve accuracy of inspecting a printed matter.SOLUTION: An inspection system comprises: a printing machine that executes printing on the basis of print data; a camera that captures a printed matter printed by the printing machine to generate a camera image; and a processing device that converts the print data into a print image for collation using a learned model preliminarily trained by machine learning and collates the print image for collation with a camera image for collation based on the camera image to determine whether the printed matter is good. A method for generating the learned model for the inspection system includes: acquiring a data set for learning including learning print data indicating an image of a test chart including a plurality of color patterns and a learning camera image generated by photographing, with the camera, the printed matter printed by the printing machine on the basis of the learning print data; and generating the learned model by machine learning based on the data set for learning.SELECTED DRAWING: Figure 16

Description

Patent Act Article 30, Paragraph 2 application filed Sale date: November 30, 2021 Place of sale: Fujifilm Business Innovation Co., Ltd. Ebina Office (2274 Hongo, Ebina City, Kanagawa Prefecture 243-0417) Sale date: December 24, 2021 Place of sale: Kyodo Printing Co., Ltd. Information Output Center (1427-2 Oaza Omido, Kamisato-cho, Kodama-gun, Saitama Prefecture) Date of sale: February 14, 2021 Place of sale: Co., Ltd. TLP Hanyu Factory (2-69 Onuma, Hanyu, Saitama 348-0016)

The present invention relates to a method and apparatus for generating a trained model for a printed matter inspection system.

2. Description of the Related Art Various inspection devices have been developed and used that inspect the quality or quality of printing by comparing image data obtained by photographing printed matter with a camera and data from the printing source.

For example, Patent Document 1 discloses an inspection device that performs inspection based on the difference between a master image generated from print data and a read image generated by reading a printed matter with a reading device. It is stated that by generating a master image from print data, it is possible to efficiently inspect printed matter using variable data printing (VDP), for example.

Patent Document 2 reads out and compares an image to be inspected obtained by photographing a printed matter and a reference image generated in advance for each unit of printed matter, and if a discrepancy is found, it is determined that an error has occurred in printing. Discloses the testing method.

Patent Document 3 discloses an inspection device that inspects the printed content and printing condition of a color printed matter in which different data is printed on a page-by-page basis. This inspection device inspects the print content and printing condition by comparing and checking a background pattern common to printed matter and variable data that differs for each printed matter with a reference image.

Patent Document 4 discloses a method for determining the quality of printed matter by comparing reference data for comparison generated by converting reference data according to the conditions of the printed matter and imaged data for comparison based on imaged data obtained by imaging the printed matter. Discloses an inspection method for making the determination. The reference data for comparison is converted from the reference data using a learning model generated by applying deep learning. It is described that this makes it possible to determine the quality of printed matter using reference data according to the conditions of the printed matter.

Japanese Patent Application Publication No. 2015-174307 JP2019-132661A Japanese Patent Application Publication No. 2003-54096 JP2020-186938A

In conventional print inspection methods, inspection accuracy may be affected due to differences in the format of the image data obtained by photographing the print and the original data for printing, ink bleeding during printing, deformation of the paper, etc. There were cases where the value decreased.

The present disclosure provides a novel inspection technique that can improve the accuracy of image inspection of printed matter.

An inspection system according to one aspect of the present invention is used in combination with a printing machine that includes a print controller that converts input printing source data into head data, and a print head that prints based on the head data. . The inspection system includes a camera that takes a picture of a printed matter printed by the printing machine to generate a camera image, and a processing device. The processing device performs a first spatial filtering process on the print source data or the head data to generate a first filter image. The processing device matches the first filter image with the camera image, or matches the first filter image with a second filter image generated by performing a second spatial filtering process on the camera image. By doing so, the quality of the printed matter is determined and the determination result is output.

A method according to another aspect of the present invention includes: a printing machine that executes printing based on print data; a camera that generates a camera image by photographing printed matter printed by the printing machine; The quality of the printed matter is determined by converting it into a verification print image using a learned model trained in advance through learning, and comparing the verification print image with a verification camera image based on the camera image. A method of generating the learned model for an inspection system comprising a processing device. The method includes learning print data showing an image of a test chart including a plurality of color patterns, and learning print data generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. The method includes obtaining a learning data set including a camera image, and generating the learned model by machine learning based on the learning data set.

General or specific aspects of the invention may be implemented by an apparatus, a system, an integrated circuit, a computer program, a recording medium, or any combination thereof.

According to the present invention, it is possible to improve the accuracy of image inspection of printed matter.

1 is a diagram schematically showing an example of a conventional inspection system. FIG. 3 is a diagram schematically showing that paper and printed parts are deformed by printing. 1 is a diagram schematically showing the configuration of an inspection system according to an exemplary embodiment of the present invention. FIG. 2 is a schematic diagram for explaining an overview of determination processing by a processing device. FIG. 2 is a diagram for explaining the functions of the inspection system. FIG. 3 is a diagram schematically showing the flow of printing and inspection. FIG. 3 is a diagram schematically showing an outline of color conversion processing. 3 is a flowchart illustrating an example of processing executed by a processing device. It is a flowchart which shows another example of the process which a processing apparatus performs. It is a flowchart which shows yet another example of the process which a processing apparatus performs. FIG. 3 is a diagram showing the flow of processing for converting head data into a first filter image. FIG. 3 is a diagram showing an example of a pixel value of one pixel in head data. FIG. 3 is a diagram showing an example of a Gaussian filter and an averaging filter. FIG. 3 is a diagram showing an example of a color conversion table. FIG. 2 is a diagram schematically showing an example of a spatial filter and a method for adjusting parameters of the spatial filter. FIG. 3 is a diagram illustrating an example of a color conversion table and an example of a method for adjusting its parameters. It is a figure which shows an example of the matching process of a 1st filter image and a camera image. 3 is a flowchart showing the flow of a method for generating a trained model. It is a figure which shows an example of a test chart. It is a figure which shows another example of a test chart. It is a figure which shows yet another example of a test chart.

Before describing the embodiments of the present invention, the findings on which the present invention is based will be explained.

Digital printing machines that perform variable printing print different content on each page. Such printing machines are widely used for various applications such as address printing, bill printing, and statement printing, for example.

In variable printing, different content is printed on each page, so inspection cannot be performed simply by comparing it with a single image prepared in advance. For this reason, a method can be considered in which the inspection is performed by comparing and collating the image data of the printing source and the image data obtained by imaging the printed matter.

However, the method of comparing two images has the following problems, for example.
-Generally, printing processes data using subtractive color system (YMCK), and cameras process data using additive color system (RGB), so it is not possible to directly compare the data of the two.
- The image developed in the printing machine and the image developed in the inspection device do not necessarily match, and there are differences.
-Due to the characteristics of the paper or ink, ink bleeding or paper deformation may occur, resulting in differences between the image to be printed and the image actually printed.

Because of these problems, it has been difficult for conventional inspection devices to inspect color printed matter with high accuracy. Therefore, in the past, inspection systems were often used that performed verification by focusing on the color of important parts, such as characters, for example.

FIG. 1 is a diagram schematically showing an example of a conventional inspection system. The inspection system shown in FIG. 1 includes a printing machine having a print controller 110 and a print head 120, a processing device 210, and a camera 220 (imaging device). The print head 120 includes four types of heads corresponding to four colors: cyan (C), magenta (M), yellow (Y), and black (K). The print controller 110 converts print source data (for example, a PDF file, etc.) input from an external device into head data and outputs it to each print head 120. The head data is, for example, data of halftone images of each color of CMYK. Each print head 120 executes printing on the transported paper 10 according to the head data. The printed paper 10 is photographed by a camera 220. The camera 220 photographs the printed paper 10 and generates camera image data (hereinafter also referred to as "camera image"). The processing device 210 extracts data of a specific color (for example, the color of important parts such as characters) from each of the head data and the camera image, and compares the data to determine the quality of printing.

In such a system, the object of inspection is limited to a specific color component of a printed matter, and other color components cannot be inspected. Furthermore, it is difficult to detect with high accuracy defects in printed matter due to ink bleeding or paper deformation during printing.

Generally, paper used for printing expands and contracts by absorbing moisture and oil. When ink containing water or oil is jetted or applied onto paper, the paper may expand or contract, and the printed matter may be deformed relative to the original digital data. For example, an A4 size sheet may be deformed by about 1 mm.

FIG. 2 is a diagram schematically showing that the paper and the printed portion are deformed by printing. As shown in FIG. 2, areas with a large amount of ink have a large amount of shrinkage, and areas with a small amount of ink have a small amount of shrinkage. Since the degree of deformation and the form of deformation change depending on the printed content, it is difficult to predict deformation in advance. In particular, in variable printing in which different content is printed on each page, the amount of ink differs on each page, so the way the ink shrinks can vary irregularly on each page. As a result, it may become difficult to match the camera image with the original data. In addition, the captured image may be deformed due to the optical characteristics (mainly distortion) of the lens within the camera. Although the deformation in this case is regular, matching with the original data can be difficult as well.

In order to solve the above problems, the present inventors came up with the configuration of the embodiment of the present invention described below. Exemplary embodiments of the present invention will be described below. However, more detailed explanation than necessary may be omitted. For example, detailed explanations of well-known matters or redundant explanations of substantially the same configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. The inventors have provided the accompanying drawings and the following description to enable those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter recited in the claims. do not have. Identical or similar components are provided with the same reference numerals throughout this specification.

(Embodiment 1)
FIG. 3 is a diagram schematically showing the configuration of an inspection system according to an exemplary embodiment of the present invention. This inspection system includes a printing press 100 and an inspection device 200. FIG. 3 also shows a computer 300 connected to the printing press 100 and the inspection device 200. Computer 300 is any information processing device such as a server computer, a personal computer (PC), or a mobile terminal. Computer 300 may be included in the inspection system or may be an element external to the inspection system. The printing press 100 may also be an external element of the inspection system.

The printing press 100 shown in FIG. 1 includes a print controller 110, a print head 120, and a conveyor 130. The conveyance machine 130 conveys the roll-shaped paper 10. Print controller 110 is a device or circuit that controls the overall operation of printing press 100. The print controller 110 receives a print instruction including print source data input from the computer 300 and drives the conveyor 130 and the print head 120. Thereby, the printing machine 100 executes printing while conveying the paper 10. The print controller 110, the print head 120, and the conveyor 130 do not need to be integrated as one device, and may be configured as independent devices.

The printing machine 100 is a digital printing machine that performs printing using a known method such as an inkjet method or an electrophotographic method. The printing machine 100 prints on paper 10 based on printing source data input from the computer 300. The printing may be color printing or monochrome printing. In the following description, as an example, it is assumed that color printing is performed using four colors of ink: cyan (C), magenta (M), yellow (Y), and black (K). The printing machine 100 is not limited to CMYK, and may print using inks of other colors.

The printing press 100 in this embodiment can perform variable printing. In variable printing, different characters, figures, images, etc. are continuously printed on each page of paper 10. Although the paper 10 in this embodiment is roll paper, printing may be performed on other print media such as film.

The print controller 110 converts the input printing source data into head data and outputs it to the print head 120. The print source data may be vector format data including image and character information, such as PDF (Portable Document Format). The head data may be, for example, bitmap (raster) data indicating a halftone image expressed by a small number of gradations, about 2 to 4 gradations per color for each pixel. The print controller 110 has a RIP (Raster Image Processor) function that converts, for example, vector format printing source data into head data representing a raster image expressed as a set of points of a smaller number of gradations. RIP processing involves converting print source data such as PDF into raster data, converting RGB (red, green, blue) to CMYK (cyan, magenta, yellow, black), and converting the shading of each color into halftone dots ( This may include various types of processing such as halftone processing (halftone). Note that at least a portion of the RIP functions may be executed by a device different from the print controller 110 (for example, the computer 300). The print source data is not limited to PDF, and may be image data such as JPEG (Joint Photographic Experts Group) or PNG (Portable Network Graphics).

The print head 120 is a device that includes parts for executing printing based on input head data and a circuit for controlling those parts. Although one print head 120 is illustrated in FIG. 3, when color printing is performed, multiple types of print heads corresponding to multiple colors of ink may be provided. For example, four print heads may be provided, each corresponding to four colors of ink: cyan (C), magenta (M), yellow (Y), and black (K). If the printing machine 100 is an inkjet printer, the print heads for each color may include a plurality of inkjet heads arranged along the surface of the paper 10 in a direction perpendicular to the conveyance direction (the direction of the arrow shown in FIG. 1). Each inkjet head includes, for example, a nozzle that pressurizes or heats ink and jets it. Note that when the printing machine 100 is an electrophotographic printer, the print head 120 includes a light source such as a laser or an LED, an optical system such as a polygon mirror, parts such as a photosensitive drum, and a circuit that controls the light source and the optical system. obtain.

The inspection device 200 operates in conjunction with the printing machine 100, inspects the printed paper 10, and outputs the inspection results. Inspection device 200 is connected to printing press 100 and computer 300. Inspection device 200 includes a camera 220 and a processing device 210.

Camera 220 is placed on the conveyance path of printed paper 10. The camera 220 photographs the paper 10 after printing and generates image data (referred to as a "camera image"). Camera 220 may be configured to output a camera image for each page, for example.

The camera 220 includes an image sensor such as a CCD or CMOS, and an optical system such as a lens. The camera 220 may include a two-dimensional image sensor or a linear image sensor (line sensor) such as a contact image sensor (CIS). When the camera 220 includes a linear image sensor, the camera 220 may photograph the transported paper 10 line by line, collect data of a plurality of lines for each page, and output the data as a camera image. Alternatively, the camera 220 may output data line by line, and the processing device 210 may collect the data of a plurality of lines page by page to generate two-dimensional camera image data. The camera 220 may be provided not only on the front side of the paper 10 but also on the back side. By providing two cameras 220 on both sides of the paper 10, the printing state on both sides of the paper 10 can be inspected.

The processing device 210 includes a processor 212, a storage medium (memory) such as a RAM 214 and a ROM 216, and an input/output interface (I/F) 218. The processor 212 includes an arithmetic processing circuit such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). ROM 216 stores computer programs executed by processor 212. The program includes a group of instructions for causing the processor 212 to execute processing to be described later. The RAM 214 is a work memory for expanding the program when the processor 212 executes the program. Processor 212 is connected to computer 300, print controller 110, and camera 220 via interface 218.

The processor 212 of the processing device 210 performs a verification process based on printing source data or head data and camera image data to determine whether the printed matter is good or bad. The details of this determination process will be explained below with reference to FIG.

FIG. 4 is a schematic diagram for explaining an overview of the determination processing performed by the processing device 210. The processing device 210 acquires at least one of printing source data and head data, performs necessary conversion processing such as spatial filtering processing and color adjustment processing on the acquired printing source data or head data, and generates image data ( A first filter image) is generated. The processing device 210 also acquires a camera image from the camera 220, performs necessary conversion processing such as spatial filtering processing and color adjustment processing on the camera image, and generates a camera image for comparison (second filter image). The processing device 210 collates the first filter image and the second filter image and outputs the collation result. For example, the processing device 210 performs a process of comparing the first filter image and the second filter image for each relatively small area over the entire image, and determines whether printing is good or bad based on the degree of matching between the two. The processing device 210 outputs the determination result to an output device or a storage device, such as a display or a speaker, connected to the processing device 210, for example.

In the example shown in FIG. 4, the processing device 210 can switch based on either print source data or head data to perform the comparison. The processing device 210 may be configured to select which of print source data and head data to use for verification, for example, based on a mode setting specified by a user. In that case, the user can use an input device connected to the processing device 210 to switch between a mode using print source data and a mode using head data. Note that this selection function is not essential, and the processing device 210 may generate the first filter image for verification from only one of the print source data and head data.

In the example shown in FIG. 4, the print head 120 includes four types of print heads corresponding to four colors: cyan (C), magenta (M), yellow (Y), and black (K). This allows color printing. Print head 120 may be configured to perform monochrome printing. When monochrome printing is performed, color adjustment processing for print source data or head data and camera image data may be omitted.

According to the present embodiment, the processing device 210 performs a first filter image generated by performing a first spatial filtering process on print source data or head data, and a second spatial filtering process on a camera image obtained by photographing a printed matter. The quality of the printed material is determined based on the second filter image generated by performing the above steps. As the first spatial filtering process, a process that reflects the effects of ink bleeding and paper deformation during printing is performed. As a result, defects in printed matter can be detected with high accuracy even if ink smearing, paper deformation, etc. occur during printing. Further, in this embodiment, color adjustment processing is also performed according to the print source data or head data and the characteristics of the camera image. Thereby, inspection can be performed with high accuracy regardless of the difference in color expression between the two.

Note that the processing device 210 may omit the spatial filtering process and color adjustment process for the camera image, and directly compare the first filter image and the camera image. FIG. 5 shows an example of the system configuration in that case. By performing appropriate processing such as spatial filtering on print source data or head data (hereinafter sometimes referred to collectively as "print data"), high-precision inspection comparable to that achieved by human visual inspection can be achieved. is possible.

FIG. 6 is a diagram schematically showing the flow of printing and inspection in this embodiment. Here, an example will be described in which a filter image generated by filtering an image indicated by head data using a spatial filter is compared with a camera image. As shown in FIG. 6, the print controller 110 performs halftone processing such as dithering on the print source data to generate head data. The head data is data of halftone dots in which the shading in the printing source data is expressed as the density of dots. Such head data can be generated for each color, for example, cyan, magenta, yellow, and black. The print head 120 prints on paper by driving heads corresponding to each color based on such head data. On paper, changes occur in the image due to the effects of ink bleeding, mixing, etc. Furthermore, when a printed matter is photographed with a camera, the photographed image changes depending on the characteristics of the camera, the photographing environment, and the like. As a result, the camera image does not match the image indicated by the head data, and the two cannot be directly compared. Therefore, in the example shown in FIG. 6, the processing device 210 converts the head data into a filter image using a spatial filter that simulates the effects of ink bleeding on paper during printing, deformation of paper, etc., and converts the head data into a filter image. Match the image with the camera image. This makes it possible to compare head data and camera images, and greatly improves inspection accuracy.

Note that in the example of FIG. 6, spatial filtering is performed on the head data to generate a filter image for verification, but the filter image for verification may be generated by performing spatial filtering on the printing source data. When the print source data is data in a vector format such as PDF, the processing device 210 may perform spatial filtering after performing necessary processing such as rasterization processing and color conversion processing on the print source data. Since the printing source data is also different from the camera image, it is not possible to compare the two as is. By performing appropriate spatial filtering processing on the print source data, inspection accuracy can be improved.

In the example shown in Figure 6, the camera image is used as it is as data for comparison, but as in the example in Figure 4, the camera image is also subjected to transformation processing such as spatial filtering and color conversion, and the two are compared. You can make it easier to do so.

In this manner, in this embodiment, print data (and camera images as necessary) are processed before verification, assuming factors that may cause changes in images due to printing. The processing may be performed using a spatial filter (ie, a low-pass filter) that performs smoothing, such as a Gaussian filter. The spatial filter used is not limited to a linear filter such as a Gaussian filter, but may also be a nonlinear filter. Alternatively, a suitable filter image for matching may be generated by combining a plurality of spatial filters. The spatial filter is created in advance according to the characteristics of the printing machine 100 and the paper 10. The spatial filter may be generated as a learned model trained using machine learning, for example. The processing device 210 itself may generate the spatial filter before inspection. The processing device 210 may generate an optimal spatial filter (learned model) by learning the deformation of known print data during printing using, for example, a machine learning algorithm using a neural network. An example of a neural network that may be used is a convolutional neural network (CNN) that uses error backpropagation. The processing device 210 may generate an optimal spatial filter using, for example, a learning algorithm (pix2pix, etc.) based on a GAN (Generative Adversarial Network).

In the embodiment shown in FIGS. 4 and 5, the processing device 210 performs color conversion processing on the print data in addition to spatial filtering processing. This makes it easier to compare colors between print data and camera images. An overview of the color conversion process in this embodiment will be described below with reference to FIG.

FIG. 7 is a diagram schematically showing an overview of color conversion processing. Here, similarly to the example shown in FIG. 6, an example in which a first filter image is generated from head data will be described. In this example, the print source data is expressed in an additive RGB color space, and the head data is expressed in a subtractive CMYK color space. Since the image taken by the camera is expressed in the RGB color space, it is not possible to directly compare the head data and the camera image. Therefore, the processing device 210 generates a first filter image by performing a conversion process from the CMYK color space to the RGB color space on the head data in addition to the above-described spatial filtering process. By performing such conversion, it is possible to simulate the image taken by the camera from the head data, and it is possible to improve the accuracy of the inspection.

The color conversion is not limited to conversion from CMYK to RGB, and may be other conversions. For example, in order to bring the sensitivity to color changes closer to that of the naked eye, print data (print source data or head data) and camera images are converted to a color system such as CIE-L ^* a ^* b ^* in which each color is weighted equally. May be converted. By performing such conversion, inspection that is closer to the naked eye becomes possible. Furthermore, the color conversion of the print data may be performed using known print data and camera images, using a learned model trained by a machine learning algorithm such as CNN using error backpropagation, for example. The processing device 210 may use an algorithm such as pix2pix to generate a model that simultaneously performs the above-described spatial filtering and color conversion.

Next, the processing executed by the processing device 210 will be described in more detail with reference to FIG.

FIG. 8 is a flowchart illustrating an example of processing executed by the processor 212 of the processing device 210. In this example, after starting printing, the processing device 210 repeats the processes from steps S101 to S106 for each page of printed material until receiving an instruction to end printing (step S107) from the print controller 110. Note that the instruction to end printing may be input from a device different from the print controller 110.

In step S101, the processing device 210 acquires print source data or head data. The processing device 210 may obtain the print source data from the computer 300 as shown in FIG. 3 or from the print controller 110. The processing device 210 acquires head data from the print controller 110. The processing device 210 may acquire both print source data and head data, or may acquire only one of them. The processing device 210 may be configured to acquire either print source data or head data according to a set mode.

In step S102, the processing device 210 performs a first spatial filtering process on the print source data or head data acquired in step S101 to generate a first filter image. The first spatial filtering process may include, for example, a process using a smoothing filter such as a weighted averaging filter (eg, a Gaussian filter). The first spatial filtering process may include size compression processing using an averaging filter. The first spatial filtering process can simulate the effects of ink bleeding and paper deformation, making it easier to compare with camera images. When processing print source data in vector format, the processing device 210 may perform rasterization processing and convert it into a raster (bitmap) format image before performing the first spatial filtering processing. In addition to the first spatial filtering process, the processing device 210 may perform a first color conversion process to adjust the color of each pixel of the first filter image. When the first filter image is expressed in a four-color color space of CMYK, the processing device 210 is configured to perform color conversion on the first filter image from CMYK to RGB, which is the color space of the camera image. can be done. Color conversion may be performed based on a color conversion table created in advance. Alternatively, the processing device 210 converts the print source data or head data into a first filter image for comparison using a learned model that is trained in advance using a learning algorithm that uses error backpropagation. Good too.

In step S103, the processing device 210 acquires a camera image from the camera 220. The camera image is, for example, RGB image data.

In step S104, the processing device 210 performs a second spatial filtering process on the camera image acquired in step S103 to generate a second filtered image. The second spatial filtering process may include processing using a smoothing filter, such as a weighted averaging filter (eg, a Gaussian filter). The second spatial filtering process may include size compression processing using an averaging filter. By performing the second spatial filtering process, it is possible to generate, for example, a second filter image that has the same number of pixels and degree of blur as the first filter image. This facilitates comparison with the first filter image. In addition to the second spatial filtering process, the processing device 210 may perform a second color conversion process to adjust the color of each pixel of the camera image. In addition to the second spatial filtering process and the second color conversion process, processing such as rotation correction may be performed to facilitate comparison with the first filter image. Alternatively, the processing device 210 may convert the camera image into a second filter image for comparison using a learned model that has been trained in advance using a learning algorithm that uses backpropagation or the like.

Note that the processing in steps S103 and S104 may be performed before steps S101 and S102, or may be performed in parallel with the processing in steps S101 and S102.

In step S105, the processing device 210 compares the first filter image and the second filter image to determine whether the printing is good or bad. The processing device 210 determines the quality of printing based on the difference between the first filter image and the second filter image. For example, it is possible to compare the first filter image and the second filter image for each corresponding small area and determine that a printing error has occurred if a location where the pixel values are significantly different is detected.

In step S106, the processing device 210 outputs the determination result. For example, the processing device 210 outputs, as a determination result, information indicating whether or not there is a defect in the printing and, if there is a defect, where in the image the defect is located. The determination result may be output to an output device such as a display or a storage device connected to the processing device 210, for example. By looking at this output, the user can know that a defect has occurred in printing. The processing device 210 may create a list of test results in which the determination results are recorded for each page, and record the list in the storage device. This allows a history of printing and inspection to be kept.

In step S107, the processing device 210 determines whether there is an instruction to end printing. If there is no instruction to end printing, the process returns to step S101 and the next page is inspected. The instruction to end printing may be input from the print controller 110, for example.

Through the above operations, the processing device 210 can compare the print source image and the camera image for each page of printed matter, and determine whether there is any abnormality in printing. In this embodiment, in steps S102 and S104, spatial filtering processing and color conversion processing are performed on the print source data or head data and the camera image. Through these processes, it is possible to easily match the print source data or head data with the camera image, and improve the accuracy of inspection.

In the example of FIG. 8, the camera image is converted into the second filter image in step S104, but this process may be omitted.

FIG. 9 is a flowchart illustrating another example of the process executed by the processor 212 of the processing device 210. In the flowchart shown in FIG. 9, steps S104 and S105 in the flowchart shown in FIG. 8 are replaced with step S114. Processing other than step S114 is similar to the example of FIG. 8. In the example of FIG. 9, the processing device 210 does not perform spatial filtering processing or color conversion processing on the camera image, and uses the camera image as it is as data for matching. In step S114, the processing device 210 compares the first filter image and the camera image to determine whether the printing is good or bad. Even in this case, by appropriately creating the model (spatial filter and color conversion table) used to generate the first filter image from print data, it is possible to judge the quality of printing with high accuracy. .

FIG. 10 is a flowchart showing still another example of the processing executed by the processor 212 of the processing device 210. In this example, the processing device 210 selects and processes either print source data or head data. The flowchart shown in FIG. 10 is the same as the flowchart shown in FIG. 8 except that the operation of step S101 in FIG. 8 is replaced with steps S120, S121, and S122.

In step S120, the processing device 210 determines whether to use print source data or head data. The processing device 210 determines whether to use print source data or head data, for example, according to a mode set by a user. If it is determined that the print source data is to be used, the process proceeds to step S121, and the processing device 210 acquires the print source data. If it is determined that head data is to be acquired, the process proceeds to step S122, and the processing device 210 acquires head data. After step S121 or S122, the process advances to step S102. In step S102, the processing device 210 performs spatial filtering and color conversion processing on one of the print source data and head data acquired in step S121 or S122. The operations in steps S102 to S107 are the same as the operations in the corresponding steps in FIG.

Note that the operation of step S114 in FIG. 9 may be performed instead of the operations of steps S104 and S105 in the example of FIG.

Here, a specific example of the data conversion process in step S102 will be described with reference to FIG. 11. FIG. 11 shows the flow of processing for converting head data into a first filter image. Here, as an example, the first head data is An example of processing when generating a filter image will be described. In the example of FIG. 11, the pixel value in the image indicated by the head data is expressed by 2 bits for each color (8 bits in total).

FIG. 12A shows an example of the pixel value of one pixel in the head data. As shown in FIG. 12A, for example, when the pixel value is 210 (11010010 in binary), the amount of cyan ink is "large", the amount of magenta is "small", the amount of yellow is "no ink", and the amount of black ink is "large". is "medium". In this way, the amount of ink for each color of CMYK is expressed by 2 bits, and the larger the value of the 2 bits, the larger the amount of ink. In this embodiment, the pixel value in the head data and the print density have a nonlinear relationship, and the print density is not proportional to the pixel value.

In the example of FIG. 11, the processing device 210 refers to a gray scale conversion table recorded in memory in advance for each color of CMYK, and converts head data having a pixel value of 2 bits for each color into 8 bits for each color (32 bits in total). The image data is converted into image data (b) having a pixel value of . There is a linear relationship between the pixel value and the density in this image data, and the density is proportional to the pixel value. The processing device 210 performs filtering processing (convolution operation) on this image data using a spatial filter such as a Gaussian filter and an averaging filter (for example, a filter that averages 16 pixels). As a result, the processing device 210 smoothes the CMYK color images and generates image data (c) with the number of pixels reduced to 1/4 (300 dpi).

FIG. 12B shows an example of a Gaussian filter and an averaging filter. The numerical values of each element of these filters (kernels) are determined so as to appropriately reflect the effects of ink bleeding, paper deformation, etc. due to printing.

The processing device 210 performs color conversion on the image data (c) from CMYK to RGB based on a color conversion table recorded in advance in a storage medium such as a memory, and generates a first filter image (d). do.

FIG. 12C shows an example of a color conversion table. This color conversion table defines the correspondence between a combination of four CMYK values and a combination of three RGB values. Each numerical value in the color conversion table is set to an appropriate value depending on the characteristics of the ink, paper, camera, etc.

The first filter image in the example of FIG. 11 is a 300 dpi image in which the pixel values of each color of RGB are expressed with 8 bits (24 bits in total). The processing device 210 checks the quality of the print by comparing the first filter image with the camera image or a second filter image converted from the camera image.

Note that the resolution of each image, the number of bits of each pixel, and the numerical values in each filter and each table in the example of FIG. 11 are merely examples, and these numerical values are set to appropriate values depending on the system.

In the example of FIG. 11, the first filter image is generated from the head data, but the first filter image may be generated based on printing source data such as PDF. In that case, the processing device 210 converts the print source data into bitmap image data expressed in a predetermined color space such as RGB, and then converts the data into a bitmap image data expressed in a predetermined color space such as RGB, Filtering (and color conversion if necessary) may be performed to generate the first filter image.

Next, an example of the spatial filter used in this embodiment will be explained in more detail.

ここで、（ｘ，ｙ）は、画像の中心を原点とする座標であり、ｘ軸は画像の横方向に対応し、ｙ軸は画像の縦方向に対応する。σは標準偏差である。 FIG. 13 is a diagram schematically showing an example of a spatial filter applied to print data and a method for adjusting parameters of the spatial filter. The spatial filter in this example is a Gaussian filter based on the following two-dimensional Gaussian distribution function.

Here, (x, y) are coordinates with the center of the image as the origin, the x axis corresponds to the horizontal direction of the image, and the y axis corresponds to the vertical direction of the image. σ is the standard deviation.

The kernel size of the Gaussian filter illustrated in FIG. 13 is 5×5. The kernel size may be other sizes (eg, 3x3, 7x7, etc.). Parameters such as the kernel size and standard deviation σ are set to appropriate values depending on the characteristics of the printing press, camera, etc. used in the inspection. Although a Gaussian filter is used in this example, when other spatial filters are used, the filter parameters (values of each element) can be adjusted depending on the characteristics of the printing press, camera, etc.

Here, the pixel values at coordinates (i, j) in the print data, filter image, and camera image are expressed as A[i,j], B[i,j], and C[i,j], respectively. The pixel value B[i,j] of the filter image is approximated by the following formula using the above two-dimensional Gaussian distribution function.

The deviation d between the camera image and the filter image is expressed by the following formula.

Each parameter of the spatial filter is determined in advance to an appropriate value so as to minimize this deviation d. Each parameter of the spatial filter may be determined using machine learning. For example, by performing supervised learning using a predetermined machine learning algorithm and a large number of training datasets corresponding to pairs of print data A[i,j] and camera image C[i,j], , each parameter of the spatial filter may be determined. As a result of this learning, the spatial filter actually used in the inspection may be a filter different from the Gaussian filter.

Next, an example of the color conversion process and the method of adjusting the color conversion table in this embodiment will be described in more detail.

FIG. 14 is a diagram showing an example of a color conversion table and an example of a method for adjusting its parameters. In this example, the processing device 210 converts CMYK print data into an RGB filter image based on a color conversion table recorded in a storage medium such as a memory. This color conversion table is data that defines the relationship between the combination of four color values of CMYK and the combination of three color values of RGB. Instead of a color conversion table, a function or model that implements a similar conversion may be used. Each parameter in the color conversion table is determined to minimize the deviation between the camera image and the filter image. The processing device 210 executes supervised learning using a predetermined machine learning algorithm and a large number of learning data sets corresponding to pairs of print data A[i,j] and camera images C[i,j]. By doing so, the parameters of the color conversion table may be adjusted.

In this embodiment, the spatial filtering process and the color conversion process are performed separately, but a conversion process that includes these processes may also be performed. Such conversion processing may be, for example, conversion using a learned model trained using a machine learning algorithm such as a neural network. The processing device 210 may generate a trained model for generating a pseudo camera image from print data using an algorithm such as GAN, for example.

Next, an example of matching processing between the first filter image converted from the print data and the camera image will be described.

FIG. 15 is a diagram illustrating an example of a matching process between the first filter image and the camera image. In this example, the processing device 210 divides the first filter image and the camera image into a plurality of regions, and compares the two regions for each region. In the example of FIG. 15, each image is divided into 12 regions. Overlapping regions are provided between the regions. The processing device 210 compares the region to be cut out from the first filter image multiple times while moving it one pixel at a time in each direction, for example, vertically and horizontally, determines the cutout region with the highest degree of matching with the camera image, and combines that region with the camera image. You may also compare the corresponding areas in . An offset may be provided in advance to the coordinates in order to compensate for the effects of predictable deformations such as deformations of the camera image due to lens aberrations in the camera. The processing device 210 can calculate the difference between the pixel value of the first filter image and the corresponding pixel value of the camera image for each pixel, and detect printing defects based on the difference in pixel values. For example, the processing device 210 calculates an index value indicating the difference between images, such as SSD (Sum of Squared Difference) or SAD (Sum of Absolute Difference), for each of the above regions, and when the value exceeds a threshold value, , it is possible to detect that there is a defect in the printing in that area. Note that when comparing the first filter image with the second filter image obtained by performing spatial filtering or color conversion processing on the camera image, the quality of printing can be determined using the same method.

As described above, the inspection system in this embodiment is combined with a printing press 100 that includes a print controller 110 that converts input printing source data into head data, and a print head 120 that prints based on the head data. used. The inspection system includes a camera 220 that photographs a printed matter printed by the printing press 100 to generate a camera image, and a processing device 210. The processing device 210 generates a first filter image by performing a first spatial filtering process on the print source data or head data, and matches the first filter image with the camera image, or compares the first filter image with the camera image. The quality of the printed material is determined by comparing it with the second filter image generated by performing the second spatial filtering process, and the determination result is output.

With such a configuration, it is possible to compensate for the effects of ink smearing, paper deformation, or camera lens aberration during printing, thereby improving the accuracy of image inspection of printed matter.

Note that in this embodiment, spatial filtering processing and color conversion processing are performed on the print source data or head data, but the color conversion processing may be omitted depending on the printed matter. For example, when monochrome printing is inspected, the color conversion process may be omitted. This point also applies to color conversion processing for camera images. Further, although the printing apparatus 100 in this embodiment is an inkjet printer, the technology of this embodiment may be applied to an electrophotographic printer such as a laser printer, or a printing machine that performs plate printing.

(Embodiment 2)
Next, an example of a method for generating, by machine learning, a trained model used to perform transformation processing such as the above-mentioned spatial filtering and color transformation will be described.

FIG. 16 is a flowchart showing the flow of the trained model generation method in this embodiment. This method includes the steps of generating a training dataset (step S200) and generating a trained model based on the training dataset (step S210). Step S200 includes steps S201 to S205. Step S210 includes steps S211 to S213. Step S200 and step S210 may be executed by the same device or by different devices. In the following description, it is assumed, as an example, that the computer 300 shown in FIG. 3 executes step S200, and the processing device 210 executes step S210. Note that the respective processes of steps S200 and S210 may be executed not only by the computer 300 and the processing device 210 but also by other devices.

In step S201, the computer 300 acquires learning print data from the storage device. The learning print data may be, for example, data showing an image similar to a printed matter actually inspected, or data showing an image of a test chart including a plurality of color patterns specially created for learning. The learning print data is created in advance and recorded in a storage device within the computer 300 or an external storage device. The learning print data corresponds to the print source data shown in FIG. Note that the learning print data may be data in the same format as the head data.

In step S202, the computer 300 instructs the printing press 100 to execute printing based on the learning print data. In response to this, the printing machine 100 executes printing on the paper 10.

In step S203, the computer 300 instructs the camera 220 to take a photograph of the printed matter based on the learning print data. In response to this, the camera 220 generates image data of the printed matter (hereinafter referred to as "learning camera image").

In step S204, the computer 300 associates the learning camera image and the learning print data and records them in the storage device.

In step S205, the computer 300 determines whether an instruction to end the generation of learning data has been issued. The termination instruction may be input to computer 300 based on a user's operation, for example. Alternatively, the termination instruction may be issued when a preset number of printing and photographing operations are completed. The computer 300 repeats the operations from steps S201 to S205 for different learning print data until a termination instruction is issued.

Through the above operations, a learning data set including one or more pairs of learning print data and learning camera images is generated and recorded in the storage device. This learning data set is used to train a machine learning model for converting print data into a pseudo camera image (hereinafter sometimes referred to as a "verification print image"). In order to enhance the effectiveness of learning, it is preferable to prepare as many pairs of learning print data and learning camera images as possible.

Next, details of step S210 for generating a trained model using the training data set will be described.

In step S211, the processing device 210 acquires a learning data set from the storage device. If the learning data set includes a plurality of sets of learning print data and learning camera images, data of all sets are acquired.

In step S212, the processing device 210 uses a predetermined machine learning algorithm to perform machine learning based on the learning data set, thereby creating a model (a learned model) for converting the print data into a print image for verification. ) is generated. As the machine learning algorithm, various algorithms used for converting an input image into another image using deep learning, for example, can be used. For example, an algorithm using a convolutional neural network (CNN), such as GAN or NST (Neural Style Transfer), can be used. The generated trained model may be a model that executes the spatial filtering process and color conversion process described in the first embodiment, for example. Generation of the learned model includes adjusting a plurality of parameters in the model so as to reduce the deviation between the learning print data and the corresponding learning camera image.

In step S213, the processing device 210 records the generated trained model in a storage device (eg, ROM 216).

Through the above operations, it is possible to generate a learned model for converting print data into a print image for verification. When performing an inspection based on this learned model, the processing device 210 executes an image conversion program implementing the learned model to convert print data to be inspected into a print image for comparison. The processing device 210 compares and verifies the verification print image with a camera image or an image obtained by processing rotation correction, color correction, etc. on the camera image (collectively referred to as "verification camera image"), and processes the printed matter. Determine the quality of the product.

Next, an example of a test chart used as learning print data used for model generation will be described. The test chart may include, for example, a pattern in which a plurality of regions of different colors are arranged one-dimensionally or two-dimensionally. As used herein, "color" means a combination of hue, lightness, and saturation.

17A to 17C show examples of images of test charts. FIG. 17A shows an example of a test chart in which a plurality of color blocks of different colors are two-dimensionally arranged. This test chart consists of areas of many square color blocks arranged in a direction corresponding to the conveyance direction of the printed material (vertical direction in the figure) and a direction corresponding to the direction perpendicular to the conveyance direction (horizontal direction in the figure). including. Each color block is small, for example smaller than 10 mm x 10 mm. The number of color blocks is, for example, 1000 or more. The colors of these color blocks may be selected to cover most of the colors printable by printing press 100, for example. It is not necessary that all the colors of the color blocks are different, and blocks of the same color may be included.

FIG. 17B shows an example of a test chart in which a plurality of color bars of different colors are arranged one-dimensionally. This test chart includes a plurality of color bars of different colors extending in a direction (horizontal direction in the figure) corresponding to the direction perpendicular to the conveyance direction of the printed matter. The plurality of color bars are arranged along a direction corresponding to the transport direction (vertical direction in the figure). The width (dimension in the vertical direction) of each color bar is small, for example smaller than 10 mm. The number of color bars is, for example, 30 or more. The colors of these color bars may be selected to cover most of the colors printable by printing press 100, for example. All colors of the color bar do not need to be different, and color bars of the same color may be included.

If printing machine 100 is an inkjet printer, print head 120 may include multiple heads for each color of CMYK, for example. Multiple heads for each color may be arranged along a direction perpendicular to the transport direction. Since the ink ejection condition may differ depending on the head, the density and spread of the ink may differ depending on the position of the head even for the same color. Therefore, in the example of FIG. 17B, a test chart including a plurality of color bars extending in a direction corresponding to the direction in which the heads are arranged is used. By using such a test chart, it is possible to effectively learn how each color changes depending on the position in the arrangement direction of the heads.

The test chart illustrated in FIGS. 17A and 17B includes a plurality of two-dimensional codes 70 including information indicating the type of the test chart. The two-dimensional code 70 is a code, such as a QR (Quick Response) code, in which information is expressed as two-dimensional shading information according to a predetermined rule. The two-dimensional code 70 can be used to identify the type of test chart and the position of each color area in a model learning device. In the example of FIGS. 17A and 17B, a plurality of two-dimensional codes 70 are arranged at predetermined positions around a color block or color bar. By arranging the two-dimensional code 70 in this manner, positioning during verification can be facilitated. At least a portion of the two-dimensional code 70 may be arranged along a direction (lateral direction) corresponding to a direction perpendicular to the conveyance direction of the printed material. Such a two-dimensional code 70 functions as a mark during verification and facilitates identification of each color area. The two-dimensional code 70 may include information for specifying the position of each color region included in the color chart. For example, the two-dimensional code 70 may include information indicating the correspondence between position coordinates in the coordinate system set in the color chart and colors at the position coordinates. As in the example of FIGS. 17A and 17B, when a plurality of two-dimensional codes 70 are included in the color chart, each two-dimensional code 70 is associated with the two-dimensional code 70 in addition to its own position coordinate information. It may also include information on the position coordinates of regions of multiple colors. By referring to such information, the processing device 210 can specify the position of each color area using the two-dimensional code 70 as a landmark. This facilitates positioning and improves learning accuracy.

Note that instead of the two-dimensional code, a one-dimensional code (barcode) may be provided on the color chart. A similar function can be achieved even when barcodes are used. Further, instead of the two-dimensional code and the one-dimensional code, for example, characteristic marks, characters, or symbols may be used to realize the same function.

FIG. 17C is a diagram showing still another example of the test chart. This test chart includes a plurality of color bars of different colors. The brightness of each color bar changes continuously along a direction (horizontal direction) corresponding to a direction perpendicular to the transport direction. By using such a test chart, it is possible to efficiently learn a plurality of colors having different lightness.

The number of test charts illustrated in FIGS. 17A to 17C is not limited to one type, and multiple types of test charts may be used. In that case, for example, the operations from steps S201 to S205 shown in FIG. 16 are repeated for each test chart. A model that can perform image conversion with higher accuracy may be generated by combining learning using a test chart and learning using test data close to a normal inspection target.

As described above, the present disclosure includes the systems, devices, methods, and programs described in the following items.

[Item a1]
An inspection system used in combination with a printing machine including a print controller that converts input printing source data into head data, and a print head that prints based on the head data, the inspection system comprising:
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
Performing a first spatial filtering process on the print source data or the head data to generate a first filter image, and comparing the first filter image and the camera image, or comparing the first filter image and the camera image. a processing device that determines the quality of the printed matter by comparing it with a second filter image generated by performing a second spatial filtering process, and outputs a determination result;
An inspection system equipped with

[Item a2]
The inspection system according to item a1, wherein the processing device performs the first spatial filtering process on the print source data to generate the first filter image.

[Item a3]
The inspection system according to item a1, wherein the processing device performs the first spatial filtering process on the head data to generate the first filter image.

[Item a4]
The inspection system according to any one of items a1 to a3, wherein the processing device determines the quality of the printed matter by comparing the first filter image with the camera image.

[Item a5]
The inspection system according to any one of items a1 to a3, wherein the processing device determines the quality of the printed matter by comparing the first filter image and the second filter image.

[Item a6]
The processing device generates the first filter image by performing the first spatial filtering process and the first color conversion process on the printing source data or the head data, according to any one of items a1 to a5. inspection system.

[Item a7]
The processing device generates the second filter image by performing the second spatial filtering process and the second color conversion process on the camera image,
determining the quality of the printed matter by comparing the first filter image and the second filter image;
The inspection system according to any one of items a1 to a3, a5, and a6.

[Item a8]
The processing device has a mode in which the first spatial filtering process is performed on the print source data to generate the first filter image, and a mode in which the first spatial filtering process is performed on the head data to generate the first filter image. The inspection system according to any one of items a1 to a7, which is capable of switching between modes.

[Item a9]
a printing machine including a print controller that converts input printing source data into head data; and a print head that prints based on the head data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
a processing device;
A processing device used in an inspection system comprising:
performing a first spatial filtering process on the print source data or the head data to generate a first filter image;
By comparing the first filter image with the camera image, or comparing the first filter image with a second filter image generated by performing a second spatial filtering process on the camera image, the printed matter is Determine the quality of the product and output the determination result.
Processing equipment.

[Item a10]
a printing machine including a print controller that converts input printing source data into head data; and a print head that prints based on the head data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
a processing device;
An inspection method executed by the processing device in an inspection system comprising:
performing a first spatial filtering process on the print source data or the head data to generate a first filter image;
By comparing the first filter image with the camera image, or comparing the first filter image with a second filter image generated by performing a second spatial filtering process on the camera image, the printed matter is determining the quality of the product and outputting the determination result;
Inspection methods including.

[Item a11]
a printing machine including a print controller that converts input printing source data into head data; and a print head that prints based on the head data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
a processing device;
A program used in an inspection system comprising:
performing a first spatial filtering process on the print source data or the head data to generate a first filter image;
By comparing the first filter image with the camera image, or comparing the first filter image with a second filter image generated by performing a second spatial filtering process on the camera image, the printed matter is Determine the quality of the product and output the determination result.
A program that does something.

[Item b1]
a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
A method for generating the learned model for an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. Obtaining a training dataset containing
Generating the learned model by machine learning based on the learning data set;
method including.

[Item b2]
The method according to item b1, wherein the test chart includes a pattern in which a plurality of regions of different colors are arranged one-dimensionally or two-dimensionally.

[Item b3]
The test chart includes a plurality of color bars of different colors extending in a direction corresponding to a direction perpendicular to the conveyance direction of the printed matter, and the plurality of color bars are arranged in a direction corresponding to the conveyance direction. The method described in item b1 or b2.

[Item b4]
The method according to any one of items b1 to b3, wherein the test chart includes one or more barcodes or two-dimensional codes containing information indicating the type of the test chart.

[Item b5]
The test chart includes a plurality of two-dimensional codes including information indicating the type of the test chart, and at least a part of the plurality of two-dimensional codes are arranged in a direction corresponding to a direction perpendicular to the conveyance direction of the printed material. The method according to any one of items b1 to b4, wherein

[Item b6]
The learned model is a model that performs spatial filtering processing and color conversion processing on the print data,
Generation of the learned model includes adjusting a plurality of parameters in the model so as to reduce a deviation between the learning print data and the corresponding learning camera image.
The method according to any one of items b1 to b5.

[Item b7]
The printing machine includes:
a print controller that converts input print source data into head data;
a print head that performs printing based on the head data;
Equipped with
the print data is the print source data or the head data;
The method according to any one of items b1 to b6.

[Item b8]
a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
A program for generating the learned model in an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. Obtaining a training dataset containing
Learning the trained model by machine learning based on the learning dataset;
A program to run.

[Item b9]
a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
An apparatus for generating the learned model, which is used in combination with an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. a memory for storing a training dataset including;
a processor that generates the learned model by machine learning based on the learning data set;
A device comprising:

The technology of the present disclosure can be used to inspect whether the print result of printed matter is appropriate. For example, it can be suitably used for inspecting printed matter printed by a digital printing machine that performs color variable printing.

10 Paper 70 Test chart 100 Printing machine 110 Print controller 120 Print head 130 Conveyor
200 Inspection device 210 Processing device 212 Processor 214 RAM
216 ROM
220 camera 300 computer

Claims (9)

a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
A method for generating the learned model for an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. Obtaining a training dataset containing
Generating the learned model by machine learning based on the learning data set;
method including. The method according to claim 1, wherein the test chart includes a pattern in which a plurality of regions of different colors are arranged one-dimensionally or two-dimensionally. The test chart includes a plurality of color bars of different colors extending in a direction corresponding to a direction perpendicular to the conveyance direction of the printed matter, and the plurality of color bars are arranged in a direction corresponding to the conveyance direction. The method according to claim 1 or 2. 4. A method according to any of claims 1 to 3, wherein the test chart includes one or more barcodes or two-dimensional codes containing information indicating the type of the test chart. The test chart includes a plurality of two-dimensional codes including information indicating the type of the test chart, and at least a part of the plurality of two-dimensional codes are arranged in a direction corresponding to a direction perpendicular to the conveyance direction of the printed matter. 5. The method according to any one of claims 1 to 4, wherein: The learned model is a model that performs spatial filtering processing and color conversion processing on the print data,
Generation of the learned model includes adjusting a plurality of parameters in the model so as to reduce a deviation between the learning print data and the corresponding learning camera image.
A method according to any one of claims 1 to 5. The printing machine includes:
a print controller that converts input print source data into head data;
a print head that performs printing based on the head data;
Equipped with
the print data is the print source data or the head data;
A method according to any one of claims 1 to 6. a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
A program for generating the learned model in an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. Obtaining a training dataset containing
Learning the trained model by machine learning based on the learning dataset;
A program to run. a printing machine that executes printing based on print data;
a camera that generates a camera image by photographing the printed matter printed by the printing machine;
By converting the print data into a verification print image using a learned model trained in advance by machine learning, and performing verification between the verification print image and a verification camera image based on the camera image, a processing device that determines the quality of printed matter;
An apparatus for generating the learned model, which is used in combination with an inspection system comprising:
learning print data showing an image of a test chart including a plurality of color patterns; and a learning camera image generated by the camera photographing a printed matter printed by the printing machine based on the learning print data. a memory for storing a training dataset including;
a processor that generates the learned model by machine learning based on the learning data set;
A device comprising:

Images