JP7272857B2 - Inspection method, inspection device, program and printing device - Google Patents

Description

The present invention relates to an inspection method, an inspection apparatus, a program, and a printing apparatus.

2. Description of the Related Art A defect inspection apparatus is widely used for inspecting defects in an inspection target by comparing inspection data obtained by imaging the inspection target using an imaging device such as a scanner and reference data as reference data. there is

Patent Literature 1 describes a character inspection device that inspects the quality of characters printed on an article. The device described in the same document compares a test character image signal based on an image of a character to be inspected with a reference character image signal, and determines whether or not the test character image signal is non-defective. In this apparatus, when replacement of reference characters is specified, replacement with characters determined as non-defective is performed.

Japanese Patent Application Laid-Open No. 2002-200000 discloses a visual inspection apparatus that inspects an inspection target for defects based on imaging data obtained by imaging the inspection target. The device described in the document captures an image of an inspection target, detects defect candidates from the captured image data, and calculates feature amounts of the defect candidates. In addition, the apparatus classifies defect candidate types using a classifier, and judges whether the inspection target is good or bad. The classifier installed in the device is generated by performing machine learning using learning data.

JP-A-5-293951 Japanese Patent No. 6403261

As an application of the conventional defect inspection apparatus, there is a print defect inspection apparatus that inspects print defects such as streaks and ink defects on printed matter generated using a printing apparatus. The printed matter defect inspection apparatus acquires in advance reference data obtained by imaging the printed matter printed under the same printing conditions as those of the printed matter to be inspected, and acquires inspection data obtained by imaging the printed matter. The quality of the printed matter is determined by comparing it with the reference data.

However, it is difficult to compare and inspect the printed matter in the case of variable printing in which the pattern changes, the first printed matter immediately after the printing conditions are changed, and the like. In the print defect inspection that compares with the reference data described above, it is relatively easy to compare image data of prints to which the same printing conditions are applied, because the basic image characteristics of both are the same. It is possible to carry out processing at

On the other hand, when the printing conditions of the printed material to be inspected, such as the type of paper, the type of mesh, and the color profile, are different from the printing conditions of the printing on which the reference data is based, and when the digital data input to the printing device is used as the reference data , it is difficult to compare the two because the characteristics of the inspection data differ from those of the reference data.

If the type of paper is different, the luminance value of the entire image will be different. The printed matter is inspected for defects by applying the reference data generated from the printed image using gloss paper, and the same pattern as the printed matter using gloss paper is printed using high-quality paper, and the printed matter using high-quality paper. When performing defect inspection, the following problems occur.

Since there is a large difference between the reference data using gloss paper and the inspection data of printed materials using high-quality paper, when comparing the two, the inspection performance is reduced to allow for a certain range of differences between the two. There is a need. On the other hand, when inspecting defects in printed matter without degrading inspection performance, it is necessary to generate reference data using high-quality paper.

When the digital data input to the printing device is used as the reference data, between the 4 channels to which CMYK, which is the digital data to be input to the printing device, are applied, and the 3 channels, to which the RGB which is inspection data is applied, The data format, etc. are different. In this case, preprocessing is required to match the formats and the like of both. CMYK represents cyan, magenta, yellow and black. RGB stands for red, green and blue.

In the prior art, a method is applied in which a table is created for matching the color characteristics of the two, and the table is used to match the color characteristics of the two. On the other hand, there are wide-ranging differences in printing conditions, such as differences in printing apparatuses and differences in printing paper from country to country. There is a problem that creating a table for each printing condition is time consuming and complicated.

The inventions described in Patent Literature 1 and Patent Literature 2 improve the inspection performance by applying inspection results obtained using an inspection apparatus that already has a certain standard. Further, the invention described in Patent Document 1 aims to improve the inspection performance by replacing the reference character with a character determined to be non-defective. However, the invention described in the document does not pay attention to the technical problem that it is difficult to perform inspection by comparing the reference data and the inspection data, and it is difficult to determine whether the product is non-defective. is.

The invention described in Patent Document 2 is a classifier that classifies feature amounts of defect candidates, and includes a classifier generated by applying machine learning. On the other hand, it is difficult for the classifier in the invention described in Patent Document 2 to perform the inspection by comparing the reference data and the inspection data itself, and it is difficult to solve the technical problem that it is difficult to determine whether the product is non-defective.

SUMMARY OF THE INVENTION The present invention has been made in view of such circumstances, and aims to provide an inspection method, an inspection apparatus, a program, and a printing apparatus capable of judging the quality of printed matter by applying reference data according to the conditions of the printed matter. aim.

In order to achieve the above object, the following aspects of the invention are provided.

An inspection method according to a first aspect includes a reference data acquisition step of acquiring reference data applied to printing of a printed matter, and comparison reference data of generating comparison reference data by converting the reference data according to conditions of the printed matter. a generation step; a comparison image data generation step of generating comparison image data using image data obtained by imaging a printed matter using an imaging device; and an inspection step of judging quality, wherein the comparison reference data generation step is an inspection method of converting reference data into comparison reference data using a first learning model generated by applying deep learning. .

According to the first aspect, the reference data is converted into the reference data for comparison using the first learning model generated by applying deep learning, and the imaging data for comparison and the imaging data for comparison based on the imaging data obtained by imaging the inspection object The quality of the object to be inspected is judged by comparing it with the reference data. This makes it possible to determine the quality of the inspection target based on the comparison reference data corresponding to the printed matter to be inspected.

The reference data may apply print input data that is input to the printing device during printing. The print input data can apply image data to which a prescribed format has been applied. A captured image of a print can be applied to the reference data.

A second aspect is the inspection method of the first aspect, wherein in the comparison reference data generating step, the similarity of the comparison reference data to the comparison imaging data is higher than the similarity of the reference data to the comparison imaging data. A first learning model generated by performing deep learning may be used.

According to the second aspect, it is possible to determine the quality of a printed matter to which comparison reference data having a higher similarity to the comparison imaging data than the reference data is applied.

A third aspect is the inspection method according to the first aspect or the second aspect, wherein the first learning model is configured to be generated by performing deep learning using a plurality of comparative imaging data with different conditions of printed matter as learning data. good too.

According to the third aspect, it is possible to generate comparison reference data according to various printed material conditions.

Print conditions may include printing conditions such as print media type, mesh type and color profile. The print conditions may include print information such as print element characteristic information and print element correction information.

A fourth aspect is the inspection method according to any one of the first to third aspects, wherein the first learning model is obtained by imaging printed matter printed by applying one or more printing conditions. It may be configured to be generated by implementing deep learning.

According to the fourth aspect, it is possible to generate the first learning model reflecting the printing conditions.

An example of one or more print conditions is an example of using any print medium as a print condition. Any print medium can be any net or any color profile.

A fifth aspect is the inspection method according to any one of the first to fourth aspects, wherein the first learning model comprises one or more convolution layers, and receives learning reference data in the same format as the reference data. , may be generated by performing deep learning for learning learning data outputting learning imaging data in the same format as the comparison imaging data.

According to the fifth aspect, the first learning model can be generated by performing deep learning using the learning reference data in the same format as the reference data and the learning imaging data in the same format as the comparison imaging data.

According to a sixth aspect, in the inspection method of the fifth aspect, the first learning model may be generated by performing deep learning using print information as learning data.

According to the sixth aspect, it is possible to generate the first learning model reflecting the print information.

A seventh aspect is the inspection method according to the fifth aspect or the sixth aspect, wherein the learning imaging data is generated using a printed matter on which a first correction process for correcting variations in the printing device that generates the printed matter has been performed. may be

According to the seventh aspect, it is possible to generate learning imaging data in which variations in printing apparatuses are corrected. Printing device variability may include variability within a printing device and/or between printing devices in a plurality of printing devices.

According to an eighth aspect, in the inspection method of the seventh aspect, the first correction process may be configured such that the correction of the ejection characteristics is performed as the correction of the variation of the printing device.

According to the eighth aspect, it is possible to generate learning imaging data in which ejection characteristics are corrected.

A ninth aspect may be configured such that, in the inspection method of the seventh aspect or the eighth aspect, the first correction process corrects variations in the transport speed of the printing medium in the printing device as variations in the printing device.

According to the ninth aspect, it is possible to generate the learning imaging data that reflects the correction of the conveyance speed variation of the print medium.

A tenth aspect is the inspection method according to any one of the fifth aspect to the ninth aspect, wherein the learning imaging data is an image of the printed matter on which a second correction process for correcting variations in an imaging device that images the printed matter has been performed. The configuration may be generated using data.

According to the tenth aspect, it is possible to generate learning imaging data in which variations in imaging devices are corrected.

An eleventh aspect may be configured such that, in the inspection method of the tenth aspect, the second correction process corrects at least one of the MTF and CTF of the imaging device as variation of the imaging device.

According to the eleventh aspect, learning imaging data in which at least one of the MTF and CTF of the imaging device is corrected can be generated.

A twelfth aspect may be configured such that, in the inspection method of the tenth aspect or the eleventh aspect, the second correction process corrects the shading characteristics in the imaging data of the imaging device as variations of the imaging device.

According to the twelfth aspect, it is possible to generate learning imaging data in which the shading characteristics of the imaging data of the imaging device are corrected.

A thirteenth aspect is the inspection method according to any one of the first to twelfth aspects, wherein the comparison reference data generating step generates the comparison reference data based on the printing conditions applied when the printed matter was generated. It is good also as a structure which generates.

According to the thirteenth aspect, it is possible to generate comparison reference data that reflects the printing conditions applied to the printing of the printed matter.

A fourteenth aspect is the inspection method according to the thirteenth aspect, wherein the step of generating comparison imaged data applies the first printing condition to the comparison reference data and the second printing condition when the printing conditions are different. Image data of printed matter to which the second printing condition is applied is converted to image data of printed matter to which the first printing condition is applied, using a second learning model generated by performing learning using the imaged data of the printed matter as learning data. It may be configured to include an imaging data conversion step of converting to .

According to the fourteenth aspect, the second learning model generated by performing learning using the learning reference data to which the first printing condition is applied and the learning imaging data to which the second printing condition is applied as learning data can be used to generate comparison imaging data from imaging data to which the second printing condition is applied.

A fifteenth aspect is the inspection method of the thirteenth aspect or the fourteenth aspect, wherein the comparison reference data generating step applies at least one of the printing medium conditions, the mesh conditions, and the color profile conditions to the printing conditions, It may be configured to generate comparison reference data.

According to the fifteenth aspect, it is possible to generate comparison reference data that reflects at least one of the print medium conditions, the mesh conditions, and the color profile conditions.

A sixteenth aspect is the inspection method according to any one of the first to fifteenth aspects, wherein the comparison reference data generating step includes a first step using the first learning model and a second step not applying the learning model. It is good also as a structure containing.

According to the sixteenth aspect, the content of deep learning can be simplified. Also, the required amount of learning data can be reduced.

A seventeenth aspect is the inspection method according to any one of the first aspect to the sixteenth aspect, wherein the comparison reference data generating step converts the color channel of the comparison reference data according to the color channel of the comparison imaging data. may be configured.

According to the seventeenth aspect, it is possible to generate the comparison reference data converted according to the color channel of the comparison imaging data.

The reference data for comparison may match the color channel with the imaging data for comparison.

An eighteenth aspect is the inspection method according to any one of the first aspect to the seventeenth aspect, wherein the comparison reference data generation step converts the resolution of the comparison reference data according to the resolution of the comparison imaging data. good too.

According to the eighteenth aspect, it is possible to generate the comparison reference data converted according to the resolution of the comparison imaging data.

The reference data for comparison may have the same resolution as the imaging data for comparison.

An inspection apparatus according to a nineteenth aspect includes a reference data acquisition unit that acquires reference data applied to printing of a printed matter, and comparison reference data that converts the reference data according to conditions of the printed matter to generate comparison reference data. a generation unit; a comparison image data generation unit that generates comparison image data using image data obtained by imaging a printed matter using an imaging device; and an inspection unit for judging quality, wherein the comparison reference data generation unit is an inspection device that converts reference data into comparison reference data using a first learning model generated by applying deep learning. .

According to the nineteenth aspect, the same effects as those of the first aspect can be obtained.

In the 19th aspect, the same matters as those specified in the 2nd to 18th aspects can be appropriately combined. In that case, the components that perform the processes and functions specified in the inspection method can be grasped as the components of the inspection apparatus that perform the corresponding processes and functions.

A program according to a twentieth aspect provides a computer with a reference data acquisition function of acquiring reference data applied to printing of a printed matter, and a comparison reference for generating comparison reference data by converting the reference data according to the conditions of the printed matter. A data generation function, a comparison imaging data generation function for generating comparison imaging data using imaging data obtained by imaging a printed matter using an imaging device, and comparing the comparison reference data with the comparison imaging data to A program for realizing an inspection function for judging quality, wherein the comparison reference data generation function converts reference data into comparison reference data using a first learning model generated by applying deep learning. is.

According to the twentieth aspect, the same effects as those of the first aspect can be obtained.

In the twentieth aspect, the same matters as those specified in the second to eighteenth aspects can be appropriately combined. In that case, the components that carry out the processes and functions specified in the inspection method can be grasped as the components of the program that carry out the corresponding processes and functions.

A printing apparatus according to a twenty-first aspect includes a transport unit that transports a print medium, a print unit that prints on the print medium transported using the transport unit to generate a printed material, and a print that is generated using the print unit. and an inspection device for inspecting printed matter, wherein the inspection device includes a reference data acquisition unit that acquires reference data applied to printing of the printed matter, and a reference data that is converted according to the conditions of the printed matter for comparison. A comparison reference data generation unit that generates reference data, a comparison image data generation unit that generates comparison image data using image data obtained by imaging a printed matter using an imaging device, and a comparison reference data and a comparison image. an inspection unit that compares the data and determines whether the printed material is good or bad, and the comparison reference data generation unit uses the first learning model generated by applying deep learning to generate the reference data for comparison. A printing device that converts to reference data.

According to the twenty-first aspect, the same effects as those of the first aspect can be obtained.

In the 21st aspect, the same matters as those specified in the 2nd to 18th aspects can be appropriately combined. In that case, the components that perform the processes and functions specified in the inspection method can be grasped as the components of the printing apparatus that perform the corresponding processes and functions.

According to the present invention, the reference data is converted into the reference data for comparison using the first learning model generated by applying deep learning, and the imaging data for comparison and the reference for comparison are based on the imaging data obtained by imaging the inspection target. data to determine whether the object to be inspected is good or bad. This makes it possible to determine the quality of the inspection target based on the comparison reference data corresponding to the printed matter to be inspected.

FIG. 1 is a flow chart showing the procedure of the inspection method according to the first embodiment. FIG. 2 is a functional block diagram of the inspection device according to the first embodiment. FIG. 3 is a functional block diagram of the reference data converter shown in FIG. FIG. 4 is an explanatory diagram of deep learning applied to the first deep learning model. FIG. 5 is a functional block diagram of a reference data converter according to the first modified example. FIG. 6 is a functional block diagram of a reference data converter according to the second modification. FIG. 7 is an explanatory diagram of the problem solved by the inspection apparatus according to the second embodiment. FIG. 8 is a functional block diagram of an inspection device according to the second embodiment. FIG. 9 is an explanatory diagram of deep learning applied to the second deep learning model. FIG. 10 is an explanatory diagram of the first deep learning model applied to the inspection apparatus according to the third embodiment. FIG. 11 is an overall configuration diagram of an inkjet printing apparatus according to the embodiment. FIG. 12 is a functional block diagram of the inkjet printer shown in FIG.

Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In this specification, the same components are denoted by the same reference numerals, and overlapping descriptions are omitted as appropriate.

[First embodiment]
[Configuration example of inspection method]
FIG. 1 is a flow chart showing the procedure of the inspection method according to the first embodiment. The inspection method described below is a method for inspecting printed matter, in which the printed matter is the object of inspection, and the imaged data for comparison generated from the imaged data of the printed matter is compared with the reference data for comparison generated from the reference data for the printed matter. Then, the quality of the printed matter is determined. Note that the printed matter represents a print medium on which printing has been performed based on the reference data input to the printing device.

The reference data for comparison is generated by applying a conversion process to the reference data of the printed matter using a deep learning model generated by performing deep learning corresponding to various printing conditions. As a result, the inspection of the printed material is performed according to various conditions related to printing.

In the reference data acquisition step S10 shown in FIG. 1, reference data to be input to the printing apparatus is acquired. As the reference data, digital data such as PDF format and TIFF format are applied. As the reference data, captured data generated by capturing an image of a printed matter printed in advance for proofreading using an imaging device may be applied. PDF is an abbreviation for Portable Document Format. TIFF is an abbreviation for Tagged Image File Format.

This embodiment shows a mode of inputting 4-channel TIFF-format reference data to which cyan, magenta, yellow, and black are applied to a printing apparatus. After the reference data acquisition step S10, the process proceeds to the reference data conversion step S12.

In the reference data conversion step S12, the reference data acquired in the reference data acquisition step S10 is converted to generate comparison reference data. A deep learning model is applied to generate the reference data for comparison. As an example of a deep learning model, FIG. 3 illustrates a first deep learning model 114 . Note that the first deep learning model 114 described in the embodiment corresponds to an example of the first learning model.

The first deep learning model 114 shown in FIG. 3 can compare 4-channel TIFF format image data to which CMYK is applied with 3-channel image data to which RGB is applied, which is read image data of an imaging device. image data. After the reference data conversion step S12, the imaging data acquisition step S14 is performed. Note that the reference data conversion step S12 described in the embodiment corresponds to an example of the comparison reference data generation step.

In the imaging data acquisition step S14, the imaging data of the printed matter is acquired. A scanner device can be applied as an imaging device that is applied to imaging printed matter. After the imaged data acquisition step S14, the process proceeds to the imaged data correction step S16. Note that the imaging data acquisition step S14 described in the embodiment corresponds to an example of the comparison imaging data generation step.

In the imaging data correction step S16, correction processing is performed on the imaging data of the printed matter acquired in the imaging data acquisition step S14. Correction processing includes resizing, rotation correction, and the like. If the imaging data of the printed matter need not be corrected, the imaging data correction step S16 is omitted. After the imaging data correction step S16, the process proceeds to the inspection step S18. Note that the imaging data correction step S16 described in the embodiment corresponds to an example of the comparison imaging data generation step.

The imaging data acquisition step S14 and the imaging data correction step S16 may be performed before the reference data acquisition step S10, or may be performed in parallel with the reference data acquisition step S10 and the reference data conversion step S12.

The inspection step S18 compares the reference data for comparison based on the reference data with the imaging data for comparison based on the imaging data of the printed matter, and determines the quality of the printed matter based on the comparison result. The comparison processing between the reference data for comparison and the imaging data for comparison can be performed by applying a known method such as comparing the feature amount of each pixel or the feature amount of each area including a plurality of pixels. After the inspection step S18, the process proceeds to the inspection result reporting step S20.

In the inspection result reporting step S20, the inspection result is reported. Examples of forms of notification include notification of visual information using a display device and notification of audio information using a speaker. After the inspection result notification step S20, the process proceeds to an inspection end determination step S22.

In the inspection end determination step S22, it is determined whether or not a specified inspection end condition is satisfied. If it is determined in the inspection termination determination step S22 that the specified termination condition is not satisfied, the determination is No. In the case of No determination, the process proceeds to the imaging data acquisition step S14, and each step from the imaging data acquisition step S14 to the inspection completion determination step S22 is performed until a Yes determination is made in the inspection completion determination step S22.

On the other hand, if it is determined in the inspection termination determination step S22 that the specified termination condition is satisfied, the determination is Yes. In the case of Yes determination, the procedure of the inspection method is terminated. Examples of prescribed end conditions include the case where a predetermined number of printed materials have been inspected, and the case where a signal indicating the end of the inspection is input from the outside.

[Configuration example of inspection device]
FIG. 2 is a functional block diagram of the inspection device according to the first embodiment. An inspection apparatus 10 shown in the figure includes a reference data acquisition unit 12 and a reference data conversion unit 14 . The inspection apparatus 10 also includes an imaging data acquisition unit 20 and an imaging data correction unit 22 . Furthermore, the inspection apparatus 10 includes a determination section 30 and an inspection result notification section 32 .

The reference data acquisition unit 12 acquires the reference data 11 input to the printing device. The reference data acquisition unit 12 transmits the reference data 11 to the reference data conversion unit 14 . The reference data can be 4-channel TIFF format digital data to which CMKY is applied. The reference data may be a TIFF format with five or more channels including one or more spot colors in CMKY, or a TIFF format with three channels to which RGB is applied.

Also, the reference data is not limited to the TIFF format. Image data in PDF format may be applied to the reference data. Image data in formats such as JPEG and BMP may be applied to the reference data. JPEG is an abbreviation for Joint Photographic Experts Group. BMP is an abbreviation for Bitmap.

The image data used as the reference data may be image data before RIP processing, or may be image data after RIP processing. Note that RIP is an abbreviation for Raster Image Processor.

Image data used as reference data may be image data after correction processing such as gamma correction processing, density unevenness correction processing, and abnormal recording element correction processing. Image data used as reference data may be dot data after halftone processing.

Further, instead of digital data input to the printing device, the reference data is image data of a printed matter printed under the same conditions as the image data of the printed matter to be inspected, or image data of a printed matter printed under different conditions. Imaging data may be applied.

The reference data conversion unit 14 converts the reference data 11 using a deep learning model to generate comparison reference data 13 . The reference data conversion unit 14 transmits the comparison reference data 13 to the determination unit 30 . Note that the reference data conversion unit 14 described in the embodiment corresponds to an example of a comparison reference data generation unit.

The imaging data acquisition unit 20 acquires imaging data 21 of a printed matter from an imaging device 40 such as a scanner device or a camera. The imaging data acquisition unit 20 transmits the imaging data 21 to the imaging data correction unit 22 .

The imaging data correction unit 22 performs correction processing such as size conversion on the imaging data 21 transmitted from the imaging data acquisition unit 20 to generate comparison imaging data 23 . Examples of correction processing include one-dimensionalization processing and filter processing of the imaging data 21 .

The imaging data correction unit 22 may output the imaging data 21 as the comparison imaging data 23 without performing the correction process on the imaging data 21 when the imaging data does not need to be corrected. The imaging data correction unit 22 transmits the comparison imaging data 23 to the determination unit 30 . Note that the imaging data acquisition unit 20 described in the embodiment corresponds to an example of a comparison imaging data generation unit. Also, the captured image data correction unit 22 described in the embodiment corresponds to an example of a comparison captured image data generation unit.

The determination unit 30 compares the comparison reference data 13 and the comparison imaging data 23 to determine whether the printed matter is good or bad. The determination unit 30 may determine the presence or absence of printing abnormality such as ink shortage and identify the type of printing abnormality based on the comparison result between the comparison reference data 13 and the comparison imaging data 23 . The determination unit 30 transmits the inspection result 31 to the inspection result notification unit 32 .

The determination unit 30 compares result information such as the intensity value of the detected defect, the size of the defect, and the length of the defect with the inspection threshold value corresponding to the result information, and determines if there is a defect exceeding the inspection threshold value. It can be determined that there is a printing abnormality in The inspection threshold value may be a fixed value held in advance by the determination unit 30, or may be arbitrarily set by the user of the inspection apparatus 10 via a user interface or the like.

A user or the like of the inspection apparatus 10 may directly set the value of the inspection threshold. A user or the like of the inspection apparatus 10 may selectively set a plurality of inspection thresholds defined according to inspection levels. As an application example of a plurality of inspection thresholds, a plurality of stepwise inspection thresholds according to inspection levels, such as a large inspection threshold, a medium inspection threshold, and a small inspection threshold, are defined, and the user of the inspection apparatus 10 or the like responds to the inspection level. Then, a mode in which a large inspection threshold, a medium inspection threshold, and a small inspection threshold are selected.

A fixed value may be applied to each of the plurality of inspection thresholds according to the inspection level, or the user of the inspection apparatus 10 may be able to change them each time. Only one inspection threshold may be held for each inspection level, or a plurality of inspection thresholds may be held for each inspection level according to printing conditions.

Examples of the printing conditions include the printing conditions 124 shown in FIG. 3, such as paper type. Other examples of printing conditions include front side printing or back side printing. Still another example of printing conditions is the type of ink.

The inspection result notification unit 32 notifies the inspection result 31 . The inspection result reporting unit 32 has a display device such as a monitor device, and can report the inspection result using character information or the like. The test result notification unit 32 has an audio output device such as a speaker, and can notify the test result using audio information. The inspection result reporting unit 32 may transmit data representing inspection results to an external device using a communication line (not shown). Examples of external devices include computers and printers.

[Configuration example of the first deep learning model]
FIG. 3 is a functional block diagram of the reference data converter shown in FIG. The reference data conversion unit 14 has a first deep learning model 114 . The first deep learning model 114 is a deep learning model that performs learning to bring the reference data 11 closer to the comparison imaging data 23 based on the imaging data 21 of the printed matter to be inspected.

The reference data converter 14 acquires the printing conditions 124 . The reference data conversion unit 14 can convert the reference data 11 into the comparison reference data 13 using the first deep learning model 114 according to the printing conditions.

The printing conditions 124 include at least one of paper type, screen type, resolution, and color profile. The paper type is the type of print medium. Examples of paper types include gloss paper and high-quality paper.

The halftone type indicates the type of halftone dot. A halftone dot is sometimes called a halftone or the like. Examples of halftone dots include AM screens, FM screens and hybrid screens. AM is an abbreviation for Amplitude Modulation. FM is an abbreviation for Frequency Modulation.

An example of resolution is print resolution expressed using dots per inch. The number of dots per inch may be described as dots per inch or DPI (dots per inch). Examples of color profiles include CMKY profiles and RGB profiles.

The reference data conversion unit 14 may include a plurality of first deep learning models 114 generated by learning each printing condition. The first deep learning model 114 may be configured to switch processing according to the printing conditions 124 .

The reference data conversion unit 14 shown in FIG. 3 transforms the reference data 11 to which TIFF format digital data is applied into comparison reference data 13 that can be compared with 3-channel comparison imaging data 23 to which RGB is applied. Convert.

[Specific example of the first deep learning model]
As a specific example of the first deep learning model 114, a deep learning model for printing the same content using both glossy paper and fine paper will be described. The first deep learning model 114 is generated by performing deep learning using input data of a printing apparatus when printing a printed matter using glossy paper and imaging data of a printed matter using high-quality paper as learning data.

The reference data conversion unit 14 receives the reference data 11 used when inspecting the printed matter using high-quality paper when inspecting the printed matter using high-quality paper. Generate the reference data for comparison 13 to be applied.

As a result, the reference data 11 used when inspecting printed matter using high-quality paper can be used when inspecting printed matter using high-quality paper, and input data for printed matter using high-quality paper can be obtained. It is possible to carry out inspection of the printed matter in which a constant level is maintained.

Here, the paper type is exemplified as the printing condition, but the same applies to the printing condition such as the halftone type illustrated in FIG. In other words, input data for printing to which arbitrary printing conditions are applied can be used as reference data for inspection of printed matter to which different printing conditions are applied, and printed matter can be inspected at a constant level.

A first deep learning model 114 may apply a neural network having an input layer, an intermediate layer and an output layer. Examples of neural networks include convolutional neural networks. A convolutional neural network may comprise one or more convolutional layers and one or more pooling layers in the intermediate layers. Note that the convolutional neural network may be described as CNN using the abbreviation of Convolution Neural Network.

The first deep learning model 114 can apply generative models such as GAN (Generative Adversarial Network), CycleGAN and VAE (Variational Auto Encoder). It should be noted that GAN may be called a generative adversarial model. A VAE may be referred to as a variational autoencoder.

[Specific examples of deep learning]
FIG. 4 is an explanatory diagram of deep learning applied to the first deep learning model. The 4-channel reference data 11 to which CMYK is applied and the 3-channel imaging data 21 to which RGB is applied have different color channels. Also, the resolution of the imaging data 21 depends on the performance of the imaging device, and generally the resolution of the imaging data 21 is different from the resolution of the reference data 11 . Then, it is difficult to compare the reference data 11 and the imaging data 21 .

Therefore, supervised deep learning is performed using learning data 100 in which learning reference data 111 is input and learning imaging data 123 is output, and a first deep learning model 114 is generated. The first deep learning model 114 performs color channel and resolution conversion processing when converting the reference data 11 into the comparison reference data 13 .

For example, various TIFF-format learning reference data 111 are input to a printing device, and printed matter is generated for each input data. An imaging device is used to capture an image of a printed matter, and learning imaging data 123 to which RGB is applied is generated. The learning reference data 111 and the learning imaging data 123 are input to the first deep learning model 114 to perform deep learning and generate the trained first deep learning model 114 .

As a result, conversion processing of reference data, which conventionally required manual design of parameters and took time and effort, is automated by applying the first deep learning model 114, and conversion processing of reference data is easily performed. It becomes possible.

Also, the learned first deep learning model 114 carries out a process of bringing the reference data closer to the captured data of the printed matter. As a result, it becomes possible to inspect the printed material while maintaining at least the same level of inspection as in the past. The input of learning data may be called input data, and the output of learning data may be called output data and correct data.

[Additional use of printed information in deep learning]
Deep learning in generating first deep learning model 114 may additionally utilize printed information 125 . Image defects due to nozzle conditions can occur in inkjet printing devices. In the inkjet printing apparatus, the occurrence of image defects is suppressed by correcting non-ejection and ejection bending for each nozzle. Printing information 125 such as nozzle information and correction parameters representing the performance of each nozzle, such as non-ejection and ejection bending, can be incorporated into deep learning, and more accurate deep learning can be performed.

When ejection failure information for each nozzle is incorporated into deep learning as the print information 125, there is a characteristic that the image structure of an area printed using normal nozzles in the image data and an area subjected to corrective ejection are different. When the characteristics of the two different image structures are reproduced by applying the conversion operation of the deep learning model, discharge failure information for each nozzle is used as one of the inputs during learning. As a result, it is possible to generate a deep learning model corresponding to changes in the image structure for the imaging data positions corresponding to the printing positions of the ejection failure nozzles.

[Modified example of output data in deep learning]
As the output data applied to deep learning, the image data after correction obtained by correcting the image data of the printed matter by applying the correction information 126 can be applied. Even when the same printing conditions are applied to imaged data captured during printing, there are two disturbance factors classified as variations in the printing apparatus and variations in the imaging apparatus.

As an example of the variation in printing apparatuses, there is variation in the transport speed of the print medium in the transport direction of the print medium. In addition, the speed here may include the meaning of speed. Fluctuations in the transport speed of the print medium in the transport direction of the print medium cause fluctuations in the imaging position in the transport direction of the print medium, causing a problem that the position of the printed matter in the imaging data differs slightly for each print medium.

As a correction for this problem, it is possible to correct the imaging position in the transport direction of the print medium using an encoder signal applied to the transport control of the print medium provided in the print medium transport unit. It should be noted that the process of correcting variations in printing apparatuses for each printing apparatus described in the embodiment corresponds to an example of the first correction process.

An example of variation in imaging devices is local differences in resolving power in imaging data due to MTF characteristics and CTF characteristics of imaging devices. Note that MTF is an abbreviation for Modulation Transfer Function. CTF is an abbreviation for Contrast Transfer Function. Correction processing for making the resolution of imaging data constant is performed using a correction table generated based on the MTF characteristics acquired in advance, with respect to local variations in resolution caused by the MTF characteristics. The same applies to local variations in resolving power due to CTF characteristics.

As another example of variations in imaging devices, shading occurs in which the edges of captured data become dark, and when a print is captured using illumination with uniform brightness, the difference between the edges and non-edges of the captured data. There is a difference in brightness. A shading characteristic correction is performed to correct the brightness of the entire captured data to uniform brightness for the difference in brightness in the captured data caused by the shading characteristic.

If there is variation in the imaged data, there is a problem that the variations in the imaged data are forcibly distinguished during learning using the imaged data as output data, and a deep learning model that is vulnerable to disturbance factors is generated. Therefore, the corrected imaging data obtained by correcting various variations of the imaging data is output. As a result, variations in the output data are reduced and more accurate deep learning can be performed as compared with the case where the imaging data before correction is used as the output data. It should be noted that the correction of variations in the imaging apparatus described in the embodiment corresponds to an example of second correction processing.

[Other modifications of output data in deep learning]
As output data in deep learning, captured image data after correction for variations between printing apparatuses can be applied. A specific example of variation between printing apparatuses is variation in ejection characteristics of each head in an inkjet printing apparatus. An example of ejection characteristics is streak-like density unevenness along a specific direction that occurs in a printed matter.

Density unevenness may occur in printed matter due to variations between heads. In addition, differences in color tone may occur due to variations between heads. Ink jet printing apparatuses perform density unevenness correction and color tone correction to correct variations between apparatuses.

When deep learning is performed by generating output data for deep learning from printed matter printed without correcting variations between apparatuses, learning to reproduce variations is performed. Then, if the deep learning model that has undergone the above learning is applied to a printing device that does not print the printed matter that serves as learning data, the printed matter may have variations different from the learning output data. A problem that is weak against disturbance factors caused by variations in Therefore, when generating learning output data, correction is performed to reduce variations between devices in printing. This can generate general-purpose and high-performance deep learning models. Note that the correction of variations between printing apparatuses described in the embodiment corresponds to an example of first correction processing.

[First modification of the reference data converter]
FIG. 5 is a functional block diagram of a reference data converter according to the first modified example. A reference data conversion unit 14A shown in the figure includes a plurality of deep learning models. In the figure, a first deep learning model 114A and a third deep learning model 114B are illustrated as multiple deep learning models. The reference data conversion unit 14A can have three or more deep learning models. Note that the third deep learning model 114B described in the embodiment corresponds to an example of the first deep learning model.

The reference data conversion unit 14A can selectively switch between the first deep learning model 114A and the third deep learning model 114B. When multiple deep learning models are provided, deep learning for each deep learning model is simplified, and learning performance can be improved.

For example, the first deep learning model 114A may be applied to transform color information, and the third deep learning model 114B may be applied to transform resolution information. Other deep learning models, not shown, may be provided to implement other conversions of color information conversion and resolution information conversion.

[Second Modification of Reference Data Conversion Unit]
FIG. 6 is a functional block diagram of a reference data converter according to the second modification. The reference data conversion unit 14C shown in the figure includes a first deep learning model 114C and an image processing unit 115 . The image processing unit 115 performs part of the conversion processing from the reference data 11 to the comparison reference data 13 performed in the first deep learning model 114 shown in FIG. 3 and the like.

Examples of processing applied to the image processing unit 115 include brightness correction and the like. A deep learning model is a statistical process, and a large amount of high-quality training data is required to generate a highly accurate deep learning model. Therefore, the image processing unit 115 performs some relatively simple processing such as luminance correction. This simplifies deep learning, maintains constant performance of the reference data conversion unit 14C, and enables reduction of learning data applied to deep learning.

Note that the first deep learning model 114C corresponds to an example of a component that implements the first step. Also, the image processing unit 115 described in the embodiment corresponds to an example of a component that performs the second step of not applying the learning model.

[Description of hardware configuration]
Each processing unit shown in FIGS. 2 to 6 can implement the functions of the inspection apparatus 10 by executing a prescribed program using hardware described below. Various processors can be applied to the hardware of each processing unit. Examples of processors include CPUs (Central Processing Units) and GPUs (Graphics Processing Units). The CPU is a general-purpose processor that executes programs and functions as each processing unit. A GPU is a processor specialized for image processing. The hardware of the processor is applied to an electric circuit in which electric circuit elements such as semiconductor elements are combined. Each control unit includes a ROM (Read Only Memory) in which programs and the like are stored and a RAM (Random Access Memory) as a work area for various calculations.

Two or more processors may be applied to one processing unit. The two or more processors may be the same type of processor or different types of processors. Also, one processor may be applied to a plurality of processing units.

The inspection apparatus 10 described with reference to FIGS. 2 to 6 can be realized by executing a specified program using the hardware described above. A program is synonymous with software. The same applies to the second embodiment and the third embodiment.

[Modified example of reference data and imaging data]
In this embodiment, the reference data 11 and the imaging data 21 exemplify a mode in which image data is applied. The reference data 11 and the imaging data 21 may not be image data. For example, the reference data 11 and the imaging data 21 may apply a one-dimensional profile of image data.

Printed matter printed using a single-pass inkjet printing apparatus may have streak-like defects extending in the transport direction of the print medium. From the standpoint of noise removal and streak-like defect information retention in the imaged data 21 of the printed matter, a blurring filter that acts only in the transport direction of the print medium may be used for the imaged data 21 of the printed matter.

When the blurring filter is used for the image data 21 of the print, a plurality of pixels along the transport direction of the print medium at each position in the width direction of the print medium orthogonal to the transport direction of the print medium are obtained for the image data 21 of the print. create a one-dimensional profile of the pixel values of .

A one-dimensional profile similar to that of the imaging data 21 is created for the reference data 11, and the one-dimensional profile based on the reference data 11 and the one-dimensional profile based on the imaging data 21 are compared to detect streak-like defects in the printed matter. can.

When a one-dimensional profile is applied as the comparison reference data 13, a deep learning model can be generated by performing deep learning using the one-dimensional profile as learning data. Deep learning, which uses one-dimensional profiles as learning data, has a reduced number of dimensions compared to deep learning, which uses image data as learning data, and can shorten the learning period and improve the performance of the deep learning model.

[Additional learning of deep learning model]
The first deep learning model 114 and the like shown in FIG. 3 may perform additional learning using the imaging data 21 inspected using the inspection apparatus 10 as the learning data 100 . This may improve the conversion performance of the first deep learning model 114 .

[Action and effect of the first embodiment]
According to the inspection method and inspection apparatus according to the first embodiment, it is possible to obtain the following effects.

[1]
Using a deep learning model generated by performing learning using reference data for comparison based on image data of printed matter instead of inspection results as learning data, input data at the time of printing is converted into reference data for comparison, and fixed Acquisition of inspection performance and efficiency of learning are achieved.

A deep learning model is used to eliminate the need for manual definition of feature values in conversion from reference data to reference data for comparison. The application of the deep learning model causes a drawback that requires a large amount of learning data compared to manual feature value definition, etc. is dramatically facilitated.

As a result, compared to the conversion of reference data using a table or the like, it is possible to significantly reduce the effort spent on parameter tuning and the like.

[2]
The first deep learning model performs learning such that the reference data for comparison has a higher similarity to the imaging data for comparison than the reference data. Thereby, the first deep learning model can generate reference data for comparison that has relatively high similarity to the imaging data for comparison.

[3]
The first deep learning model performs deep learning using comparative imaging data with different conditions of printed materials including at least one of printing conditions and printing information as learning data. Thereby, it is possible to generate the reference data for comparison that reflects the conditions of the printed matter.

[4]
The first deep learning model is a deep learning model having one or more convolution layers, and uses learning reference data in the same format as the reference data and learning imaging data in the same format as the comparison imaging data as learning data. Implement deep learning. Thereby, the first deep learning model can be generated by performing deep learning using the learning reference data in the same format as the reference data and the learning imaging data in the same format as the comparison imaging data.

[5]
The first deep learning model performs deep learning by adding print information to learning data. This can generate a first deep learning model that reflects the printed information.

[6]
The first deep learning model performs deep learning using output data of learning data corrected for variations within the printing apparatus, variations within the imaging apparatus, and variations among printing apparatuses. As a result, deep learning can be performed in which disturbance factors caused by variations in printing apparatuses are eliminated.

[7]
The reference data converter includes a first deep learning model and a third deep learning model. This may simplify deep learning for each deep learning model and improve learning performance.

[8]
The reference data converter includes a first deep learning model and an image processor. The image processor performs relatively simple image processing. This simplifies the first deep learning model. Learning data can be reduced.

[Second embodiment]
Next, a second embodiment will be described. The inspection method and inspection apparatus according to the second embodiment can perform highly accurate inspection even when the printing conditions for generating the learning data and the printing conditions for generating the printed matter to be inspected are different.

FIG. 7 is an explanatory diagram of the problem solved by the inspection apparatus according to the second embodiment. In FIG. 7, the first deep learning model 114 implemented in the inspection apparatus 10 according to the first embodiment is applied, and the comparison reference data 13A under the gloss condition and the imaging data 21A under the fine paper condition are generated. A case of comparison with comparison image data 23A under high-quality paper conditions is shown.

In such a case, a difference in data occurs due to the difference in the printing conditions of the high-quality paper conditions, making it difficult to compare the two. The inspection apparatus 10A described below can solve such problems and perform highly accurate inspection.

FIG. 8 is a functional block diagram of an inspection device according to the second embodiment. An inspection apparatus 10A shown in the figure includes an imaging data conversion unit 42 . The imaging data conversion unit 42 has a second deep learning model 142 .

The second deep learning model 142 is generated by performing learning to approximate the captured image data 21A of the printed matter to which the second printing condition is applied to the comparison reference data 13A to which the first printing condition is applied. That is, the second deep learning model 142 converts the imaged data 21A of the printed material to which the second printing condition is applied into the comparison imaged data 23B to which the first printing condition is applied.

The first printing condition is a printing condition applied to learning data when generating the first deep learning model 114 . The second printing condition is a printing condition different from the first printing condition, and is a printing condition applied to the printed material to be inspected. FIG. 8 exemplifies a gloss condition in which gloss paper is used as the print medium as the first print condition. In addition, as a second printing condition, a high-quality paper condition in which high-quality paper is used as the printing medium is exemplified.

The inspection apparatus 10A shown in FIG. 8 converts the imaged data 21A of the printed matter to which the second printing condition different from the first printing condition applied to the reference data 11A is applied to the imaged data 21A of the printed matter to which the first printing condition is applied. It converts into the corresponding comparison imaging data 23B. This facilitates comparison between the comparison reference data 13A and the comparison imaging data 23B corresponding to the first printing condition, and can improve the inspection accuracy of printed matter to which the second printing condition is applied.

The first printing condition and the second printing condition can apply other than the paper type such as halftone type and color profile. The second deep learning model 142 may collectively convert a plurality of first printing conditions and a plurality of second printing conditions.

FIG. 9 is an explanatory diagram of deep learning applied to the second deep learning model. The second deep learning model 142 performs learning using the learning data 100A that receives the learning image data 121A to which the second printing condition is applied and outputs the learning image data 123A to which the first printing condition is applied. Generated by implementation.

Note that the imaging data conversion unit 42 described in the embodiment is an example of a component that performs the imaging data conversion process. Also, the second deep learning model 142 described in the embodiment corresponds to an example of the second learning model.

[Advantages of Inspection Apparatus and Inspection Method According to Second Embodiment]
When TIFF-format digital data input to a printing apparatus is applied as the reference data 11A before conversion, the reference data greatly deviates from the captured image data of the printed matter. As a result, the first deep learning model 114 becomes complicated, the required amount of learning data increases, and the period required for learning becomes longer. Therefore, generation of the first deep learning model 114 for each printing condition poses a problem of efficiency.

On the other hand, the second deep learning model 142 that converts the imaging data 21A of the printed matter has a relatively small difference between the imaging data with different printing conditions when the same printing device is used, and the generation of the first deep learning model 114 , the second deep learning model 142 can be generated by short-term learning using less learning data.

When the printing condition is paper type, the second deep learning model 142 mainly performs color tone conversion. As a result, the second deep learning model 142 can be generated in a short period of time using less training data than the first deep learning model 114 that performs multiple transformations.

In learning when generating the second deep learning model 142, learning data without defects and learning data in which defects are intentionally inserted in the same position are used as input data and output data, and conversion to erase defects is suppressed. Compared to learning when generating the first deep learning model 114, it takes time and effort to generate the input data and output data of the learning data, but the learning required for generating the second deep learning model 142 Since the number of learning data is small and the learning period is short, there is no problem.

[Action and effect of the second embodiment]
According to the inspection method and inspection apparatus according to the second embodiment, it is possible to obtain the following effects.

[1]
The imaging data conversion unit has a second deep learning model. The second deep learning model performs deep learning to approximate the captured image data of the printed matter to which the second printing condition is applied to the comparison reference data to which the first printing condition is applied. This makes it possible to perform an inspection with a certain degree of accuracy by using reference data to which printing conditions different from those applied to printed matter are applied.

[2]
The second deep learning model is generated by implementing deep learning in which learning imaging data 121A to which the second printing condition is applied is input and learning imaging data 123A to which the first printing condition is applied is output. . As a result, a second deep learning model can be generated that enables conversion into comparison imaging data whose characteristics are closer to those of the comparison reference data in which printed matter is different.

[Third embodiment]
FIG. 10 is an explanatory diagram of a fourth deep learning model applied to the inspection apparatus according to the third embodiment. A fourth deep learning model 214 shown in the figure is generated by performing unsupervised deep learning using the learning imaging data 123 as the learning data 200 . An example of the fourth deep learning model 214 is GAN.

When a GAN is applied to the fourth deep learning model 214, only the generator among the generators and discriminators that constitute the GAN may be installed in the reference data converter of the inspection device.

The fourth deep learning model 214 according to the third embodiment can also be applied to the second deep learning model 142 according to the second embodiment. That is, the deep learning model corresponding to the second deep learning model 142 shown in FIG. 8 can be generated by performing deep learning in which the learning imaging data 123A shown in FIG. 9 is applied to the learning data.

[Action and effect of the third embodiment]
According to the inspection method and inspection apparatus according to the third embodiment, unsupervised deep learning is performed to obtain a deep learning model applied to the reference data conversion unit 14 shown in FIG. 2 and the imaging data conversion unit 42 shown in FIG. can generate a deep learning model that is applied to

[Example of application to inkjet printer]
〔overall structure〕
FIG. 11 is an overall configuration diagram of an inkjet printing apparatus according to the embodiment. The inkjet printing apparatus 300 includes a printing section 310 , a drying section 330 , an inspection section 350 and a stacking section 360 . Each part will be described in detail below.

[Printing department]
The printing unit 310 prints on the printing medium P supplied from a paper feeding unit (not shown). The printing unit 310 includes an inkjet head 312C, an inkjet head 312M, an inkjet head 312Y, and an inkjet head 312K corresponding to cyan ink, magenta ink, yellow ink, and black ink, respectively. Since the inkjet head 312C, the inkjet head 312M, the inkjet head 312Y, and the inkjet head 312K can apply the same configuration, they may be collectively referred to as the inkjet head 312 in some cases.

The inkjet head 312 is supplied with ink from an ink tank through an ink supply path. The inkjet head 312 has an ink channel, a pressure chamber and a nozzle portion. In the inkjet head 312, the ink channel communicates with the ink supply channel through the ink supply port. The ink channel communicates with the pressure chamber via the supply throttle. The pressure chamber communicates with the nozzle portion. A nozzle opening is formed at the tip of the nozzle portion.

A drop-on-demand system is applied to the inkjet head 312 . The inkjet head 312 has a piezoelectric element as a pressure generating element. A piezoelectric element is placed on the wall of the pressure chamber. When a drive voltage is supplied to the piezoelectric element, the pressure chamber is compressed according to the drive voltage, and ink droplets are ejected from the nozzle opening. A thermal method may be applied to the ejection method of the inkjet head 312 . Illustrations of ink tanks, ink supply paths, ink flow paths, pressure chambers, nozzles, supply throttles, and piezoelectric elements are omitted.

A line type is applied to the structure of the inkjet head 312 . The line-type inkjet head 312 has a plurality of nozzle openings arranged over a length corresponding to the entire length of the print medium P in the medium width direction perpendicular to the medium transport direction. A matrix arrangement can be applied to the arrangement of the plurality of nozzle openings.

The line-type inkjet head 312 can perform single-pass printing in which the print medium P and the inkjet head 312 are scanned relative to each other only once, and the entire print area of the print medium P is printed. Note that the inkjet head 312 may apply a serial method.

The printing section 310 comprises a print drum 314 . The print drum 314 has a cylindrical shape and supports the print medium P on the outer peripheral surface 314A. A medium support area for supporting the print medium P on the outer peripheral surface 314A of the print drum 314 includes a plurality of suction holes. The specified size applies to the suction holes. Also, the plurality of suction holes are arranged by applying a prescribed arrangement pattern. A plurality of suction holes are connected to the pump through channels. The illustration of the print medium P supported by the outer peripheral surface 314A of the print drum 314, the plurality of suction holes, the flow path, and the pump is omitted.

A rotating shaft 314B of the print drum 314 is connected to the rotating shaft of the motor through a connecting member. Rotation of the motor causes print drum 314 to rotate about axis of rotation 314B. As a result, the print medium P sucked and supported by the outer peripheral surface 314A of the print drum 314 is transported along the medium transport direction.

Print drum 314 includes a plurality of grippers 316 . FIG. 11 illustrates an embodiment with two grippers 316 . Gripper 316 includes a plurality of gripping claws. A plurality of gripping claws are arranged along the axial direction of the print drum 314 . A plurality of gripping claws are supported so as to be openable and closable using claw support members.

[Drying section]
The drying section 330 performs a drying process on the printing medium P printed by the printing section 310 . Drying section 330 includes a chain gripper 332 . Chain gripper 332 comprises a pair of chains 334 and a plurality of grippers 336 . The pair of chains 334 are endless and are wound around a pair of sprockets (not shown). Only one of the pair of chains 334 is shown in FIG.

A plurality of grippers 336 are attached to the chain 334 at regular intervals. Gripper 336 includes a plurality of gripping claws. A plurality of gripping claws are attached so as to be able to open and close using a claw supporting member that is passed between two chains 334 .

The chain gripper 332 receives the print medium P from the print drum 314, grips the leading edge of the received print medium P, and transports the print medium P along the medium transport direction. The chain gripper 332 opens the claws to release the print medium P when the print medium P reaches the stacking unit 360 . Arrow lines shown in FIG. 11 indicate the medium transport direction in the printing section 310 and the medium transport direction in the drying section 330 .

The transport path of the print medium P using the chain gripper 332 includes a downward transport area that transports the print medium P obliquely downward, a horizontal transport area that transports the print medium P horizontally from the end of the downward transport area, and a horizontal It includes an upward transport area that transports the print medium P obliquely upward from the end of the transport area.

The drying section 330 includes a first paper guide 340 , a second paper guide 342 and a third paper guide 344 . The first paper guide 340 , the second paper guide 342 , and the third paper guide 344 are mechanisms that guide the conveyance of the print medium P using the chain gripper 332 .

The first paper guide 340 supports the print medium P in the downward transport area. The second paper guide 342 supports the print medium P in the horizontal transport area. The third paper guide 344 supports the print medium P in the upward transport area.

A conveyor belt device 346 is applied to the second paper guide 342 . The transport belt device 346 is a device that transports the print medium P by running the transport belt 346A in a state where the print medium P is attracted to the endless transport belt 346A.

The transport belt device 346 has a plurality of suction holes on the print medium supporting surface 346B of the transport belt 346A. The specified size applies to the suction holes. Also, the plurality of suction holes are arranged by applying a prescribed arrangement pattern. The plurality of adsorption holes are connected with the pump through adsorption channels. When the pump is operated, suction pressure is generated in the plurality of suction holes. An adsorption pressure acts on the print medium P, and the print medium P is adsorbed and supported by the conveying belt 346A. Illustrations of suction holes, suction channels, and pumps are omitted.

The running speed of the transport belt 346A may be the same as the running speed of the chain gripper 332. There may be a speed difference between the running speed of the conveyor belt 346A and the running speed of the chain gripper 332 .

The transport belt 346A has a length that can absorb a plurality of print media P in the same period. The transport belt 346A shown in FIG. 11 has a length that can absorb two print media P in the same period. A form that can absorb more than one print medium P is also possible.

The drying section 330 includes a heat drying processing device 348 . The heating/drying device 348 applies heat to the print medium P printed by the printing unit 310 to evaporate the ink solvent and dry the print medium P. FIG.

〔Inspection unit〕
The inspection unit 350 includes an inspection device 352 . Inspection device 352 includes an imaging device and an illumination device. The inspection device 352 is configured by adding an imaging device and a lighting device to the inspection device 10 shown in FIG. Illustration of an imaging device and a lighting device is omitted. The inspection device 352 performs pass/fail judgment of the printed matter based on the imaging data of the printed matter.

The imaging device reads an image printed on the print medium P and generates imaging data of the printed matter. The imaging device transmits the imaging data of the printed matter to the processing section corresponding to the imaging data acquisition section 20 shown in FIG. The illumination device irradiates the printed matter with illumination light when the printed matter is read using the imaging device. The details of the inspection device 352 are the same as those of the inspection device 10 and the like shown in FIG. A detailed description of the inspection device 352 is omitted here.

[Collecting part]
The stacking section 360 includes a stacking tray 362 . The stacking tray 362 stacks the print media P that have been dried in the drying section 330 . The stacking tray 362 is provided with an elevating mechanism that maintains the stacking position of the print media P at a constant height.

[Functional Blocks of Inkjet Printer]
FIG. 12 is a functional block diagram of the inkjet printer shown in FIG. The inkjet printing device 300 has a system controller 400 . The system controller 400 has a CPU 402 , a ROM 404 and a RAM 406 . The ROM 404 and RAM 406 may be provided outside the system controller 400 .

The system controller 400 functions as a general control section that controls each section of the inkjet printing apparatus 300 in an integrated manner. In addition, the system controller 400 functions as an arithmetic unit that performs various kinds of arithmetic processing. The system controller 400 may execute programs to control each section of the inkjet printing apparatus 300 .

Furthermore, the system controller 400 functions as a memory controller that controls data reading and data writing in memories such as the ROM 404 and the RAM 406 .

The inkjet printing device 300 includes a communication section 410 and an image memory 412 . The communication unit 410 has a communication interface (not shown). The communication unit 410 can transmit and receive data to and from a host computer 414 connected to the communication interface.

The image memory 412 functions as a temporary storage unit for various data including image data. Data is read from and written to the image memory 412 through the system controller 400 . Image data fetched from the host computer 414 via the communication unit 410 is temporarily stored in the image memory 412 .

The inkjet printing apparatus 300 includes a transport control section 420 , a print control section 422 , a drying control section 424 , an accumulation control section 426 and an inspection section 350 .

The image memory 412 functions as a temporary storage unit for various data including image data. Data is read from and written to the image memory 412 through the system controller 400 . Image data fetched from the host computer 414 via the communication unit 410 is temporarily stored in the image memory 412 .

The transport control unit 420 controls the operation of the transport unit 428 for the print medium P in the inkjet printing apparatus 300 according to commands from the system controller 400 . The transport section 428 shown in FIG. 12 includes the print drum 314 and chain gripper 332 shown in FIG.

The print control unit 422 controls the operation of the print unit 310 according to commands from the system controller 400 . That is, the print control unit 422 controls ink ejection from the inkjet head 312C, the inkjet head 312M, the inkjet head 312Y, and the inkjet head 312K shown in FIG.

The print control unit 422 includes an image processing unit. The image processing unit forms dot data from input image data. The image processing section includes a color separation processing section, a color conversion processing section, a correction processing section, and a halftone processing section (not shown). Note that illustration of the image processing unit, color separation processing unit, color conversion processing unit, correction processing unit, and halftone processing unit is omitted.

The color separation processing section performs color separation processing on the input image data. For example, if the input image data is expressed in RGB, the color separation processing unit separates the input image data into data for each RGB color.

The color conversion processing unit converts the RGB-separated image data for each color into CMYK corresponding to ink colors.

The correction processing unit performs correction processing on the image data for each color converted into CMYK. Examples of correction processing include gamma correction processing, density unevenness correction processing, and abnormal recording element correction processing.

The halftone processing unit converts image data represented by multiple gradations, for example, 0 to 255, into dot data represented by binary or multivalued ternary or more values less than the number of gradations of the input image data. do.

The halftoning section applies predetermined halftoning rules. Examples of halftoning rules include dithering and error diffusion. The halftone processing rule may be changed according to the image forming conditions, the contents of the image data, and the like.

The print control unit 422 includes a waveform generation unit, a waveform storage unit, and a drive circuit. Note that illustration of the waveform generation unit, the waveform storage unit, and the driving circuit is omitted. The waveform generator generates a waveform of the driving voltage. The waveform storage unit stores the waveform of the drive voltage. A drive circuit generates a drive voltage having a drive waveform corresponding to dot data. The drive circuit supplies a drive voltage to the inkjet head 312 shown in FIG.

That is, the ejection timing and ink ejection amount for each pixel position are determined based on the dot data generated through processing using the image processing unit. A drive voltage corresponding to the ejection timing and ink ejection amount of each pixel position, and a control signal for determining the ejection timing of each pixel are generated. A drive voltage is supplied to the inkjet head 312 to cause the inkjet head 312 to eject ink. The ink ejected from the inkjet head 312 forms dots.

The drying control section 424 operates the drying section 330 according to a command from the system controller 400 . The drying control unit 424 controls the temperature of the drying gas, the flow rate of the drying gas, the injection timing of the drying gas, and the like. Also, the drying control section 424 controls the operation of the suction belt conveying device provided in the drying section 330 .

The stacking control unit 426 operates the stacking unit 360 according to commands from the system controller 400 . The stacking control unit 426 controls the operation of the lifting mechanism according to the increase or decrease of the print media P when the stacking tray 362 shown in FIG. 11 includes the lifting mechanism.

The inspection unit 350 inspects the printed matter according to a command from the system controller 400 . The inspection unit 350 transmits inspection results to the system controller 400 . The system controller 400 stores the inspection result of the printed matter in an inspection result storage unit (not shown).

The inkjet printing device 300 includes an operation unit 430 . The operation unit 430 includes operation members such as operation buttons, a keyboard, and a touch panel. The operation unit 430 may include multiple types of operation members. Illustration of the operating member is omitted.

Information input via the operation unit 430 is sent to the system controller 400 . The system controller 400 causes various processes to be executed according to information sent from the operation unit 430 .

The inkjet printing device 300 includes a display section 432 . The display unit 432 includes a display device such as a liquid crystal panel and a display driver. Illustration of the display device and the display driver is omitted. The display unit 432 causes the display device to display various types of information such as various setting information and abnormality information of the apparatus according to commands from the system controller 400 . The display unit 432 shown in FIG. 12 can function as the inspection result notification unit 32 shown in FIG.

The inkjet printing device 300 includes an image reading section 433 . The image reading unit 433 reads the print medium P printed by the printing unit 310 and generates read data. The read data is used for correction of the printing unit 310 and the like.

The inkjet printing apparatus 300 can correct the printing unit 310 using read data of the test pattern printed on the printing medium P. FIG. Correction of the printing unit 310 includes ejection failure correction, color correction, and the like. The image reading unit 433 shown in FIG. 12 can be arranged after the printing unit 310 shown in FIG.

The inkjet printing device 300 includes a parameter storage section 434 . The parameter storage unit 434 stores various parameters used in the inkjet printing device 300 . Various parameters stored in the parameter storage unit 434 are read out via the system controller 400 and set in each unit of the apparatus.

The inkjet printing device 300 includes a program storage unit 436 . The program storage unit 436 stores programs used in each unit of the inkjet printer 300 . Various programs stored in the program storage unit 436 are read out via the system controller 400 and executed in each unit of the apparatus.

Each processing unit shown in FIG. 12 can apply the same hardware as each processing unit in the inspection apparatus 10 described using FIGS. 2 to 6 . Also, each function of the inkjet printing apparatus 300 described using FIGS. 11 and 12 can be realized by executing a prescribed program using prescribed hardware.

[Example of application to other devices]
In this embodiment, the inkjet printing apparatus 300 is exemplified as an apparatus to which the inspection apparatus 10 is applied, but the inspection apparatus 10 can also be applied to an electrophotographic printing apparatus and a plate printing apparatus. Moreover, the inspection apparatus 10 can also be applied to an industrial liquid ejection apparatus for drawing wiring patterns, resist patterns, and the like. Furthermore, the inspection device 10 can also be applied to a liquid coating device having a jet nozzle.

[Example of application to program invention]
A program corresponding to the inspection apparatus and inspection method disclosed herein can be configured. That is, the present specification provides a computer with a reference data acquisition function for acquiring reference data, a reference data conversion function for converting the reference data into reference data for comparison, an image data acquisition function for acquiring image data of a printed matter, and a comparison data acquisition function. Disclosed is a program that realizes an inspection function of comparing reference data and comparative imaging data based on imaging data.

Note that the reference data conversion function described in the embodiment corresponds to an example of the comparison reference data generation function. Also, the imaging data acquisition function described in the embodiment corresponds to an example of the imaging data generation function for comparison. Furthermore, when a correction function for correcting imaging data is provided, the correction function described in the embodiment corresponds to an example of a comparison imaging data generation function.

In the embodiments of the present invention described above, it is possible to appropriately change, add, or delete constituent elements without departing from the gist of the present invention. The present invention is not limited to the embodiments described above, and many modifications are possible within the technical concept of the present invention by those skilled in the art.

10 inspection apparatus 10A inspection apparatus 11 reference data 11A reference data 12 reference data acquisition unit 13 reference data for comparison 13A reference data for comparison 14 reference data conversion unit 14A reference data conversion unit 14C reference data conversion unit 20 imaging data acquisition unit 21 imaging data 21A imaging data 22 imaging data correction unit 23 imaging data for comparison 23A imaging data for comparison 23B imaging data for comparison 30 determination unit 31 inspection result 32 notification unit 40 imaging device 42 imaging data conversion unit 100 learning data 111 reference data for learning 114 One deep learning model 114A First deep learning model 114B Third deep learning model 114C First deep learning model 115 Image processing unit 123 Learning imaging data 124 Printing conditions 125 Printing information 142 Second deep learning model 200 Learning data 214 First deep layer Learning model 300 Inkjet printing device 310 Printing unit 312 Inkjet head 312C Inkjet head 312M Inkjet head 312Y Inkjet head 312K Inkjet head 314 Printing drum 314A Outer peripheral surface 314B Rotary shaft 316 Gripper 330 Drying unit 332 Chain gripper 334 Chain 336 Gripper 340 First paper guide 342 Second paper guide 344 Third paper guide 346 Conveyor belt device 346A Conveyor belt 346B Print medium support surface 348 Heat drying device 350 Inspection unit 352 Inspection device 360 Stacking unit 362 Stacking tray 400 System controller 402 CPU
404 ROMs
406 RAM
410 Communication unit 412 Image memory 414 Host computer 420 Transport control unit 422 Print control unit 424 Drying control unit 426 Integration control unit 428 Transport unit 430 Operation unit 432 Display unit 434 Parameter storage unit 436 Program storage unit P Print media S10 to S22 Inspection method each process of

Claims (19)

a reference data acquisition step of acquiring reference data applied to printing of printed matter;
a comparison reference data generation step of generating comparison reference data by converting the reference data according to printing conditions acquired by the printing apparatus;
a comparative imaging data generation step of generating comparative imaging data using imaging data obtained by imaging the printed matter using an imaging device;
an inspection step of comparing the reference data for comparison and the imaging data for comparison to determine the quality of the printed matter;
including
In the comparison reference data generation step, a plurality of first learning models generated by applying deep learning to each of the plurality of printing conditions are switched according to the acquired printing conditions , and the reference data is changed to the An inspection method that converts to reference data for comparison. The reference data for comparison generating step receives reference data for learning as an input, inputs the reference data for learning to the printing device, and generates a printed matter, which is captured by the imaging device for learning. The deep learning is performed using data as an output, and the similarity of the reference data for comparison with the imaged data for comparison is higher than the similarity of the reference data with the imaged data for comparison. The inspection method of claim 1, wherein a first learning model is used. The first learning model is a plurality of printed materials with different printing conditions, and learns a plurality of learning imaging data generated by imaging printed materials generated by inputting the learning reference data to a printing device. 3. The inspection method according to claim 2, wherein the data is generated by performing the deep learning. The first learning model is printed matter printed by applying one or more printing conditions, and learns the imaging data for learning obtained by imaging the printed matter generated by inputting the reference data for learning to a printing device. 4. The inspection method according to claim 2, wherein the data is generated by performing the deep learning. The first learning model includes one or more convolution layers, receives the learning reference data in the same data format as the reference data, and outputs the learning imaging data in the same data format as the comparison imaging data. 4. The inspection method according to claim 2 or 3, which is generated by implementing the deep learning for learning learning data to be. The first learning model adds nozzle information representing characteristics of nozzles provided in an inkjet head in an inkjet printing apparatus or printing information including correction parameters for the characteristics of the nozzles as learning data, and performs the deep learning. 6. The inspection method according to claim 5, wherein the method is generated. The imaging data for learning is a printing device that generates a printed material based on the reference data for learning, and in a case where a different printing device generates the printed material for each of the reference data for learning, the variation of the printing device is corrected. 7. The inspection method according to claim 6, wherein the inspection method is generated using printed matter on which one correction process has been performed. 8. The inspection method according to claim 7, wherein in the first correction process, correction of ejection characteristics of the nozzles is performed as correction of variations in the printing apparatus. 9. The inspection method according to claim 7, wherein the first correction process corrects variations in the transport speed of the print medium in the single-pass printing apparatus as variations in the printing apparatus. 10. The imaging data for learning is generated using imaging data of a printed matter on which a second correction process for correcting at least one of MTF and CTF of an imaging device that images the printed matter is performed. or the inspection method according to item 1. 10. The imaging data for learning is generated using imaging data of a printed matter on which a second correction process for correcting shading characteristics in the imaging data of an imaging device that images the printed matter is performed. The inspection method described in the section. The inspection method according to any one of claims 1 to 11, wherein the comparison reference data generating step generates the comparison reference data based on the printing conditions applied when generating the printed matter. 13. The method according to claim 12, wherein the comparison reference data generation step applies at least one of printing medium conditions, mesh conditions, and color profile conditions to the printing conditions to generate the comparison reference data. Inspection method. The inspection method according to any one of claims 1 to 13 , wherein the comparison reference data generating step includes a first step using the first learning model and a second step not applying the learning model. 15. The inspection method according to any one of claims 1 to 14 , wherein the comparison reference data generation step converts color channels of the comparison reference data according to color channels of the comparison imaging data. 16. The inspection method according to any one of claims 1 to 15 , wherein the comparison reference data generation step converts the resolution of the comparison reference data according to the resolution of the comparison imaging data. a reference data acquisition unit that acquires reference data applied to printing of printed matter;
a comparison reference data generation unit that converts the reference data according to printing conditions acquired by the printing apparatus to generate comparison reference data;
a comparison imaging data generation unit that generates comparison imaging data using imaging data obtained by imaging the printed matter using an imaging device;
an inspection unit that compares the reference data for comparison with the imaging data for comparison and determines whether the printed matter is good or bad;
including
The comparison reference data generation unit switches a plurality of first learning models generated by applying deep learning to each of the plurality of printing conditions according to the acquired printing conditions , and converts the reference data to the An inspection device that converts to reference data for comparison. to the computer,
a reference data acquisition function for acquiring reference data applied to printing of printed matter;
a comparison reference data generating function for generating comparison reference data by converting the reference data according to printing conditions acquired by the printing apparatus;
A comparative imaging data generation function for generating comparison imaging data using imaging data obtained by imaging the printed matter using an imaging device, and comparing the comparison reference data and the comparison imaging data to determine the quality of the printed matter. A program that realizes an inspection function that determines
The comparison reference data generation function switches a plurality of first learning models generated by applying deep learning to each of the plurality of printing conditions according to the acquired printing conditions , and converts the reference data to the A program that converts to reference data for comparison. a transport unit that transports the print medium;
a printing unit that prints on the print medium conveyed using the conveying unit to generate a printed matter;
an inspection device that inspects the printed material generated using the printing unit;
with
The inspection device is
a reference data acquisition unit that acquires reference data applied to printing of printed matter;
a comparison reference data generation unit that converts the reference data according to printing conditions acquired by the printing apparatus to generate comparison reference data;
a comparison imaging data generation unit that generates comparison imaging data using imaging data obtained by imaging the printed matter using an imaging device;
an inspection unit that compares the reference data for comparison with the imaging data for comparison and determines whether the printed matter is good or bad;
including
The comparison reference data generation unit switches a plurality of first learning models generated by applying deep learning to each of the plurality of printing conditions according to the acquired printing conditions , and converts the reference data to the A printing device that converts to reference data for comparison.

Images