Method for creating color prediction model, color prediction model creation device, and computer program

Abstract

To provide a technique for creating a color prediction model applicable to a printing device in which a user is capable of arbitrarily selecting and using one of a plurality of ink sets.SOLUTION: A method for creating a color prediction model includes the steps of: (a) creating a first learning data group relating to a first ink set; (b) creating a second learning data group relating to a second ink set; and (c) executing learning of one color prediction model using an integrated learning data group including the first learning data group and the second learning data group. The first ink set and the second ink set include a plurality of common inks which are common and one or more non-common inks which are not common. Each of the first learning data group and the second learning data group includes a common learning data group relating to the plurality of common inks and does not include learning data relating to a non-common ink set consisting of the non-common inks of the first ink set and the non-common inks of the second ink set.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a method for creating a color prediction model, a color prediction model creating device, and a computer program. [Background technology]

[0002] In a printing device, a color conversion process is performed to convert color values ​​of input image data expressed in a first color system into color values ​​in a second color system corresponding to the type of ink. In the color conversion process, a plurality of color conversion tables are referenced to associate color values ​​in the first color system with color values ​​in the second color system. When creating such a color conversion table, a color prediction model is used to predict, from an arbitrary ink amount set, the spectral reflectance of a printed matter printed with that ink amount set. Patent Document 1 discloses a color prediction model based on the well-known spectral Neugebauer model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2006-334945 A Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, printing devices have emerged that allow users to select one of a number of ink sets at their discretion, but in the past, there had been little research into the technology for creating color prediction models that could be applied to such printing devices. [Means for solving the problem]

[0005] According to a first aspect of the present disclosure, there is provided a method for creating a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The method includes the steps of (a) creating a first learning data group for a first ink set among a plurality of ink sets that are used alternatively in the same printing device, (b) creating a second learning data group for a second ink set among the plurality of ink sets, and (c) executing learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set.

[0006] According to a second aspect of the present disclosure, there is provided a color prediction model creation device that creates a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The color prediction model creation device includes a first learning data creation unit that creates a first learning data group related to a first ink set among a plurality of ink sets that are used alternatively in the same printing device, a second learning data creation unit that creates a second learning data group related to a second ink set among the plurality of ink sets, and a learning unit that executes learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group related to the plurality of common inks, and is created so as not to include learning data related to a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set.

[0007] According to a third aspect of the present disclosure, there is provided a computer program for creating a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The computer program causes a computer to execute the following processes: (a) a process of creating a first learning data group for a first ink set among a plurality of ink sets that are used alternatively in the same printing device; (b) a process of creating a second learning data group for a second ink set among the plurality of ink sets; and (c) a process of executing learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram of a color prediction model creation system according to a first embodiment. [Diagram 2] FIG. 2 is an explanatory diagram showing the configurations of two ink sets in the first embodiment. [Diagram 3] FIG. 4 is an explanatory diagram showing an example of classification of learning data of an ink amount set. [Figure 4] 5 is a flowchart showing the procedure of a process for creating a color prediction model in the first embodiment. [Diagram 5] FIG. 4 is an explanatory diagram showing an example of a color chart. [Figure 6] FIG. 11 is an explanatory diagram showing another example of a color chart. [Figure 7] FIG. 4 is an explanatory diagram showing an ink set selection window. [Figure 8] FIG. 11 is a block diagram of a color prediction model creation system according to a second embodiment. [Figure 9]FIG. 11 is an explanatory diagram showing the configurations of three ink sets in a second embodiment. [Figure 10] 10 is a flowchart showing the procedure of a process for creating a color prediction model in a second embodiment. [Figure 11] FIG. 11 is a block diagram of a color prediction model creation system according to a third embodiment. [Figure 12] 13 is a flowchart showing the procedure of a color prediction model creation process in a third embodiment. [Figure 13] 11 is a flowchart showing the processing procedure of step S150. [Figure 14] 11 is a graph showing the relationship between ink amount and cumulative color difference. [Figure 15] 13 is a flowchart showing the processing procedure of step S160 in the third embodiment. [Figure 16] 13 is a flowchart showing the processing procedure of step S160 in the fourth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] A. First embodiment: 1 is a block diagram showing a schematic configuration of a color prediction model creation system in Embodiment 1. The color prediction model creation system 500 includes a color prediction model creation device 100, a plurality of printing devices 200, and a color measurement device 300.

[0010] The color prediction model creation device 100 creates a color prediction model CM that predicts the spectral reflectance of a printed material printed with an ink amount set used in the printing device 200, from that ink amount set. An "ink amount set" refers to a combination of ink amounts of multiple types of ink that can be used in the printing device 200. A combination of multiple types of ink that can be used in the printing device 200 is called an "ink set."

[0011] The color prediction model creation device 100 is a computer including a CPU 50, a storage unit 60, and an input / output interface 70. The CPU 50, the storage unit 60, and the input / output interface 70 are connected to each other via an internal bus so as to be able to communicate bidirectionally. A display device is connected to the color prediction model creation device 100, but is not shown in the figure.

[0012] The CPU 50 executes a color prediction model creation program 61 pre-stored in the memory unit 60, thereby functioning as a first learning data creation unit 51 that creates a first learning data group, a second learning data creation unit 52 that creates a second learning data group, and a learning unit 55 that uses these learning data groups to perform learning of the color prediction model CM.

[0013] In addition to a color prediction model creation program 61, the storage unit 60 stores color chart data 62 and ink set information 63. The color chart data 62 is image data representing a color chart including a group of multiple single-color color patches and a group of multiple mixed-color color patches.

[0014] The ink set information 63 is information that represents ink sets that can be selected by the printing device 200. The printing device 200 has a plurality of ink containers 210, and each ink container 210 contains one type of ink. In this embodiment, the number of ink containers 210 is six, and six types of ink can be used. The printing device 200 is configured so that the user can arbitrarily select and use two ink sets. In the example of FIG. 1, the first printing device 200_1 uses inks of the first ink set CMYKGrR, and the second printing device 200_2 uses inks of the second ink set CMYKGrOr. Here, C is cyan, M is magenta, Y is yellow, and K is black process ink. Gr is green, R is red, and Or is orange special color ink.

[0015] The input / output interface 70 transmits images such as the color chart data 62 to the printing device 200, and also receives the color measurement results of the print image printed on the printing medium P from the color measurement device 300.

[0016] The printing device 200 is an inkjet printer that prints an image on the printing medium P by ejecting ink onto the printing medium P. Upon receiving color chart data 62 from the color prediction model creation device 100, each of the printing devices 200_1 and 200_2 prints a color chart CC on the printing medium P in response to the color chart data 62.

[0017] The color measurement device 300 measures the color of a printed matter produced by the printing device 200. The color measurement device 300 measures the spectral reflectance R(λ) of each of a plurality of color patches included in the color chart CC. The measured spectral reflectance R(λ) is supplied to the color prediction model creation device 100.

[0018] 2 is an explanatory diagram showing the configuration of two ink sets in the first embodiment. As described above, in the first embodiment, the user can select and use one of the first ink set CMYKGrR and the second ink set CMYKGrOr in the same printing device 200. In the printing device 200_1 that uses the first ink set CMYKGrR, the Or ink in the second ink set CMYKGrOr cannot be used. Similarly, in the printing device 200_2 that uses the second ink set CMYKGrOr, the R ink in the first ink set CMYKGrR cannot be used.

[0019] In this disclosure, the following terminology is used with respect to ink sets: Combined ink set: A union of inks contained in multiple ink sets that can be alternatively used in the same printing device 200. In the example of Figure 2, the combined ink set is CMYKGrROr. Common ink set: A combination of inks commonly included in a plurality of ink sets that can be alternatively used in the same printing apparatus 200. In the example of Fig. 2, the common ink set is CMYKGr. Non-common ink set: a combination of inks that is not used in any of a plurality of ink sets that can be alternatively used in the same printing apparatus 200. In the example of Fig. 2, the non-common ink set is ROr.

[0020] As described below, when training the color prediction model CM, a combined training data set related to the ink amount set of the combined ink set is used. The combined training data set is configured not to include training data related to non-common ink sets.

[0021] FIG. 3 is an explanatory diagram showing an example of classification of learning data for ink amount sets. The ink amount of each ink in each ink amount set IS1 to IS8 is expressed as a duty. 100% duty means that that ink is applied solidly. The learning data for each ink amount set IS1 to IS8 is data that includes the value of the ink amount set and the spectral reflectance of a color patch printed with that ink amount set.

[0022] The first two ink amount sets IS1 and IS2 in Figure 3 are composed only of the ink amounts of the first ink set CMYKGrR, so their learning data belongs to the learning data group LD1e dedicated to the first ink set. The ink amount sets IS3 and IS4 are composed only of the ink amounts of the second ink set CMYKGrOr, so their learning data belongs to the learning data group LD2e dedicated to the second ink set. The ink amount sets IS5 and IS6 are composed only of the ink amounts of the common ink set CMYKGr, so their learning data belongs to the common learning data group CLD that is used in common by the first ink set and the second ink set. The ink amount sets IS7 and IS8 include the ink amounts of the non-common ink set ROr, so their learning data is not used.

[0023] The learning data group LD1 for the first ink set is configured to include a learning data group LD1e dedicated to the first ink set and a common learning data group CLD. The learning data group LD2 for the second ink set is configured to include a learning data group LD2e dedicated to the second ink set and a common learning data group CLD. The integrated learning data group TLD is configured to include a learning data group LD1e dedicated to the first ink set, a learning data group LD2e dedicated to the second ink set, and a common learning data group CLD.

[0024] The training data group is preferably created to include a teacher data group and a verification data group. The verification data group is created as a training data group relating to an ink amount set different from the ink amount set of the teacher data group. However, in the following description, the term "training data group" refers to a teacher data group unless otherwise specified, and the verification data group is mentioned as necessary. The same applies to other embodiments.

[0025] FIG. 4 is a flowchart showing the procedure of the process of creating a color prediction model in the first embodiment. In step S10, the user inputs the creation conditions of the color prediction model CM. The creation conditions of the color prediction model CM include, for example, the types of ink contained in the first ink set and the second ink set, the ink amount duty limit value, and the printing conditions of the color chart. The ink amount duty limit value is set to include the ink amount duty limit value for each ink and the total ink amount duty limit value. The printing conditions include the type of printing medium P and the printing resolution. In step S20, the printing device 200 prints the color chart CC on the printing medium P using the color chart data 62.

[0026] FIG. 5 is an explanatory diagram showing an example of a color chart CC. In this example, a first color chart CC1 for the first ink set is printed by the first printing device 200_1, and a second color chart CC2 for the second ink set is printed by the second printing device 200_2. These color charts CC1 and CC2 are printed on different printing media. The first color chart CC1 includes color patches dedicated to the first ink set and color patches for the common ink set. The second color chart CC2 includes only color patches dedicated to the second ink set. That is, color patches for the common ink set are included only in the first color chart CC1, and are not included in the second color chart CC2. In this embodiment, the process of printing the color charts and measuring the colors can be simplified.

[0027] FIG. 6 is an explanatory diagram showing another example of the color chart CC. In this example, the first color chart CC1 printed by the first printing device 200_1 includes color patches dedicated to the first ink set and color patches for the common ink set. The second color chart CC2 printed by the second printing device 200_2 includes color patches dedicated to the second ink set and color patches for the common ink set. The color patches for the common ink set included in the first color chart CC1 and the color patches for the common ink set included in the second color chart CC2 are the same. In this case, it is preferable that the spectral reflectance of the common learning data group CLD is set to the average value of the spectral reflectance measured using the first color chart CC1 and the spectral reflectance measured using the second color chart CC2. In this aspect, the error of the spectral reflectance in the common learning data group CLD can be reduced. The printing of the color chart and the method of determining the spectral reflectance shown in FIG. 5 and FIG. 6 can be applied to other embodiments.

[0028] In step S30 of Fig. 4, the color measurement device 300 measures the spectral reflectance R(λ) of each color patch of the color chart CC. In step S40, the first learning data creation unit 51 generates a first learning data group LD1 for the first ink set. As illustrated in Fig. 3, the first learning data group LD1 is created to include a learning data group LD1e dedicated to the first ink set and a common learning data group CLD for the common ink set. Each learning data includes an ink amount set and a spectral reflectance R(λ).

[0029] In step S50, the second learning data creation unit 52 extracts a common learning data group CLD from the first learning data group LD1 for the first ink set. In step S60, the second learning data creation unit 52 generates a second learning data group LD2 for the second ink set. The second learning data group LD2 is created so as to include a learning data group LD2e dedicated to the second ink set and the common learning data group CLD.

[0030] As can be understood from the above description, each of the first learning data group LD1 and the second learning data group LD2 is created to include the common learning data group CLD related to the common ink set, and not to include learning data related to the non-common ink set, thereby making it possible to reduce the number of data in the learning data group.

[0031] In step S70, the learning unit 55 executes learning of one color prediction model CM using an integrated learning data group TLD including the first learning data group LD1 and the second learning data group LD2. The color prediction model CM can be, for example, a regression model of multidimensional output configured by a neural network. By executing machine learning using the integrated learning data group TLD, such a color prediction model CM becomes an inference model that can predict the spectral reflectance from any ink amount set of the ink amount set of the first ink set and the ink amount set of the second ink set.

[0032] The multiple inputs of the color prediction model CM created in step S70 are ink amount sets for the integrated ink set CMYKGrROr. Because this color prediction model CM can be applied to two ink sets that are used alternatively, the learning time can be reduced compared to when learning color prediction models dedicated to each ink set separately.

[0033] 7 is an explanatory diagram showing a selection window for an ink set used when estimating spectral reflectance using a trained color prediction model CM. In this selection window W1, the user can select either the first ink set CMYKGrR or the second ink set CMYKGrOr. When the first ink set CMYKGrR is selected, the spectral reflectance for the ink amount set of the first ink set CMYKGrR is estimated using the color prediction model CM. On the other hand, when the second ink set CMYKGrOr is selected, the spectral reflectance for the ink amount set of the second ink set CMYKGrOr is estimated using the color prediction model CM.

[0034] As described above, in the first embodiment, the first training data group and the second training data group are each created so as not to include training data related to a non-common ink set, so that the amount of data can be reduced and the training time can be shortened.

[0035] B. Second embodiment: Fig. 8 is a block diagram of a color prediction model creation system in the second embodiment. The only differences from the first embodiment shown in Fig. 1 are that a third learning data creation unit 53 is added and that three printing devices 200_1 to 200_3 are included, and the other configurations are the same as those of the first embodiment. In the second embodiment, the printing device 200 is configured so that the user can arbitrarily select and use one of three ink sets.

[0036] FIG. 9 is an explanatory diagram showing the configuration of three ink sets in the second embodiment. The first ink set CMYKGrR and the second ink set CMYKGrOr are the same as those ink sets in the first embodiment. The third ink set CMYKGrB contains a B (blue) spot color ink. In the example of FIG. 9, the integrated ink set is CMYKGrROrB. Also, the common ink set is CMYKGr, and the non-common ink set is ROrB. The third printing device 200_3 performs printing using the third ink set CMYKGrB.

[0037] Fig. 10 is a flowchart showing the procedure of the process of creating a color prediction model in the second embodiment. The only difference from the procedure in the first embodiment shown in Fig. 4 is that step S100 is added between step S60 and step S70, and the other steps are almost the same as those in Fig. 4.

[0038] In step S100, the third learning data creation unit 53 generates a third learning data group LD3 for the third ink set. The third learning data group LD3 is created to include a learning data group dedicated to the third ink set and a common learning data group CLD. That is, the first learning data group LD1, the second learning data group LD2, and the third learning data group LD3 are each created to include the common learning data group CLD, and not to include learning data related to a non-common ink set.

[0039] In step S70 of FIG. 10, the learning unit 55 executes learning of one color prediction model CM using an integrated learning data group TLD including the first learning data group LD1, the second learning data group LD2, and the third learning data group LD3.

[0040] It should be noted that even when the same printing device 200 is capable of selectively using four or more ink sets, the color prediction model CM can be created in a similar manner to that shown in FIG. 4 or FIG.

[0041] As described above, in the second embodiment, the first learning data group, the second learning data group, and the third learning data group are each created so as not to include learning data related to non-common ink sets, thereby making it possible to reduce the amount of data and also shorten the learning time.

[0042] C. Third embodiment: Fig. 11 is a block diagram of a color prediction model creation system in the third embodiment. The only difference from the first embodiment shown in Fig. 1 is that a tentative learning data candidate creation unit 110, a learning data creation unit 120, and a learning unit 130 are provided instead of the first learning data creation unit 51, the second learning data creation unit 52, and the learning unit 55, and the other configurations are the same as those of the first embodiment. In the third embodiment, as in the first embodiment, the printing device 200 is configured so that the user can arbitrarily select and use one of the first ink set CMYKGrR and the second ink set CMYKGrOr.

[0043] The provisional learning data candidate creation unit 110 creates a provisional learning data candidate group using the simplified color prediction model SM. Here, the "provisional learning data candidate group" refers to a data group used in the processing of the learning data creation unit 120, and is different from the final learning data group created by the learning data creation unit 120. The term with "provisional" attached means that the spectral reflectance is not actually measured but is predicted by the simplified color prediction model SM.

[0044] The learning data creation unit 120 includes a first learning data creation unit 121 and a second learning data creation unit 122. The first learning data creation unit 121 creates a first learning data group LD1 for the first ink set through repeated learning of the first temporary color prediction model PM1. The second learning data creation unit 122 creates a second learning data group LD2 for the second ink set through repeated learning of the second temporary color prediction model PM2.

[0045] The learning unit 130 executes learning of one color prediction model CM using an integrated learning data group TLD including the first learning data group LD1 and the second learning data group LD2.

[0046] The first learning data creation unit 121, the second learning data creation unit 122, and the learning unit 130 in the third embodiment roughly correspond to the first learning data creation unit 51, the second learning data creation unit 52, and the learning unit 55 in the first embodiment.

[0047] The meanings of the terms used in the third embodiment are as follows: Although the reference numbers of the terms for the first ink set and the reference numbers of the terms for the second ink set are suffixed with the additional reference numbers "1" and "2", respectively, these additional reference numbers will be omitted in the following explanation of the terms.

[0048] Provisional learning data group PLD: A learning data group determined by repeated learning of the provisional color prediction model PM. As mentioned above, the reason for the "provisional" is that the spectral reflectance is not actually measured, but predicted by the simplified color prediction model SM. The provisional learning data group PLD determined by repeated learning of the provisional color prediction model PM is then turned into the final learning data group LD by actually measuring the spectral reflectance.

[0049] Provisional training data candidate group PLC: A group of candidates for data to become the provisional training data group PLD. The provisional training data group PLD is selected from the provisional training data candidate group PLC.

[0050] Simple color prediction model SM: A color prediction model for predicting the spectral reflectance used in the provisional training data candidate group PLC. It is preferable that the simple color prediction model SM is a model that can be trained more easily than the final color prediction model CM. "Easily trainable" means that the amount of data in the training data group is small and the training time is short.

[0051] Temporary color prediction model PM: A color prediction model used when determining the temporary learning data group PLD by repeated learning. The temporary color prediction model PM is preferably a neural network having the same configuration as the final color prediction model CM. "Same configuration" means that the number of input dimensions and the number of output dimensions are the same, and also that the configurations of the input layer, intermediate layer, and output layer are the same. Since the number of input dimensions of the temporary color prediction model PM is the same as that of the integrated ink set, for example, when learning or predicting the first temporary color prediction model PM1 for the first ink set, it is preferable to set the input values ​​for inks not included in the first ink set to 0. The same applies to the second temporary color prediction model PM2 for the second ink set.

[0052] Fig. 12 is a flowchart showing the procedure of the process of creating a color prediction model in the third embodiment. Steps S110 to S130 are almost the same as the processes of steps S10 to S30 in Fig. 4. That is, in step S110, the user inputs the conditions for creating the color prediction model CM. In step S120, the printing device 200 prints a color chart CC on the printing medium P using the color chart data 62. In step S130, the color measurement device 300 measures the spectral reflectance R(λ) of each color patch of the color chart CC.

[0053] In step S140, the tentative learning data candidate creation unit 110 creates a simplified color prediction model SM using the measurement results of the spectral reflectance for the color chart CC. The simplified color prediction model SM is a model that predicts the spectral reflectance from an ink amount set, and can be configured as a color prediction model based on a known spectral Neugebauer model, for example. As a color prediction model based on the spectral Neugebauer model, the printing model described in the above Patent Document 1 can be used. Note that a color prediction model other than the spectral Neugebauer model may be used as the simplified color prediction model. However, it is preferable that the simplified color prediction model SM requires a smaller amount of data in the learning data group required for learning than the color prediction model CM that is finally created, and that the learning time is also shorter.

[0054] In step S150, the tentative learning data candidate creation unit 110 creates a tentative learning data candidate group PLC using the simplified color prediction model SM.

[0055] 13 is a flowchart showing the processing procedure of step S150. In step S51, the tentative learning data candidate creation unit 110 determines the lattice point positions of the tentative learning data candidate group PLC from the colorimetric results of the single-color patches. The colorimetric value Lab of the single-color patch can be calculated from its spectral reflectance. The lattice point positions of the tentative learning data candidate group PLC can be determined using the cumulative color difference of multiple single-color patches. The lattice point positions refer to positions in a color solid represented by an ink amount set.

[0056] FIG. 14 is a graph showing the relationship between the ink amount and the accumulated color difference for each of cyan ink and magenta ink. The horizontal axis of FIG. 14 indicates the ink amount, and the vertical axis indicates the accumulated color difference ΣΔE. The accumulated color difference ΣΔE is a value obtained by accumulating the color difference between adjacent single-color color patches in a gradation patch in which a plurality of single-color color patches are arranged. The lattice point positions of the provisional learning data candidate group PLC are indicated by black circles, and are determined so that the differences in the accumulated color difference ΣΔE are equal to each other. In the graph 81c for cyan ink, the lattice point positions of the provisional learning data candidate group PLC are points where the cyan ink amount is 0%, 5%, 15%, 45%, and 100%. In the graph 81m for magenta ink, the lattice point positions of the provisional learning data candidate group PLC are points where the magenta ink amount is 0%, 7%, 17%, 37%, and 100%. In this way, the lattice point positions for each ink are set separately. Similarly, the lattice point positions of the provisional learning data candidate group PLC are determined for each of the other inks included in the combined ink set.

[0057] When a provisional training data candidate group and a provisional verification data candidate group are generated as the provisional training data candidate group PLC, the lattice point positions of the two groups are determined to be at least partially different. For example, the lattice point positions of the provisional training data candidate group can be determined to be positions where the differences in accumulated color differences ΣΔE are equal to each other as shown in Fig. 14, and the lattice point positions of the provisional verification data candidate group can be determined to be positions slightly shifted from the lattice point positions of the provisional training data candidate group.

[0058] 12, the tentative learning data candidate creation unit 110 determines an ink amount set of the tentative learning data candidate group PLC. For example, the ink amount set of the tentative learning data candidate group PLC is created as the sum of all combinations of lattice point positions determined in step S51 for each ink of the first ink set and all combinations of lattice point positions determined in step S52 for each ink of the second ink set. Note that if the ink amount set exceeds the duty limit value, it is preferable to adjust the ink amount so that it falls within the duty limit value.

[0059] In step S53, the tentative learning data candidate creation unit 110 estimates the spectral reflectance of the tentative learning data candidate group PLC using the simplified color prediction model SM. Note that the ink amount sets of the tentative learning data candidate group PLC may be added or adjusted by dividing the color solid in the L*a*b* space into multiple cells, and each cell may include one or more ink amount sets.

[0060] In step S54, the provisional learning data candidate creation unit 110 creates a provisional learning data candidate group PLC. Each data in the provisional learning data candidate group PLC includes the ink amount set determined in step S52 and the spectral reflectance estimated in step S53. The provisional learning data candidate group PLC is preferably created to include a provisional teacher data candidate group and a provisional verification data candidate group. Each data in the provisional learning data candidate group PLC is preferably provided with a label that can distinguish whether the data is for the first ink set or the second ink set. Furthermore, each data in the provisional learning data candidate group PLC is preferably provided with a label indicating that the data is for a common ink set.

[0061] 12, where the learning data creation unit 120 creates learning data groups LD1 and LD2 through learning of the temporary color prediction models PM1 and PM2. That is, the first learning data creation unit 121 creates the first learning data group LD1 through learning of the first temporary color prediction model PM1, and the second learning data creation unit 122 creates the second learning data group LD2 through learning of the second temporary color prediction model PM2.

[0062] 15 is a flowchart showing the processing procedure of step S160 in the third embodiment. In step S61, necessary parameters are read in. The parameters include parameters representing the configurations of the color prediction models PM1, PM2, and CM, and the value of the integer N used in steps S66 and S76.

[0063] In step S62, the learning data creation unit 120 reads the provisional learning data candidate group PLC.

[0064] In step S63, the first learning data creation unit 121 selects and sets an initial first provisional learning data group PLD1 from the provisional learning data candidate group PLC. The initial first provisional learning data group PLD1 is preferably a data group relating to a plurality of ink amount sets for the first ink set, including at least an ink amount set representing white (paper white), an ink amount set representing each single ink color, and a maximum ink amount set in mixed colors. The first provisional learning data group PLD1 is used as training data for the first provisional color prediction model PM1.

[0065] In step S64, the first learning data creation unit 121 performs machine learning of the first temporary color prediction model PM1 using the first temporary learning data group PLD1. In step S65, the first learning data creation unit 121 performs color difference evaluation of the temporary learning data candidate group PLC using the first temporary color prediction model PM1. For example, among the temporary learning data candidate group PLC, the ink amount sets of all the temporary learning data candidates related to the first ink set are input to the first temporary color prediction model PM1, and the spectral reflectance for each ink amount set is estimated. Then, the color difference between the colorimetric value Lab calculated from the estimated spectral reflectance and the colorimetric value Lab calculated from the spectral reflectance included in the temporary learning data candidate group PLC is calculated.

[0066] In step S66, the first learning data creation unit 121 adds the N provisional learning data candidates having the largest color difference calculated in step S65 to the first provisional learning data group PLD1. The value of N is preset to an integer equal to or greater than 1 in the above-mentioned step S61. The "N provisional learning data candidates having the largest color difference" are also referred to as "worst color difference candidates." The reason for adding the worst color difference candidates to the first provisional learning data group PLD1 is to efficiently reduce the prediction error of the first provisional color prediction model PM1. That is, when step S64 is next executed, machine learning is performed using the worst color difference candidates, so that the prediction error of the first provisional color prediction model PM1 can be significantly reduced.

[0067] In step S67, the first learning data creation unit 121 determines whether a preset termination condition is satisfied. For example, the termination condition may be that the prediction accuracy of the spectral reflectance is equal to or lower than a target value. For example, the average color difference calculated for the group of provisional verification data candidates for the first ink set may be used as the prediction accuracy of the spectral reflectance.

[0068] If the termination condition is not met, the process from step S64 onwards is executed again. On the other hand, if the termination condition is met, the iterative learning from steps S64 to S67 is terminated and the process proceeds to step S68. Note that if the termination condition is not met even after the number of repetitions reaches the upper limit, it is preferable to return to step S63, reset the initial first provisional learning data group PLD1, and execute the process from step S64 onwards.

[0069] In step S68, the first learning data creation unit 121 actually measures the spectral reflectance of the first provisional learning data group PLD1 to create the first learning data group LD1. Specifically, the ink amount set of the first provisional learning data group PLD1 is used to print a plurality of color patches with the first printing device 200_1, the color measurement device 300 is used to measure the spectral reflectance of the plurality of color patches, and the first learning data group LD1 is created using the measured spectral reflectance. The ink amount set of the first learning data group LD1 is the same as the ink amount set of the first provisional learning data group PLD1. Since the first learning data group LD1 uses the actually measured spectral reflectance, it is a data group with high reliability as a teacher data group for the final color prediction model CM.

[0070] When a verification data group is created as the first learning data group LD1 in addition to the teacher data group, a different ink amount set from the teacher data group is selected as the ink amount set of the verification data group, and the spectral reflectance is measured for them. As a method for selecting the ink amount set of the verification data group, for example, a method can be used in which the spectral reflectance of each of the provisional verification data candidates for the first ink set is estimated using the trained first provisional color prediction model PM1 and the trained second provisional color prediction model PM2, the color difference is calculated from the spectral reflectance, and a specified number of data is selected according to a selection rule according to the color difference. As the selection rule, for example, a rule of sorting in descending order of color difference and sampling a specified number of data at equal intervals, or a rule of selecting a specified number in descending order of color difference can be used. However, such a selection method uses two machine-learned provisional color prediction models PM1 and PM2, and is therefore executed after step S78 in FIG. 15.

[0071] After step S68, the process proceeds to step S72, where the second learning data creation unit 122 starts processing. In step S72, the second learning data creation unit 122 extracts a common temporary learning data group CPLD for the common ink set from the first temporary learning data group PLD1 created in the processes of steps S64 to S67. In step S73, the second learning data creation unit 122 selects and sets an initial second temporary learning data group PLD2 for the second ink set from the temporary learning data candidate group PLC. This initial second temporary learning data group PLD2 is created to include the common temporary learning data group CPLD extracted in step S72, and preferably includes a data group relating to a plurality of ink amount sets for the second ink set including at least an ink amount set representing white (paper white), an ink amount set representing each single color of ink, and a maximum ink amount set in mixed colors, as in step S63.

[0072] The processing of steps S74 to S78 is almost the same as steps S64 to S68 described above. That is, the processing of steps S74 to S78 is substantially the same as the explanation of steps S64 to S68, except that the letter or code "1" representing the first ink set is replaced with the letter or code "2" representing the second ink set. A second learning data group LD2 for the second ink set is created by the processing of steps S74 to S78.

[0073] The processing of steps S64 to S67 corresponds to a first iterative learning process of determining the first tentative learning data group PLD1 by repeating a first learning process of increasing the first tentative learning data group PLD1 so that the first prediction accuracy of the first tentative color prediction model PM1 satisfies the second target accuracy. The processing of steps S74 to S77 corresponds to a second iterative learning process of determining the second tentative learning data group PLD2 by repeating a second learning process of increasing the second tentative learning data group PLD2 so that the second prediction accuracy of the second tentative color prediction model PM2 satisfies the second target accuracy. In the third embodiment, such first iterative learning process and second iterative learning process are executed in series in this order, so that the common tentative learning data group CPLD extracted from the first tentative learning data group PLD1 can be used as it is as the common tentative learning data group of the second tentative learning data group PLD2.

[0074] In step S170 of Fig. 12, the learning unit 130 uses the integrated learning data group TLD to learn one color prediction model CM. The integrated learning data group TLD is created so as to include the first learning data group LD1 and the second learning data group LD2 created in step S160. When neural networks having the same configuration are used as the color prediction models PM1, PM2, and CM, when learning of the color prediction model CM is started, the average value of the internal parameters of the two trained temporary color prediction models PM1 and PM2 may be set as the initial value of the internal parameters of the color prediction model CM. In this way, learning of the color prediction model CM can be executed more efficiently.

[0075] 15 before step S72, step S68 may be executed simultaneously with step S78. In this case, the first tentative learning data group PLD1 and the second tentative learning data group PLD2 are integrated to generate an integrated tentative learning data group, and the integrated learning data group TLD can be generated by actually measuring the spectral reflectance of the integrated tentative learning data group. This process is also common to the process of the third embodiment in which steps S68 and S78 are executed separately in that the first learning data group LD1 and the second learning data group LD2 are essentially generated, and the two processes are equivalent.

[0076] As described above, in the third embodiment, as in the first embodiment, the first and second learning data groups are created so as not to include learning data related to non-common ink sets, so that the number of data can be reduced and the learning time can be shortened. In addition, in the third embodiment, temporary learning data groups PLD1 and PLD2 are created so that the prediction accuracy of the temporary color prediction models PM1 and PM2 meets the target accuracy, and the spectral reflectance is measured to create the learning data group, so that a final color prediction model CM can be created with a learning data group having a relatively small amount of data. Note that, as in the second embodiment described above, the third embodiment can also be extended to a case where three or more ink sets can be used alternatively. This point is also the same for the fourth embodiment described below.

[0077] D. Fourth embodiment: Fig. 16 is a flowchart showing the processing procedure of step S160 in the fourth embodiment. The device configuration of the fourth embodiment is the same as that of the third embodiment shown in Fig. 11, and the overall processing procedure is also the same as that shown in Fig. 12. The main difference from the third embodiment is only the detailed procedure of step S160. Specifically, the fourth embodiment differs from the third embodiment in three points: steps S63 to S67 and steps S73 to S77 are executed in parallel, step S72 is omitted, and steps S81, S82, S91, and S92 are added. The rest is almost the same as the third embodiment.

[0078] In steps S63 and S73, an initial first provisional learning data group PLD1 and an initial second provisional learning data group PLD2 are selected and set from the provisional learning data candidate group PLC. However, the processing contents of steps S63 and S73 in the fourth embodiment are slightly different from those of the third embodiment. That is, in steps S63 and S73 in the fourth embodiment, selection is performed so that the common provisional learning data group CPLD included in the first provisional learning data group PLD1 and the common provisional learning data group CPLD included in the second provisional learning data candidate group PLC2 are the same data group.

[0079] The processing contents of the added steps S81, S82, S91, and S92 are as follows. In step S81, the first learning data creation unit 121 determines whether or not data for the common ink set is present among the N worst color difference candidates added in step S66. If the worst color difference candidates do not include data for the common ink set, the process proceeds from step S81 to step S67. On the other hand, if the worst color difference candidates include data for the common ink set, the process proceeds from step S81 to step S82, and the worst color difference candidate is added to the provisional learning data group for the other ink set. The phrase "added to the provisional learning data group for the other ink set" means that if the worst color difference candidate was added to the first provisional learning data group PLD1 for the first ink set, the worst color difference candidate is added to the second provisional learning data group PLD2 for the second ink set in step S82. When step S82 is executed, the worst color difference candidate is added to the second provisional learning data group PLD2, so the process proceeds to step S74, where learning of the second provisional color prediction model PM2 is executed again.

[0080] As in step S81, in step S91, the second learning data creation unit 122 determines whether or not data for a common ink set is present among the N worst color difference candidates added in step S76. If the worst color difference candidates do not include data for a common ink set, the process proceeds from step S91 to step S77. On the other hand, if the worst color difference candidates include data for a common ink set, the process proceeds from step S91 to step S92, where the worst color difference candidate is added to the tentative learning data group for the other ink set. That is, in step S92, the worst color difference candidate is added to the first tentative learning data group PLD1 for the first ink set. When step S92 is executed, the worst color difference candidate is added to the first tentative learning data group PLD1, so the process proceeds to step S64, where learning of the first tentative color prediction model PM1 is executed again.

[0081] The above-described fourth embodiment also provides substantially the same effects as those of the third embodiment. Moreover, in the fourth embodiment, the first iterative learning process of steps S64 to S67 and the second iterative learning process of steps S74 to S77 are performed in parallel, so that the first provisional learning data group PLD1 and the second provisional learning data group PLD2 can be determined in a shorter time.

[0082] Other forms: The present disclosure is not limited to the above-mentioned embodiment, and can be realized in various forms without departing from the spirit of the present disclosure. For example, the present disclosure can be realized in the following forms. The technical features in the above-mentioned embodiments corresponding to the technical features in each form described below can be appropriately replaced or combined in order to solve some or all of the problems of the present disclosure, or to achieve some or all of the effects of the present disclosure. Furthermore, if the technical feature is not described as essential in this specification, it can be appropriately deleted.

[0083] (1) According to a first aspect of the present disclosure, there is provided a method for creating a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The method includes the steps of (a) creating a first learning data group for a first ink set among a plurality of ink sets that are used alternatively in the same printing device, (b) creating a second learning data group for a second ink set among the plurality of ink sets, and (c) executing learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set. According to this method, the first training data group and the second training data group are each created so as not to include training data related to a non-common ink set, making it possible to reduce the amount of data. Also, since a single color prediction model that can be applied to two or more ink sets that are used alternatively is created, the training time can be reduced.

[0084] (2) In the above method, the multiple inputs of the color prediction model may be multiple ink amounts for multiple types of ink contained in the multiple ink sets. According to this method, color prediction for multiple ink sets can be performed using one color prediction model.

[0085] (3) The above method may further include a step of creating a third learning data group for a third ink set including inks that are not common to either the first ink set or the second ink set, and step (c) may include a step of creating the integrated learning data group so as to include the first learning data group, the second learning data group, and the third learning data group. According to this method, it is possible to create one color prediction model that can be applied to three or more ink sets that are used alternatively.

[0086] (4) The above method may further include a step of creating a tentative training data candidate group using a spectral reflectance predicted by a simplified color prediction model that can be learned more easily than the color prediction model. In this case, the step (a) may include a step of (a1) determining the first tentative training data group by repeating a first learning process that executes learning of a first tentative color prediction model using a first tentative training data group selected from the tentative training data candidate group and increases the first tentative training data group so that a first prediction accuracy of the first tentative color prediction model satisfies a first target accuracy, and (a2) creating the first training data group by actually measuring the spectral reflectance of the first tentative training data group. Furthermore, the step (b) may include the steps of: (b1) determining the second provisional training data group by repeating a second learning process in which a second provisional color prediction model is trained using a second provisional training data group selected from the provisional training data candidate group, and the second provisional color prediction model increases the second provisional training data group so that a second prediction accuracy of the second provisional color prediction model satisfies a second target accuracy; and (b2) creating the second training data group by actually measuring the spectral reflectance of the second provisional training data group. According to this method, a provisional training data group is created so that the prediction accuracy of the provisional color prediction model satisfies the target accuracy, and then the spectral reflectance is measured to create a training data group. Therefore, it is possible to create a single final color prediction model from a training data group with a relatively small amount of data.

[0087] (5) In the above method, the steps (a1) and (b1) may be executed serially in this order, and the step (b1) may include a step of extracting a common tentative learning data group for the plurality of common inks from the first tentative learning data group and including the common learning data group in the second tentative learning data group. According to this method, the common provisional training data group extracted from the first provisional training data group can be used as it is as the common provisional training data group for the second provisional training data group.

[0088] (6) In the above method, the steps (a1) and (b1) may be executed in parallel, and each of the steps (a1) and (b1) may include a step of sharing a common tentative learning data group for the plurality of common inks. According to this method, the first provisional training data group and the second provisional training data group can be determined in a shorter time.

[0089] (7) In the above method, step (a2) may include printing a first color chart according to the ink volume set of the first provisional learning data group in order to actually measure the spectral reflectance of the first provisional learning data group, and step (b2) may include printing a second color chart on a printing medium different from that of the first color chart in accordance with the ink volume set of the second provisional learning data group in order to actually measure the spectral reflectance of the second provisional learning data group. According to this method, the spectral reflectance can be actually measured using the first color chart and the second color chart.

[0090] (8) In the above method, the color patches reproduced using only the common inks may be printed on only one of the first color chart and the second color chart. This method simplifies the printing and color measurement processes.

[0091] (9) In the above method, a plurality of color patches reproduced using only the plurality of common inks may be printed on the first color chart and the second color chart, respectively, and the spectral reflectance of the common learning data group may be set to an average value of a first spectral reflectance measured using the first color chart and a second spectral reflectance measured using the second color chart. According to this method, it is possible to reduce errors in the spectral reflectance in the common training data group.

[0092] (10) According to a second aspect of the present disclosure, there is provided a color prediction model creation device that creates a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The color prediction model creation device includes a first learning data creation unit that creates a first learning data group related to a first ink set among a plurality of ink sets that are used alternatively in the same printing device, a second learning data creation unit that creates a second learning data group related to a second ink set among the plurality of ink sets, and a learning unit that executes learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group related to the plurality of common inks, and is created so as not to include learning data related to a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set.

[0093] (11) According to a third aspect of the present disclosure, there is provided a computer program for creating a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set. The computer program causes a computer to execute the following processes: (a) a process of creating a first learning data group for a first ink set among a plurality of ink sets that are used alternatively in the same printing device; (b) a process of creating a second learning data group for a second ink set among the plurality of ink sets; and (c) a process of executing learning of one color prediction model using an integrated learning data group that includes the first learning data group and the second learning data group. The first ink set and the second ink set each include a plurality of common inks that are in common and one or more non-common inks that are not in common. Each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set that is composed of the non-common inks of the first ink set and the non-common inks of the second ink set.

[0094] The present disclosure may be realized in various forms other than an image processing device, a printing system, and a computer program, for example, an image processing method, a non-transitory storage medium on which a computer program is recorded, and the like. [Explanation of symbols]

[0095] 50...CPU, 51...first learning data creation unit, 52...second learning data creation unit, 53...third learning data creation unit, 55...learning unit, 60...storage unit, 61...color prediction model creation program, 62...color chart data, 63...ink set information, 70...input / output interface, 81c...graph, 81m...graph, 100...color prediction model creation device, 110...temporary learning data candidate creation unit, 120...learning data creation unit, 121...first learning data creation unit, 122...second learning data creation unit, 130...learning unit, 200...printing device, 210...ink container, 300...colorimetric device, 500...color prediction model creation system

Claims

1. A method for creating a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set, comprising the steps of: (a) creating a first training data group for a first ink set among a plurality of ink sets that are alternatively used in the same printing device; (b) creating a second training data group for a second ink set among the plurality of ink sets; (c) executing training of one color prediction model using an integrated training data group including the first training data group and the second training data group; Equipped with the first ink set and the second ink set each include a plurality of common inks that are in common with each other and one or more non-common inks that are not in common with each other; wherein each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set consisting of the non-common inks of the first ink set and the non-common inks of the second ink set.

2. 2. The method of claim 1 , A method according to claim 1, wherein the inputs of the color prediction model are ink quantities for inks in the ink sets.

3. 10. The method of claim 1 further comprising: creating a third learning data group for a third ink set including inks that are not common to either the first ink set or the second ink set; The method, wherein step (c) includes creating the integrated training data set so as to include the first training data set, the second training data set, and the third training data set.

4. 10. The method of claim 1 further comprising: creating a group of provisional training data candidates using spectral reflectances predicted by a simplified color prediction model that can be trained more easily than the color prediction model; The step (a) (a1) determining the first provisional learning data group by repeating a first learning process in which a first provisional color prediction model is learned using a first provisional learning data group selected from the provisional learning data candidate group, and the first provisional learning data group is increased so that a first prediction accuracy of the first provisional color prediction model satisfies a first target accuracy; (a2) creating the first training data group by actually measuring the spectral reflectance of the first tentative training data group; Including, The step (b) (b1) determining the second temporary learning data group by repeating a second learning process in which a second temporary color prediction model is learned using a second temporary learning data group selected from the temporary learning data candidate group, and the second temporary learning data group is increased so that a second prediction accuracy of the second temporary color prediction model satisfies a second target accuracy; (b2) creating the second training data group by actually measuring the spectral reflectance of the second provisional training data group; A method comprising:

5. 5. The method of claim 4, The steps (a1) and (b1) are carried out in series in this order; The method, wherein the step (b1) includes a step of extracting a common tentative learning data group for the plurality of common inks from the first tentative learning data group and including the common learning data group in the second tentative learning data group.

6. 5. The method of claim 4, The step (a1) and the step (b1) are carried out in parallel, The method, wherein each of the steps (a1) and (b1) includes a step of sharing a common temporary learning data group for the plurality of common inks.

7. 5. The method of claim 4, The step (a2) includes a step of printing a first color chart according to the ink amount set of the first provisional learning data group in order to actually measure the spectral reflectance of the first provisional learning data group, The method, wherein the step (b2) includes a step of printing a second color chart on a printing medium different from the first color chart according to the ink amount set of the second provisional learning data group in order to actually measure the spectral reflectance of the second provisional learning data group.

8. 8. The method of claim 7, A method, wherein a plurality of color patches reproduced using only the plurality of common inks are printed on only one of the first color chart and the second color chart.

9. 8. The method of claim 7, a plurality of color patches reproduced only with the plurality of common inks are printed on the first color chart and the second color chart, respectively; a spectral reflectance of the common training data group is set to an average value of a first spectral reflectance measured using the first color chart and a second spectral reflectance measured using the second color chart.

10. A color prediction model creation device that creates a color prediction model that predicts a spectral reflectance from an ink amount set of a plurality of types of ink that constitute an ink set, comprising: a first learning data creation unit that creates a first learning data group related to a first ink set among a plurality of ink sets that are alternatively used in the same printing device; a second learning data creation unit that creates a second learning data group related to a second ink set among the plurality of ink sets; a learning unit that executes learning of one color prediction model using an integrated learning data group including the first learning data group and the second learning data group; Equipped with the first ink set and the second ink set each include a plurality of common inks that are in common with each other and one or more non-common inks that are not in common with each other; a color prediction model creation device, wherein each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set consisting of the non-common inks of the first ink set and the non-common inks of the second ink set.

11. A computer program for creating a color prediction model for predicting a spectral reflectance from an ink amount set of a plurality of types of ink constituting an ink set, comprising: (a) creating a first training data group for a first ink set among a plurality of ink sets that are alternatively used in the same printing device; (b) creating a second training data group for a second ink set among the plurality of ink sets; (c) a process of executing learning of one color prediction model using an integrated learning data group including the first learning data group and the second learning data group; Run the following on your computer: the first ink set and the second ink set each include a plurality of common inks that are in common with each other and one or more non-common inks that are not in common with each other; a computer program, wherein each of the first learning data group and the second learning data group includes a common learning data group for the plurality of common inks, and is created so as not to include learning data for a non-common ink set constituted by the non-common inks of the first ink set and the non-common inks of the second ink set.

Images