Training data generation device, training data generation method, and program

Abstract

To provide a learning data generation apparatus, a learning data generation method, and a program that are capable of performing data augmentation specialized for individual identification of print materials.SOLUTION: A learning data generation apparatus includes: a correction scheme determination unit that retrieves a learning data group including at least one piece of learning data, the learning data being a pair of a captured image of a print material having a print source image printed thereon and identification information for the print material captured on the captured image, and that determines a correction scheme for the captured image based on the print source image and the captured image included in the learning data group; and a captured image correction unit that retrieves first learning data, and generates, as second learning data, a pair of a corrected captured image resulting from correction of the captured image included in the first learning data with the correction scheme and the identification information included in the first learning data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a training data generation device, a training data generation method, and a program. [Background technology]

[0002] Conventionally, various techniques have been proposed for identifying individuals using printed matter. For example, a technology has been disclosed that extracts features corresponding to the unevenness patterns that occur in one-dimensional barcodes or two-dimensional codes printed on artificial objects, including printed matter, and uses the extracted features to identify individuals (see Patent Document 1 below).

[0003] When individual identification of printed materials is realized by machine learning, the more training data there is, the higher the accuracy of identification can be, but the workload required to collect the training data increases. Therefore, a technology has been disclosed that reduces the workload required to collect training data by generating pseudo-new training data from the collected training data (data augmentation) (see Non-Patent Document 1 below). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 5765749 [Non-patent literature]

[0005] [Non-Patent Document 1] Connor Shorten, Taghi M. Khoshgoftaar, “A survey on Image Data Augmentation for Deep Learning,” Journal of Big Data 6, Article number 60, 6 July 2019 Summary of the Invention [Problem to be solved by the invention]

[0006] However, while data augmentation as described in Non-Patent Document 1 is useful for improving the performance of machine learning, it is a general technique and may not provide sufficient performance improvement for specific applications. This is because a general-purpose technique such as the data augmentation described in Non-Patent Document 1 does not fully consider factors characteristic of specific applications, and may generate training data that is far removed from training data that can actually be obtained for the specific application. Therefore, when machine learning is used for a specific application, it is desirable to be able to generate training data by data augmentation that is closer to training data that can actually be obtained for that application.

[0007] In view of the above-mentioned problems, an object of the present invention is to provide a training data generation device, a training data generation method, and a program that are capable of performing data expansion specialized for individual identification of printed matter. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems, a training data generation device according to one embodiment of the present invention is characterized by comprising: a correction method determination unit that acquires a training data group including at least one training data pair of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the training data group; and a captured image correction unit that acquires first training data and generates, as second training data, a pair of a corrected captured image obtained by correcting the captured image included in the first training data using the correction method and the identification information included in the first training data.

[0009] A training data generation method according to one embodiment of the present invention includes a correction method determination process in which a correction method determination unit acquires a training data group including at least one training data set that is a pair of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter that appears in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the training data group; and a captured image correction process in which a captured image correction unit acquires first training data and generates, as second training data, a pair of a corrected captured image obtained by correcting the captured image included in the first training data using the correction method and the identification information included in the first training data.

[0010] A program according to one aspect of the present invention causes a computer to function as: a correction method determination means that acquires a learning data group including at least one learning data set that is a pair of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group; and a captured image correction means that acquires first learning data and generates, as second learning data, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data using the correction method and the identification information included in the first learning data. [Effects of the Invention]

[0011] According to the present invention, data expansion can be performed specifically for identifying individual printed matter. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a block diagram illustrating an example of an overview and functional configuration of a training data generation device according to a first embodiment. [Figure 2] FIG. 3 is a diagram illustrating an example of creating learning data according to the first embodiment. [Figure 3] FIG. 2 is a diagram showing an example of the data configuration of an original print image according to the first embodiment. [Figure 4]FIG. 2 is a diagram illustrating an example of a data configuration of learning data before data extension according to the first embodiment. [Figure 5] FIG. 2 is a diagram illustrating an example of a data configuration of training data after data extension according to the first embodiment. [Figure 6] 4 is a flowchart showing an example of a processing flow according to the first embodiment. [Figure 7] 10 is a flowchart showing an example of a processing flow according to the second embodiment. [Figure 8] 10 is a flowchart showing an example of a processing flow according to the third embodiment. [Figure 9] 10 is a flowchart showing an example of a processing flow according to the fourth embodiment. [Figure 10] 13 is a flowchart showing an example of a processing flow according to the fifth embodiment. [Figure 11] FIG. 20 is a diagram showing an example of the relationship between the difference and the distribution of variations in a group of same individual difference images according to the sixth embodiment. [Figure 12] FIG. 20 is a diagram showing an example of the relationship between the difference and the distribution of variations in a group of different individual difference images according to the sixth embodiment. [Figure 13] 13 is a flowchart showing an example of a processing flow according to the sixth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0014] <<1. First Embodiment>> A first embodiment of the present invention will be described with reference to FIGS.

[0015] <1-1. Functional configuration of the learning data generation device> The functional configuration of a training data generation device 10 according to the first embodiment will be described with reference to Fig. 1 to Fig. 5. Fig. 1 is a block diagram showing an example of the functional configuration of a training data generation device 10 according to the first embodiment.

[0016] The original print image 30 shown in Fig. 1 is an image to be printed on a print medium. The original print image 30 is, for example, an image to be printed, such as a one-dimensional code or a two-dimensional code, but is not limited to such examples. The print medium is, for example, printing paper 31, but is not limited to such examples.

[0017] A printed matter 40 is a printed matter in which an original printing image 30 is printed on printing paper 31. Multiple printed matters 40 are printed for one original printing image 30. Even when the same original printing image 30 is printed on each printed matter 40, variations in the printing condition can occur due to a variety of factors. Therefore, in order to distinguish between printed matters 40 with different printing conditions, each printed matter 40 is assigned a different individual ID. The individual ID is identification information for uniquely identifying the printed matter 40. Note that factors that cause variations in the printing condition include, for example, differences in the condition of the printing surface of the printing paper 31, differences in vibration, conveying speed, humidity, etc. that occur when the printing paper 31 is transported during the printing process, deterioration of the printing plate, and differences in the position of the original on a printing plate with multiple impositions.

[0018] The imaging device 20 outputs a captured image 50 obtained by capturing an image of the printed matter 40. The number of times that one printed matter 40 is captured by the imaging device 20 is not particularly limited, and may be one or multiple times. Furthermore, one printed matter 40 may be captured by different imaging devices 20.

[0019] The learning data 60 is data that pairs a captured image 50 with the individual ID of the printed matter 40 that appears in the captured image 50. If there are multiple captured images 50 of the same printed matter 40, it is possible to create multiple pieces of learning data 60, each paired with the same individual ID and each captured image 50. If there are multiple captured images 50 of different printed matters 40, it is possible to create multiple pieces of learning data 60, each paired with the individual ID of each printed matter 40 and each captured image 50.

[0020] Here, a creation example of the learning data 60 according to the first embodiment will be described in more detail with reference to Fig. 2. Fig. 2 is a diagram showing an example of creation of the learning data 60 according to the first embodiment. Fig. 2 shows an example of creating learning data 60a to 60e based on the original print image 30a (IMG0001).

[0021] 2, original printing image 30a is printed on printing paper 31a and printing paper 31b. A printed matter 40a, in which original printing image 30a is printed on printing paper 31a, is assigned ID0101 as an individual ID. A printed matter 40b, in which original printing image 30a is printed on printing paper 31b, is assigned ID0102 as an individual ID. The printed matter 40a is photographed only once by the imaging device 20a. The imaging device 20a outputs a photographed image 50a (IMG1010). Then, learning data 60a is created by pairing the photographed image 50a with the ID0101 (individual ID) of the printed matter 40a that appears in the photographed image 50a. The printed matter 40b is photographed twice, once by the imaging device 20a and once by an imaging device 20b different from the imaging device 20a. The imaging device 20a outputs a photographed image 50b (IMG1020) and a photographed image 50c (IMG1030). The imaging device 20b outputs a photographed image 50d (IMG1040) and a photographed image 50e (IMG1050). Then, learning data 60b to 60e are created by pairing the photographed images 50b to 50e with the ID0102 (individual ID) of the printed matter 40b appearing in each photographed image.

[0022] 1, the original printing image 30 and the training data 60 are input to the training data generation device 10. The training data generation device 10 performs data extension of the training data 60 based on the input original printing image 30 and training data 60. As shown in FIG. 1, the training data generation device 10 includes a data input / output unit 110, a storage unit 120, a control unit 130, and a display unit 140.

[0023] (1) Data input / output unit 110 The data input / output unit 110 has a function of controlling input and output of data to and from an external device. For example, it accepts input of the original print image 30 and learning data 60 from the external device. The data input / output unit 110 writes and stores the accepted input of the original print image 30 and learning data 60 in the storage unit 120. The data input / output unit 110 may also output the learning data 60 (including the learning data after data extension) stored in the storage unit 120 to the external device.

[0024] (2) Storage section 120 The storage unit 120 has a function of storing various types of information. The storage unit 120 is configured by a storage medium provided as hardware in the training data generation device 10, such as a hard disk drive (HDD), a solid state drive (SSD), a flash memory, an electrically erasable programmable read-only memory (EEPROM), a random access read / write memory (RAM), a read-only memory (ROM), or any combination of these storage media. As shown in FIG. 1, the storage unit 120 includes an original print image storage unit 121 and a learning data storage unit 122.

[0025] (2-1) Original Print Image Storage Unit 121 The original print image storage unit 121 has a function of storing the original print image 30. Here, the data configuration of the original print image 30 according to the first embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the data configuration of the original print image 30 according to the first embodiment.

[0026] As shown in Fig. 3, the original print image 30 is stored in association with the individual ID of the printed matter 40 on which the original print image 30 is printed. For example, the original print image 30 is stored in association with the individual ID, such as IMG0001 and ID0101, IMG0001 and ID0102, IMG0002 and ID0201, and IMG0002 and ID0202. Note that IMG0001 and ID0101 indicate the relationship between the original print image 30a and the printed matter 40a in Fig. 2. Also, IMG0001 and ID0102 indicate the relationship between the original print image 30a and the printed matter 40b in Fig. 2.

[0027] (2-2) Learning Data Storage Unit 122 The learning data storage unit 122 has a function of storing the learning data 60 . Here, the data structure of the training data according to the first embodiment will be described with reference to Fig. 4 and Fig. 5. Fig. 4 is a diagram showing an example of the data structure of the training data before data extension according to the first embodiment. Fig. 5 is a diagram showing an example of the data structure of the training data after data extension according to the first embodiment.

[0028] As shown in FIG. 4, the training data 60 before data augmentation is stored by associating individual IDs with captured images 50. For example, the training data 60 is stored as ID0101 and IMG1010, ID0102 and IMG1020, ID0102 and IMG1030, ID0102 and IMG1040, and ID0102 and IMG1050. ID0101 and IMG1010 correspond to training data 60a in FIG. 2. ID0102 and IMG1020 correspond to training data 60b in FIG. 2. ID0102 and IMG1030 correspond to training data 60c in FIG. 2. ID0102 and IMG1040 correspond to training data 60d in FIG. 2. ID0102 and IMG1050 correspond to training data 60e in FIG. 2.

[0029] Suppose that data extension is performed on learning data 60a (a pair of ID0101 and IMG1010) among the learning data 60 before data extension shown in FIG. 4, and one new learning data 60 is added. In this case, as shown in FIG. 5, learning data of a pair of ID0101 and IMG1011 is newly registered in the learning data storage unit 122 as learning data obtained by data extension of learning data 60a. Also, suppose that data extension is performed on learning data 60e (a pair of ID0102 and IMG1050), and one new learning data 60 is added. In this case, as shown in FIG. 5, learning data of a pair of ID0102 and IMG1051 is newly registered in the learning data storage unit 122 as learning data obtained by data extension of learning data 60e.

[0030] (3) Control unit 130 The control unit 130 has a function of controlling the overall operation of the training data generation device 10. The control unit 130 is realized, for example, by causing a CPU (Central Processing Unit) provided as hardware in the training data generation device 10 to execute a program. As shown in FIG. 1, the control unit 130 includes an acquisition unit 131, a correction method determination unit 132, a captured image correction unit 133, and an output control unit 134.

[0031] (3-1) Acquisition part 131 The acquisition unit 131 has a function of acquiring various types of information. For example, the acquisition unit 131 acquires an original print image 30 from the original print image storage unit 121 of the storage unit 120 and outputs the acquired information to the correction method determination unit 132. The acquisition unit 131 also acquires a learning data group including at least one piece of learning data 60 from the learning data storage unit 122 of the storage unit 120 and outputs the learning data group to the correction method determination unit 132. The acquisition unit 131 also acquires first learning data 60 including a captured image 50 to be corrected from the learning data storage unit 122 of the storage unit 120 and outputs the first learning data 60 to the correction method determination unit 132 or the captured image correction unit 133. The acquiring unit 131 may acquire a learning data group including at least the first learning data 60 from the learning data storage unit 122 of the storage unit 120 and output the group to the correction method determining unit 132.

[0032] (3-2) Correction Method Determination Unit 132 The correction method determination unit 132 has a function of determining a correction method for the captured image 50. For example, the correction method determination unit 132 acquires learning data 60 and an original printing image 30 corresponding to the captured image 50 included in the learning data 60, and determines a correction method for the captured image 50 based on the acquired original printing image 30 and the captured image 50. Specifically, the correction method determination unit 132 acquires the learning data group acquired by the acquisition unit 131 and, from the original printing images 30 acquired by the acquisition unit 131, an original printing image 30 that corresponds to the captured image 50 included in the learning data group. Then, the correction method determination unit 132 determines a correction method for the captured image 50 based on the acquired original printing image 30 and the captured image 50 included in the learning data group.

[0033] In the first embodiment, as an example, it is assumed that the learning data group is composed of one first learning data 60. In this case, the correction method determination unit 132 detects the difference between the original printing image 30 and the captured image 50 included in the first learning data 60, and determines, as the correction method, an adjustment process for adjusting this difference.

[0034] The difference between the original print image 30 and the captured image 50 is, for example, a difference in brightness value, a difference in contour distance, a difference in feature amount, or the like. The difference in brightness value is the difference between the brightness value of a pixel in the original image 30 to be printed and the brightness value of a pixel in the captured image 50 at a position corresponding to the pixel in the original image 30 to be printed. The contour distance is the distance between the position in the photographed image 50 that corresponds to the position of the contour of the print target in the original print image 30, and the position of the contour of the print target in the photographed image 50. The difference in feature amount is the difference in feature amount detected when the original print image 30 and the captured image 50 are passed through a neural network.

[0035] The adjustment process is a process for increasing (strengthening) or decreasing (weakening) the difference. In the adjustment process for the luminance value, the luminance value is multiplied by x so that the difference between the luminance values ​​becomes larger or smaller. In the adjustment process for the distance to the contour, the position of the contour is corrected so that the distance becomes larger or smaller, and the brightness value (black and white) is inverted according to the determination of whether the contour is inside or outside. In the adjustment process for the features detected when passing the image through the neural network, the feature is multiplied by x so that the difference between the features becomes larger, and then the feature is back-propagated to the input image of the neural network so that it becomes the feature after being multiplied by x.

[0036] When determining the correction method, the correction method determination unit 132 generates a difference image that indicates the difference between the original print image 30 and the captured image 50. The correction method determination unit 132 generates an adjusted difference image by adjusting the difference based on the difference image. Then, the correction method determination unit 132 determines the specific content (adjustment amount of difference) of the adjustment process to be performed on the captured image 50 before correction so that the difference between the original print image 30 and the captured image 50 after correction by the adjustment process matches the difference indicated by the adjustment difference image.

[0037] (3-3) Captured image correction unit 133 The captured image correcting unit 133 has a function of correcting the captured image 50 to generate learning data 60. For example, the captured image correcting unit 133 corrects the captured image 50 using the correction method determined by the correction method determining unit 132. Specifically, the captured image correcting unit 133 acquires first learning data 60 and generates a corrected captured image by correcting the captured image 50 included in the first learning data 60 using the correction method. Then, the captured image correcting unit 133 generates second learning data 60 by pairing the generated corrected captured image with the individual ID included in the first learning data 60. The captured image correcting unit 133 writes the generated second learning data 60 into the learning data storage unit 122 of the storage unit 120 for storage. The captured image correction unit 133 may acquire the first learning data 60 from the learning data group used by the correction method determination unit 132, or may acquire the first learning data 60 acquired by the acquisition unit 131 from the learning data storage unit 122.

[0038] (3-4) Output control unit 134 The output control unit 134 has a function of controlling various outputs. For example, the output control unit 134 controls the output of the learning data 60. Specifically, the output control unit 134 may output the learning data 60 to an external device or display it on the display unit 140.

[0039] (4) Display section 140 The display unit 140 has a function of displaying various types of information. The display unit 140 is realized by, for example, a display device such as a display or a touch screen that is included as hardware in the training data generation device 10. The display unit 140 displays, for example, training data 60 in response to input from the output control unit 134.

[0040] <1-2. Processing flow> The functional configuration of the training data generation device 10 according to the first embodiment has been described above. Next, the flow of processing according to the first embodiment will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the flow of processing according to the first embodiment.

[0041] As shown in FIG. 6, first, the acquiring unit 131 acquires a learning data group including one piece of learning data 60 (first learning data 60) from the learning data storage unit 122 (step S101). Next, the acquisition unit 131 acquires the original printing image 30 corresponding to the captured image 50 included in the learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S102). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing image 30 associated with the same individual ID as the individual ID included in the acquired learning data 60.

[0042] Next, the correction method determination unit 132 generates a difference image (step S103). Specifically, the correction method determination unit 132 generates a difference image indicating the difference between the captured image 50 included in the learning data 60 acquired by the acquisition unit 131 and the original printing image 30 corresponding to the captured image 50.

[0043] Next, the correction method determination unit 132 generates an adjusted difference image (step S104). Specifically, the correction method determination unit 132 generates an adjusted difference image by adjusting the difference indicated by the generated difference image, based on the generated difference image.

[0044] Next, the correction method determination unit 132 determines a correction method (step S105). Specifically, the correction method determination unit 132 determines the specific content of the adjustment process (adjustment amount of the difference) to be performed on the captured image 50 before correction so that the difference between the original printing image 30 and the captured image 50 corrected by the adjustment process matches the difference indicated by the adjusted difference image.

[0045] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S106). Specifically, the captured image correcting unit 133 corrects the captured image 50 using the correction method determined by the correction method determining unit 132, and generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 corresponding to the captured image 50.

[0046] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S107). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0047] As described above, the training data generation device 10 according to the first embodiment includes the correction method determination unit 132 and the captured image correction unit 133. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60.

[0048] With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. This makes it possible to generate, by data expansion, a photographed image 50 that is closer to the photographed image 50 of the printed matter 40 on which the original print image 30 is actually printed. Then, by using the photographed image 50 generated by this data expansion as training data 60, the accuracy of individual identification of the printed matter 40 can be further improved.

[0049] Therefore, the training data generation device 10 according to the first embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0050] <<2. Second Embodiment>> Having described the first embodiment above, a second embodiment of the present invention will now be described with reference to FIG. In the first embodiment, an example in which the training data group is composed of one piece of first training data 60 has been described, but the present invention is not limited to such an example. In the second embodiment, an example will be described in which a training data group is composed of first training data 60 and third training data 60 having the same individual ID as the first training data 60. The number of third training data 60 constituting the training data group is not particularly limited as long as it is at least one. In the second embodiment, the number of third training data 60 is assumed to be one. In the following, descriptions that overlap with the description in the first embodiment will be omitted as appropriate.

[0051] <2-1. Functional configuration of the learning data generation device> The functional configuration of the training data generation device 10 according to the second embodiment is similar to the functional configuration of the training data generation device 10 according to the first embodiment described with reference to Figures 1 to 5. Hereinafter, functions of the second embodiment that differ from those of the first embodiment will be described.

[0052] The acquiring unit 131 acquires, from the training data storage unit 122, a training data group configured of the first training data 60 and the third training data 60 having the same individual ID as the first training data 60. The acquisition unit 131 acquires the original print images 30 corresponding to the captured images 50 included in the first learning data 60 and the third learning data 60 from the original print image storage unit 121.

[0053] The correction method determination unit 132 acquires the original printing image 30 and the learning data group acquired by the acquisition unit 131. The correction method determination unit 132 determines a correction method for the captured image 50 based on the acquired original printing image 30 and the captured image 50 included in the learning data group. In the second embodiment, the training data group is made up of two sets of training data 60: first training data 60 and third training data 60. In this case, the correction method determination unit 132 detects differences between the original printing image 30 and the captured images 50 included in each training data 60, and determines, as the correction method, a correction process in which part of the captured image 50 is corrected by a weighted sum with other captured images 50 based on each difference.

[0054] In the correction process for the luminance value, the luminance value is corrected by a weighted sum with the luminance values ​​of other captured images 50. In the correction process for the contour distance, the contour distance is corrected by a weighted sum with the distance of other captured images 50, and the brightness value (black and white) is corrected according to the determination of whether the contour is inside or outside after the correction. In the correction process for the features detected when passing the image through the neural network, the features are corrected by a weighted sum with the features of other captured images 50, and then the corrected features are propagated back to the input image of the neural network. The weight may be a real number between 0 and 1, or may be a binary value of 0 or 1.

[0055] The correction method determination unit 132 detects a first difference, which is the difference between the original printing image 30 and the captured image 50 included in the first learning data 60, and a second difference, which is the difference between the original printing image 30 and the captured image 50 included in the third learning data 60. The correction method determination unit 132 generates a first difference image indicating the detected first difference and a second difference image indicating the detected second difference. The correction method determination unit 132 calculates an intermediate difference as a weighted sum of the first difference indicated by the first difference image and the second difference indicated by the second difference image, and generates an intermediate difference image indicating the calculated intermediate difference. The correction method determination unit 132 determines, as the correction method, a correction process for matching the difference between the original printing image 30 and the captured image 50 included in the first learning data 60 with an intermediate difference. Specifically, the correction method determination unit 132 determines the specific content of the correction process to be performed on the captured image 50 before correction so that the difference between the original printing image 30 and the captured image 50 after correction by the correction method matches the difference indicated by the intermediate difference image.

[0056] <2-2. Processing flow> The functional configuration of the training data generation device 10 according to the second embodiment has been described above. Next, the flow of processing according to the second embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the flow of processing according to the second embodiment.

[0057] As shown in FIG. 7, first, the acquiring unit 131 acquires a learning data group including two pieces of learning data 60 (first learning data 60 and third learning data 60) having the same individual ID from the learning data storage unit 122 (step S201). Next, the acquisition unit 131 acquires the original printing image 30 corresponding to the captured image 50 included in the learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S202). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing image 30 associated with the same individual ID as the individual ID included in the acquired first learning data 60 and third learning data 60.

[0058] Next, the correction method determination unit 132 generates a difference image (step S203). Specifically, the correction method determination unit 132 generates a first difference image indicating a first difference, which is a difference between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in the first learning data 60, and a second difference image indicating a second difference, which is a difference between the original printing image 30 and the captured image 50 included in the third learning data 60.

[0059] Next, the correction method determination unit 132 generates an intermediate difference image (step S204). Specifically, the correction method determination unit 132 generates an intermediate difference image that indicates the intermediate difference between the difference images by calculating a weighted sum based on the generated first difference image and second difference image.

[0060] Next, the correction method determination unit 132 determines a correction method (step S205). Specifically, the correction method determination unit 132 determines the specific content of the correction process to be performed on the captured image 50 before correction so that the difference between the original printing image 30 and the captured image 50 after correction by the correction process matches the difference indicated by the intermediate difference image.

[0061] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S206). Specifically, the captured image correcting unit 133 corrects the captured image 50 included in the first learning data 60 using the correction method determined by the correction method determining unit 132, and generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 corresponding to the captured image 50.

[0062] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S207). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0063] As described above, the training data generation device 10 according to the second embodiment includes the correction method determination unit 132 and the captured image correction unit 133, similar to the first embodiment. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60. With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. This makes it possible to generate, by data expansion, a photographed image 50 that is closer to the photographed image 50 of the printed matter 40 on which the original print image 30 is actually printed. Then, by using the photographed image 50 generated by this data expansion as training data 60, the accuracy of individual identification of the printed matter 40 can be further improved. Therefore, the training data generation device 10 according to the second embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0064] <<3. Third Embodiment>> Having described the second embodiment above, a third embodiment of the present invention will now be described with reference to FIG. In the second embodiment, an example has been described in which the training data group is composed of two training data 60, namely, the first training data 60 and the third training data 60, but the present invention is not limited to this example. In the third embodiment, an example will be described in which the learning data group is an identical individual learning data group made up of a plurality (three or more) of learning data 60 having the same individual ID as the first learning data 60. The number of learning data 60 making up the identical individual learning data group is, for example, three or more, but is not limited to such an example. In the following, descriptions that overlap with the description in the first embodiment will be omitted as appropriate.

[0065] <3-1. Functional configuration of the learning data generation device> The functional configuration of the training data generation device 10 according to the third embodiment is similar to the functional configuration of the training data generation device 10 according to the first embodiment described with reference to Figures 1 to 5. Hereinafter, functions of the third embodiment that differ from those of the first embodiment will be described.

[0066] The acquiring unit 131 acquires, from the learning data storage unit 122, a first learning data set 60 and a same individual learning data group made up of a plurality of learning data sets 60 having the same individual ID as the first learning data set 60. The acquisition unit 131 acquires, from the original printing image storage unit 121, the first learning data 60 and the original printing images 30 corresponding to the captured images 50 included in each learning data 60 of the individual learning data group.

[0067] The correction method determination unit 132 acquires the original printing image 30 and the individual learning data group acquired by the acquisition unit 131. The correction method determination unit 132 determines a correction method for the captured image 50 based on the acquired original printing image 30 and the captured image 50 included in the individual learning data group. In the third embodiment, the correction method determination unit 132 detects the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the same individual learning data group, and based on each difference, identifies areas where differences are likely to appear, and determines an adjustment process or correction process for the identified areas as the correction method.

[0068] First, the correction method determination unit 132 generates a same-individual difference group, which is the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the same-individual learning data group. Specifically, the correction method determination unit 132 generates a same-individual difference image group (an example of a same-individual difference group) by generating a plurality of difference images indicating the difference between the original printing image 30 and the captured image 50 included in each learning data 60. For example, if the same individual learning data group includes N pieces of learning data 60, N difference images that constitute the same-individual difference image group are generated. The correction method determination unit 132 identifies an area where a difference is likely to appear based on the same-individual difference image group. The area is, for example, a pixel, a feature channel, etc. The area where a difference is likely to appear is, for example, a pixel with a large average difference in brightness value in the same-individual difference group, a pixel with a large average distance to the contour, a pixel with a large average difference in feature amount, or a channel with a large average difference in feature amount.

[0069] When adjustment processing is selected as the correction method, the correction method determination unit 132 detects the difference between the original printing image 30 and the captured image 50 included in the first learning data 60 for each pixel of the captured image 50 included in the first learning data 60, and determines that the correction method is to perform processing to emphasize the difference according to the degree of adjustment. The correction method determination unit 132 determines, for each pixel of the photographed image 50 included in the first learning data 60, an adjustment degree indicating the degree to which the pixel value of the photographed image 50 is adjusted based on the magnitude of the pixel value of the same individual difference group. For example, the correction method determination unit 132 increases the degree of adjustment as the magnitude of the pixel value of the same individual difference group increases. On the other hand, the correction method determination unit 132 decreases the degree of adjustment as the magnitude of the pixel value of the same individual difference group decreases. Furthermore, the correction method determination unit 132 increases the degree of adjustment as the variation in pixel values ​​of the same individual difference group decreases. The correction method determination unit 132 decreases the degree of adjustment as the variation in pixel values ​​of the same individual difference group increases. The correction method determination unit 132 then determines that the greater the degree of adjustment, the greater the change to be to the pixel value of the captured image 50 included in the first learning data 60. On the other hand, the smaller the degree of adjustment, the smaller the change to be to the pixel value of the captured image 50 included in the first learning data 60.

[0070] When correction processing is selected as the correction method, the correction method determination unit 132 sets one of the learning data included in the individual learning data group as third learning data 60, and detects a first difference that is the difference between the original printing image 30 and the captured image 50 included in the first learning data 60, and a second difference that is the difference between the original printing image 30 and the captured image 50 included in the third learning data 60. The correction method determination unit 132 calculates a weighted sum of the first difference and the second difference using a weighting function according to the adjustment degree to obtain an intermediate difference. The correction method determination unit 132 determines, as the correction method, correction processing that causes the difference between the original printing image 30 and the captured image 50 included in the first learning data 60 to match the intermediate difference.

[0071] <3-2. Processing flow> The functional configuration of the training data generation device 10 according to the third embodiment has been described above. Next, the flow of processing according to the third embodiment will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the flow of processing according to the third embodiment.

[0072] 8, first, the acquiring unit 131 acquires the same individual training data group (step S301). Specifically, the acquiring unit 131 acquires, from the training data storage unit 122, a training data group (including the first training data 60) including three or more pieces of training data 60 having the same individual ID. Next, the acquisition unit 131 acquires the original printing images 30 corresponding to the captured images 50 included in the learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S302). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing images 30 associated with the same individual IDs as the individual IDs included in each of the acquired learning data 60.

[0073] Next, the correction method determination unit 132 generates a same-individual difference image group (step S303). Specifically, the correction method determination unit 132 generates a plurality of difference images showing the differences between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in each learning data 60, and generates an image group consisting of the generated plurality of difference images as a same-individual difference image group.

[0074] Next, the correction method determination unit 132 identifies an area where a difference is likely to appear in the captured image 50 (step S304). For example, the correction method determination unit 132 identifies, as an area where a difference is likely to appear, a pixel with a large average difference in brightness value, a pixel with a large average distance to the contour, a pixel with a large average difference in feature amount, a channel with a large average difference in feature amount, or the like in the same individual difference group.

[0075] Next, the correction method determination unit 132 determines a correction method for the region where a difference is likely to appear (step S305). For example, the correction method determination unit 132 determines to perform adjustment processing or correction processing on the region where a difference is likely to appear.

[0076] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S306). Specifically, the captured image correcting unit 133 corrects an area of ​​the captured image 50 included in the first learning data 60 where a difference is likely to appear, using the correction method determined by the correction method determining unit 132. The captured image correcting unit 133 generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 corresponding to the captured image 50.

[0077] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S307). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0078] As described above, the training data generation device 10 according to the third embodiment includes the correction method determination unit 132 and the captured image correction unit 133, similar to the first embodiment. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60. With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. Therefore, the training data generation device 10 according to the third embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0079] <<4. Fourth Embodiment>> Having described the third embodiment above, a fourth embodiment of the present invention will now be described with reference to FIG. In the third embodiment, an example has been described in which an area where a difference is likely to appear is identified from a group of same-individual difference images, but the present invention is not limited to this example. In the fourth embodiment, an example will be described in which an area where variations in difference are likely to appear is identified from a group of same-individual difference images. In the following, descriptions that overlap with the description in the third embodiment will be omitted as appropriate.

[0080] <4-1. Functional configuration of the learning data generation device> The functional configuration of the training data generation device 10 according to the fourth embodiment is similar to the functional configuration of the training data generation device 10 according to the third embodiment described with reference to Fig. 8. Hereinafter, functions of the fourth embodiment that differ from those of the third embodiment will be described.

[0081] In the fourth embodiment, the correction method determination unit 132 detects the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the same individual learning data group, and based on each difference, identifies areas where variations in the difference are likely to appear, and determines that disturbance processing, which adds disturbance to the identified areas, is the correction method. Areas where variation in differences is likely to appear include, for example, pixels with large variance in brightness value differences in the same individual difference group, pixels with large variance in contour distances, pixels with large variance in feature differences, and channels with large variance in feature differences. The disturbance to be added to the identified region may be, for example, Gaussian noise, shading using gradation, or the like.

[0082] The correction method determination unit 132 determines, for each pixel of the captured image 50 included in the first learning data 60, a degree of disturbance indicating the degree of disturbance to be added to the pixel based on the magnitude and variation of the pixel values ​​in the same individual difference image group. For example, the correction method determination unit 132 increases the degree of disturbance as the variation in pixel values ​​of the same individual difference image group increases.The correction method determination unit 132 then determines to change the pixel values ​​of the captured image 50 included in the first learning data 60 by a larger amount as the degree of disturbance increases.On the other hand, the correction method determination unit 132 decreases the degree of disturbance as the variation in pixel values ​​of the same individual difference image group decreases.The correction method determination unit 132 then determines to change the pixel values ​​of the captured image 50 included in the first learning data 60 by a smaller amount as the degree of disturbance decreases.

[0083] The correction method determination unit 132 may obtain a disturbance image not included in the captured images 50 of the training data group, and determine the correction method based on the disturbance image. Specifically, the correction method determination unit 132 determines, for each pixel of the captured image 50 included in the first training data 60, to correct the luminance value of the captured image 50 included in the first training data 60 and the luminance value of the disturbance image by a weighted sum using a weighting function according to the degree of disturbance.

[0084] <4-2. Processing flow> The functional configuration of the training data generation device 10 according to the fourth embodiment has been described above. Next, the flow of processing according to the fourth embodiment will be described with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the flow of processing according to the fourth embodiment.

[0085] 9, first, the acquiring unit 131 acquires the same individual training data group (step S401). Specifically, the acquiring unit 131 acquires, from the training data storage unit 122, a training data group (including the first training data 60) that includes three or more pieces of training data 60 having the same individual ID. Next, the acquisition unit 131 acquires the original printing images 30 corresponding to the captured images 50 included in the learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S402). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing images 30 associated with the same individual IDs as the individual IDs included in each of the acquired learning data 60.

[0086] Next, the correction method determination unit 132 generates a same-individual difference image group (step S403). Specifically, the correction method determination unit 132 generates a plurality of difference images indicating the differences between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in each learning data 60, and generates an image group consisting of the generated plurality of difference images as a same-individual difference image group.

[0087] Next, the correction method determination unit 132 identifies an area in which variation in difference is likely to appear in the captured image 50 (step S404). For example, the correction method determination unit 132 identifies, as an area in which variation in difference is likely to appear, pixels with a large variance in brightness value differences in the same individual difference group, pixels with a large variance in contour distances, pixels with a large variance in feature amount differences, channels with a large variance in feature amount differences, etc.

[0088] Next, the correction method determination unit 132 determines a correction method for the region where the variation in the difference is likely to appear (step S405). For example, the correction method determination unit 132 determines to perform disturbance processing that adds disturbance to the region where the variation in the difference is likely to appear.

[0089] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S406). Specifically, the captured image correcting unit 133 corrects an area in which variations in the difference between the captured images 50 included in the first learning data 60 are likely to appear, using the correction method determined by the correction method determining unit 132. The captured image correcting unit 133 generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 corresponding to the corrected captured image 50.

[0090] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S407). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0091] As described above, the training data generation device 10 according to the fourth embodiment includes the correction method determination unit 132 and the captured image correction unit 133, similar to the first embodiment. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60. With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. Therefore, the training data generation device 10 according to the fourth embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0092] <<5. Fifth Embodiment>> Having described the fourth embodiment above, a fifth embodiment of the present invention will now be described with reference to FIG. In the fourth embodiment, an example has been described in which the training data group is a group of same-individual difference images, but the present invention is not limited to this example. In the fifth embodiment, an example will be described in which the training data group is a heterogeneous training data group made up of a plurality of training data 60 having individual IDs different from those of the first training data 60. In the following, explanations that overlap with those in the fourth embodiment will be omitted as appropriate.

[0093] <5-1. Functional configuration of the learning data generation device> The functional configuration of the training data generation device 10 according to the fifth embodiment is similar to the functional configuration of the training data generation device 10 according to the fourth embodiment described with reference to Fig. 9. Hereinafter, functions of the fifth embodiment that differ from those of the fourth embodiment will be described.

[0094] The acquiring unit 131 acquires, from the learning data storage unit 122, a different individual learning data group that is configured of first learning data 60 and a plurality of learning data 60 having individual IDs different from those of the first learning data 60. The acquisition unit 131 acquires, from the original printing image storage unit 121, the original printing image 30 corresponding to the captured image 50 included in the first learning data 60 and each learning data 60 of the different individual learning data group.

[0095] The correction method determination unit 132 acquires the original printing image 30 and the different individual learning data group acquired by the acquisition unit 131. The correction method determination unit 132 determines a correction method for the captured image 50 based on the acquired original printing image 30 and the captured image 50 included in the different individual learning data group.

[0096] In the fifth embodiment, the correction method determination unit 132 detects the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the heterogeneous learning data group, and based on each difference, identifies areas where variations in the difference are likely to appear, and determines that the correction method is disturbance processing that adds disturbance to the identified areas.

[0097] First, the correction method determination unit 132 generates a different-individual difference group, which is the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the different-individual learning data group. Specifically, the correction method determination unit 132 generates a different-individual difference image group (an example of a different-individual difference group) by generating a plurality of difference images indicating the difference between the original printing image 30 and the captured image 50 included in each learning data 60. For example, when the different-individual learning data group includes M pieces of learning data 60, M difference images constituting the different-individual difference image group are generated. The correction method determination unit 132 identifies an area where difference variations are likely to appear based on the different-individual difference image group. Areas where difference variations are likely to appear include, for example, pixels in the different-individual difference group where the variance of brightness value differences is large, pixels where the variance of contour distances is large, pixels where the variance of feature amount differences is large, and channels where the variance of feature amount differences is large.

[0098] The correction method determination unit 132 determines, for each pixel of the captured image 50 included in the first learning data 60, a disturbance degree indicating the degree of disturbance to be added to the pixel based on the magnitude and variation of the pixel values ​​in the different individual difference image group. For example, the correction method determination unit 132 increases the degree of disturbance as the variation in pixel values ​​of the different-individual difference image group increases.The correction method determination unit 132 then determines to change the pixel values ​​of the captured image 50 included in the first learning data 60 by a larger amount as the degree of disturbance increases.On the other hand, the correction method determination unit 132 decreases the degree of disturbance as the variation in pixel values ​​of the different-individual difference image group decreases.The correction method determination unit 132 then determines to change the pixel values ​​of the captured image 50 included in the first learning data 60 by a smaller amount as the degree of disturbance decreases.

[0099] <5-2. Processing flow> The functional configuration of the training data generation device 10 according to the fifth embodiment has been described above. Next, the flow of processing according to the fifth embodiment will be described with reference to Fig. 10. Fig. 10 is a flowchart showing an example of the flow of processing according to the fifth embodiment.

[0100] 10, first, the acquiring unit 131 acquires a different individual learning data group (step S501). Specifically, the acquiring unit 131 acquires, from the learning data storage unit 122, a learning data group (including the first learning data 60) including three or more pieces of learning data 60, all of which have different individual IDs. Next, the acquisition unit 131 acquires the original printing image 30 corresponding to the captured image 50 included in the first learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S502). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing image 30 associated with the same individual ID as the individual ID included in the acquired first learning data 60.

[0101] Next, the correction method determination unit 132 generates a different-individual difference image group (step S503). Specifically, the correction method determination unit 132 generates a plurality of difference images indicating the differences between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in each learning data 60, and generates an image group consisting of the generated plurality of difference images as a different-individual difference image group.

[0102] Next, the correction method determination unit 132 identifies an area where variation in difference is likely to appear in the captured image 50 (step S504). For example, the correction method determination unit 132 identifies, as an area where variation in difference is likely to appear, pixels with a large variance in brightness value differences in the different individual difference group, pixels with a large variance in contour distances, pixels with a large variance in feature amount differences, channels with a large variance in feature amount differences, etc.

[0103] Next, the correction method determination unit 132 determines a correction method for the region where the variation in the difference is likely to appear (step S505). For example, the correction method determination unit 132 determines to perform disturbance processing that adds disturbance to the region where the variation in the difference is likely to appear.

[0104] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S506). Specifically, the captured image correcting unit 133 corrects an area in which variations in the difference between the captured images 50 included in the first learning data 60 are likely to appear, using the correction method determined by the correction method determining unit 132. The captured image correcting unit 133 generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 corresponding to the corrected captured image 50.

[0105] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S507). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0106] As described above, the training data generation device 10 according to the fifth embodiment includes the correction method determination unit 132 and the captured image correction unit 133, similar to the first embodiment. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60. With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. Therefore, the training data generation device 10 according to the fifth embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0107] <<6. Sixth Embodiment>> Having described the fifth embodiment above, a sixth embodiment of the present invention will now be described with reference to Figs. In the fifth embodiment, an example has been described in which the learning data group is a different-individual learning data group, and an area where variation in difference is likely to appear (area where disturbance processing is performed) is identified based on a different-individual difference image group, but the present invention is not limited to such an example. In the sixth embodiment, an example will be described in which an area to perform adjustment processing and an area to perform disturbance processing are identified based on a same-individual difference image group and a different-individual difference image group, where the learning data groups are a same-individual learning data group and a different-individual learning data group. In the following, explanations that overlap with the explanations in the fifth embodiment will be omitted as appropriate.

[0108] <6-1. Functional configuration of the learning data generation device> The functional configuration of the training data generation device 10 according to the sixth embodiment is similar to the functional configuration of the training data generation device 10 according to the fifth embodiment described with reference to Fig. 10. Hereinafter, functions of the sixth embodiment that differ from those of the fifth embodiment will be described.

[0109] The acquiring unit 131 acquires, from the learning data storage unit 122, a learning data group configured as a union of a first learning data 60, a same-individual learning data group configured of a plurality of learning data 60 having the same individual ID as the first learning data 60, and a different-individual learning data group configured of a plurality of learning data 60 having an individual ID different from the first learning data 60. The acquisition unit 131 acquires, from the original printing image storage unit 121, the original printing image 30 corresponding to the captured image 50 included in the first learning data 60 and each of the learning data 60 of the same individual learning data group and the different individual learning data group.

[0110] The correction method determination unit 132 generates a same-individual difference group, which is the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the same-individual learning data group, and a different-individual difference group, which is the difference between the original printing image 30 and the captured image 50 included in each learning data 60 of the different-individual learning data group. The correction method determination unit 132 identifies, for each pixel of the captured image 50 included in the first learning data 60, an area where adjustment processing is to be performed and an area where disturbance processing is to be performed, based on the magnitude and variation of the pixel values ​​of the same-individual difference group and the magnitude and variation of the pixel values ​​of the different-individual difference group.

[0111] Here, the relationship between the difference and the distribution of variation in the same-individual difference image group and the different-individual difference image group will be described with reference to FIGS.

[0112] FIG. 11 is a diagram showing the relationship between the difference and the distribution of variation in the same individual difference image group according to the sixth embodiment. As shown in FIG. 11, an area where the difference from the original printing image 30 is large and the variation in the same individual difference image group is large is an area having the first feature. An area where the difference from the original printing image 30 is large and the variation in the same individual difference image group is small is an area having the second feature. Regardless of the magnitude of variation in the same individual difference image group, an area where the difference from the original printing image 30 is small is an area that has the third feature.

[0113] FIG. 12 is a diagram showing the relationship between the difference and the distribution of variations in a group of different individual difference images according to the sixth embodiment. As shown in FIG. 12, an area where the difference from the original printing image 30 is large and the variation in the different individual difference image group is large is an area that has the fourth characteristic. An area where the difference from the original printing image 30 is large and the variation in the different individual difference image group is small is an area that has the fifth characteristic. Regardless of the magnitude of variation in the different individual difference image group, an area where the difference from the original printing image 30 is small is an area that has the sixth feature.

[0114] From the first to sixth features shown in FIGS. 11 and 12, the features of the region can be identified as follows. The area that is the second feature or (the second feature and the fourth feature) is an area where individual-specific features are likely to appear. The areas of the third and fourth features are areas where individual characteristics are likely to appear (individual differences do not appear). The area that is the first feature or (the first feature and the fourth feature) is an area where features unrelated to the individual, such as differences in the shooting environment, are likely to appear. The area that is the fifth feature or (the second feature and the fifth feature) is an area where fixed differences occur regardless of the individual. The town area, which is the sixth characteristic or (the third characteristic and the sixth characteristic), is an area where no differences occur regardless of the individual.

[0115] In the sixth embodiment, the correction method determination unit 132 determines to perform adjustment processing on the area of ​​the second feature and the fourth feature, i.e., the area where features specific to an individual are likely to appear. Also, the correction method determination unit 132 determines to perform disturbance processing on the area of ​​the first feature and the fourth feature, i.e., the area where features unrelated to an individual, such as differences in the shooting environment, are likely to appear.

[0116] The correction method determination unit 132 determines the degree of adjustment or disturbance for the identified region for each pixel of the captured image 50 included in the first learning data 60, based on the magnitude and variation of the pixel values ​​in the same-individual difference group and the magnitude and variation of the pixel values ​​in the different-individual difference group. For example, the correction method determination unit 132 increases the degree of adjustment as the magnitude of the pixel value of the same individual difference group increases. Also, the correction method determination unit 132 increases the degree of adjustment as the variation in pixel values ​​of the same individual difference group decreases. Furthermore, the correction method determination unit 132 increases the degree of disturbance as the variation in pixel values ​​of the same-individual difference group increases. Also, the correction method determination unit 132 increases the degree of disturbance as the variation in pixel values ​​of the different-individual difference group increases.

[0117] The correction method determination unit 132 determines to change the pixel value of the captured image 50 included in the first learning data 60 by a larger amount as the adjustment degree and the disturbance degree increase. On the other hand, the correction method determination unit 132 determines to change the pixel value of the captured image 50 included in the first learning data 60 by a smaller amount as the adjustment degree and the disturbance degree decrease.

[0118] The correction method determination unit 132 determines to perform at least one of a first adjustment process in which the degree of adjustment is increased the smaller the variation in pixel values ​​of the same-individual difference group, and a second adjustment process in which the degree of adjustment is increased the larger the variation in pixel values ​​of the different-individual difference group, and then to perform the other adjustment process. Furthermore, the correction method determination unit 132 determines to execute at least one of a first disturbance processing in which the degree of disturbance increases the greater the variation in pixel values ​​of the same-individual difference group, and a second disturbance processing in which the degree of disturbance increases the greater the variation in pixel values ​​of the different-individual difference group, and then to execute the other disturbance processing.

[0119] <6-2. Processing flow> The functional configuration of the training data generation device 10 according to the sixth embodiment has been described above. Next, the flow of processing according to the sixth embodiment will be described with reference to Fig. 13. Fig. 13 is a flowchart showing an example of the flow of processing according to the sixth embodiment.

[0120] 13, first, the acquiring unit 131 acquires the same individual training data group (step S601). Specifically, the acquiring unit 131 acquires, from the training data storage unit 122, a training data group (including the first training data 60) that includes three or more pieces of training data 60 having the same individual ID. Next, the acquiring unit 131 acquires a different individual training data group (step S602). Specifically, the acquiring unit 131 acquires, from the training data storage unit 122, a training data group (including the first training data 60) including three or more pieces of training data 60, all of which have different individual IDs. Next, the acquisition unit 131 acquires the original printing image 30 corresponding to the captured image 50 included in the first learning data 60 of the acquired learning data group from the original printing image storage unit 121 (step S603). Specifically, the acquisition unit 131 acquires from the original printing image storage unit 121 the original printing image 30 associated with the same individual ID as the individual ID included in the acquired first learning data 60.

[0121] Next, the correction method determination unit 132 generates a same-individual difference image group (step S604). Specifically, the correction method determination unit 132 generates a plurality of difference images showing the differences between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in each learning data 60 of the same-individual learning data group, and generates an image group consisting of the generated plurality of difference images as a same-individual difference image group.

[0122] Next, the correction method determination unit 132 generates a different-individual difference image group (step S605). Specifically, the correction method determination unit 132 generates a plurality of difference images indicating the differences between the original printing image 30 acquired by the acquisition unit 131 and the captured image 50 included in each learning data 60 of the different-individual learning data group, and generates an image group consisting of the plurality of generated difference images as a different-individual difference image group.

[0123] Next, the correction method determination unit 132 identifies an area in the captured image 50 where adjustment processing is performed and an area where disturbance processing is performed (step S606). For example, the correction method determination unit 132 determines an area where features specific to an individual are likely to appear as an area where adjustment processing is performed. In addition, the correction method determination unit 132 determines an area where features unrelated to the individual, such as differences in the shooting environment, are likely to appear as an area where disturbance processing is performed.

[0124] Next, the correction method determination unit 132 determines the correction method for the region where the adjustment processing is performed and the region where the disturbance processing is performed (step S607). Specifically, the correction method determination unit 132 determines the adjustment degree of the adjustment processing and the disturbance degree of the disturbance processing.

[0125] Next, the captured image correcting unit 133 corrects the captured image 50 to generate learning data 60 (step S608). Specifically, the captured image correcting unit 133 corrects the area of ​​the captured image 50 that is to undergo adjustment processing and the area of ​​the captured image 50 that is to undergo disturbance processing, which are included in the first learning data 60, using the correction method determined by the correction method determining unit 132. The captured image correcting unit 133 generates learning data 60 (second learning data 60) that pairs the corrected captured image 50 with the individual ID of the printed matter 40 that corresponds to the captured image 50.

[0126] Then, the captured image correcting unit 133 registers the generated learning data 60 (step S609). Specifically, the captured image correcting unit 133 writes the generated learning data 60 into the learning data storage unit 122 to store it.

[0127] As described above, the training data generation device 10 according to the sixth embodiment includes the correction method determination unit 132 and the captured image correction unit 133, similar to the first embodiment. The correction method determination unit 132 acquires a learning data group including at least one learning data pair consisting of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter shown in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group. The captured image correction unit 133 acquires the first learning data 60, and generates, as the second learning data 60, a pair of a corrected captured image obtained by correcting the captured image included in the first learning data 60 using a correction method and identification information included in the first learning data 60. With this configuration, when generating second learning data 60 by data extension of the first learning data 60, the captured image 50 included in the first learning data 60 is corrected using a correction method that takes into account the variations specific to the printed matter 40, and the corrected captured image 50 is generated as the second learning data 60. Therefore, the training data generation device 10 according to the sixth embodiment makes it possible to perform data expansion specialized for individual identification of printed matter.

[0128] The above describes an embodiment of the present invention. Note that part or all of the training data generation device 10 in the above-described embodiment may be implemented by a computer. In this case, a program for implementing this function may be recorded on a computer-readable recording medium, and the program may be loaded into a computer system and executed. Note that the term "computer system" as used herein includes hardware such as an OS and peripheral devices. Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. Furthermore, the term "computer-readable recording medium" may also include devices that dynamically store programs for a short period of time, such as communication lines used when transmitting programs via networks such as the Internet or telephone lines, or devices that store programs for a certain period of time, such as volatile memory within a computer system that serves as a server or client. The program may be for implementing part of the above-described functions, or may be capable of implementing the above-described functions in combination with a program already stored in the computer system, or may be implemented using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0129] The embodiments of the present invention have been described in detail above with reference to the drawings, but the specific configuration is not limited to that described above, and various design changes can be made within the scope of the gist of the present invention. [Explanation of symbols]

[0130] 10... learning data generation device, 20, 20a, 20b... imaging device, 30, 30a... original print image, 31, 31a, 31b... printing paper, 40, 40a, 40b... printed matter, 50, 50a, 50b, 50c, 50d, 50e... captured image, 60, 60a, 60b, 60c, 60d, 60e... learning data, 110... data input / output unit, 120... storage unit, 121... original print image storage unit, 122... learning data storage unit, 130... control unit, 131... acquisition unit, 132... correction method determination unit, 133... captured image correction unit, 134... output control unit, 140... display unit

Claims

1. a correction method determination unit that acquires a learning data group including at least one learning data set that is a pair of a captured image of a printed matter on which an original print image is printed and identification information of the printed matter that appears in the captured image, and determines a correction method for the captured image based on the original print image and the captured image included in the learning data group; a captured image correcting unit that acquires first learning data, and generates, as second learning data, a set of a corrected captured image obtained by correcting the captured image included in the first learning data using the correction method and the identification information included in the first learning data; A training data generation device comprising:

2. the training data group is composed of the first training data, The correction method determination unit detects a difference between the original print image and the captured image included in the first learning data, and determines, as the correction method, an adjustment process for adjusting the difference.

2. The training data generating device according to claim 1 .

3. the training data group is composed of the first training data and third training data having the same identification information as the first training data, the correction method determination unit detects a first difference that is a difference between the original print image and the captured image included in the first learning data, and a second difference that is a difference between the original print image and the captured image included in the third learning data, calculating an intermediate difference as a weighted sum of the first difference and the second difference; A correction method for making the difference between the original image and the captured image included in the first learning data coincide with the intermediate difference is determined as the correction method.

2. The training data generating device according to claim 1 .

4. the training data group is a same-individual training data group configured of a plurality of training data having the same identification information as the first training data, the correction method determination unit generates a same-individual difference group that is a difference between the original printing image and the captured image included in each learning data of the same individual learning data group; determining an adjustment degree indicating a degree to which a pixel value is adjusted for each pixel of the captured image included in the first learning data, based on the magnitude and variation of the pixel value of the same individual difference group; determining that the greater the degree of adjustment, the greater the change in the pixel value of the captured image included in the first learning data; 2. The training data generating device according to claim 1 .

5. the training data group is a heterogeneous training data group composed of a plurality of training data having the identification information different from that of the first training data, the correction method determination unit generates a different individual difference group that is a difference between the original printing image and the captured image included in each learning data of the different individual learning data group; determining a disturbance degree indicating a degree of disturbance to be added to each pixel of the captured image included in the first learning data based on the magnitude and variation of the pixel value of the different-individual difference group; determining that the greater the degree of disturbance, the greater the change in the value of the pixel of the captured image included in the first learning data; 2. The training data generating device according to claim 1 .

6. the training data group is composed of a union of a same-individual training data group composed of a plurality of training data having the same identification information as the first training data, and a different-individual training data group composed of a plurality of training data having the identification information different from that of the first training data, the correction method determination unit generates a same-individual difference group which is a difference between the original printing image and the captured image included in each learning data of the same-individual learning data group, and a different-individual difference group which is a difference between the original printing image and the captured image included in each learning data of the different-individual learning data group, for each pixel of the captured image included in the first learning data, determining an adjustment degree indicating a degree to which the pixel value is adjusted and a disturbance degree indicating a degree of disturbance to be added to the pixel based on the magnitude and variation of the pixel value of the same-individual difference group and the magnitude and variation of the pixel value of the different-individual difference group; determining that the value of the pixel of the captured image included in the first learning data is to be changed to a greater extent as the degree of adjustment and the degree of disturbance increase; 2. The training data generating device according to claim 1 .

7. The correction method determination unit increases the degree of disturbance as the variation in the pixel values ​​of the same individual difference group increases.

7. The training data generating device according to claim 6.

8. the correction method determination unit increases the degree of adjustment as the magnitude of the pixel value of the same individual difference group increases; 8. The training data generating device according to claim 4, claim 6, or claim 7.

9. the correction method determination unit increases the degree of adjustment as the variation in pixel values ​​of the same individual difference group decreases; 9. The training data generating device according to claim 4, or any one of claims 6 to 8.

10. The correction method determination unit increases the degree of disturbance as the variation in the pixel values ​​of the different individual difference group increases.

8. The training data generating device according to claim 5, wherein the training data generating device generates training data based on the training data.

11. The correction method determination unit determines to execute at least one of a first adjustment process in which the degree of adjustment is increased as the variation in pixel values ​​of the same-individual difference group decreases, and a second adjustment process in which the degree of adjustment is increased as the variation in pixel values ​​of the different-individual difference group increases, and then to execute the other adjustment process.

8. The training data generating device according to claim 6 or 7.

12. The correction method determination unit determines to execute at least one of a first disturbance process that increases the degree of disturbance as the variation in pixel values ​​of the same-individual difference group increases, and a second disturbance process that increases the degree of disturbance as the variation in pixel values ​​of the different-individual difference group increases, and then to execute the other disturbance process.

12. The training data generating device according to claim 6, claim 7, or claim 11.

13. The correction method determination unit detects a difference between the original print image and the captured image included in the first learning data for each pixel of the captured image included in the first learning data, and determines, as the correction method, to perform an enhancement process of the difference in accordance with the adjustment degree.

12. The training data generating device according to claim 4, claim 6, claim 8, claim 9, or claim 11.

14. the correction method determination unit determines one of the learning data included in the same individual learning data group as third learning data, Detecting a first difference which is a difference between the original image to be printed and the captured image included in the first learning data, and a second difference which is a difference between the original image to be printed and the captured image included in the third learning data, a weighted sum of the first difference and the second difference calculated using a weighting function according to the adjustment degree to obtain an intermediate difference; A correction method for making the difference between the original image and the captured image included in the first learning data coincide with the intermediate difference is determined as the correction method.

12. The training data generating device according to claim 4, claim 6, claim 8, claim 9, or claim 11.

15. the correction method determination unit acquires a disturbance image not included in the captured image of the learning data group, The correction method is determined to be a weighted sum of the luminance value of the captured image included in the first learning data and the luminance value of the disturbance image included in the first learning data, using a weighting function according to the degree of disturbance.

13. The training data generating device according to claim 5, 6, 7, 10, or 12.

16. a correction method determination step in which a correction method determination unit acquires a learning data group including at least one learning data set that is a combination of a photographed image of a printed matter on which an original print image is printed and identification information of the printed matter that appears in the photographed image, and determines a correction method for the photographed image based on the original print image and the photographed image included in the learning data group; a captured image correcting step in which a captured image correcting unit acquires first learning data, and generates, as second learning data, a set of a corrected captured image obtained by correcting the captured image included in the first learning data using the correction method, and the identification information included in the first learning data; A training data generation method comprising:

17. Computer, a correction method determination means for acquiring a learning data group including at least one learning data set that is a combination of a photographed image of a printed matter on which an original print image is printed and identification information of the printed matter that appears in the photographed image, and determining a correction method for the photographed image based on the original print image and the photographed image included in the learning data group; a captured image correcting means for acquiring first learning data, and generating, as second learning data, a set of a corrected captured image obtained by correcting the captured image included in the first learning data using the correction method and the identification information included in the first learning data; A program to function as a

Images