Information processing device, control method for information processing device, and program

Abstract

To provide a technique for performing machine learning using original images while protecting personal information.SOLUTION: An information processing device includes: determination means for determining, in a first image, a concealment area in which personal information is concealed; concealment means for performing concealment processing on the first image based on the concealment area; display means for displaying a second image, which is a concealed image after the concealment processing; and learning means for performing learning using the first image.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] This invention relates to an information processing apparatus, a control method for the information processing apparatus, and a program. [Background technology]

[0002] In recent years, research in the field of image recognition has been flourishing, and various methods for recognizing object regions in images using convolutional neural networks (CNNs) have been proposed.

[0003] In machine learning, to obtain a machine learning model with high generalization performance, it is recommended to train the model using various image patterns, and the number and variety of training images affect accuracy. In particular, when it comes to human faces, there are many combinations of race, gender, age, facial expression, face orientation, lighting conditions, etc., and a wide variety of images are needed to improve the performance of a machine learning model that recognizes human faces.

[0004] Furthermore, awareness of personal information protection has been increasing in recent years, making anonymization necessary and making it difficult to collect training image data without consent. Personal information includes not only people's faces, but also documents containing personal names and house nameplates, and it is necessary to protect personal information from such images using image processing such as blacking out or blurring. With such personal information concealed images, users can upload images without fear of personal information leakage. For example, in online communities that share images for machine learning, this can stimulate image uploads by allowing machine learning engineers to ensure a variety of training images.

[0005] Patent Document 1 discloses a technology for generating images of individuals with their personal information concealed and for using these concealed images in machine learning. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2019-79357 [Overview of the project] [Problems that the invention aims to solve]

[0007] However, since Patent Document 1 uses images with concealed personal information for machine learning, there is a possibility that the accuracy of the machine learning model may decrease by training on images with concealed personal information. For example, if an unnatural image with a blurred portion is used for training, it may introduce unnecessary noise during training, leading to a decrease in accuracy. Furthermore, it is not possible to use images with concealed faces for training machine learning models that detect the position and size of a person's face, or machine learning models that perform facial recognition to determine whether the people in an image are the same person; the original image is required.

[0008] In such cases, it is necessary to use the original images, which have not undergone personal information anonymization, for machine learning. However, there is a challenge in that it is difficult to perform training using the original images from the standpoint of protecting personal information.

[0009] This invention has been made in view of the above-mentioned problems, and aims to provide a technology for performing machine learning using original images while protecting personal information. [Means for solving the problem]

[0010] An information processing apparatus according to one aspect of the present invention, which achieves the above objective, A determination means for determining a confidential area in which personal information is concealed from the first image, A means for performing an encryption process on the first image based on the aforementioned encrypted area, A means for assigning label information to the aforementioned confidential area, The second image is a secret image on which the aforementioned concealment process has been performed. The label information is superimposed on the above. Display means for displaying, A learning means that performs learning using the first image described above, characterized by comprising

Advantages of the Invention

[0011] According to the present invention, machine learning can be performed using the original image while protecting personal information.

Brief Description of the Drawings

[0012] [Figure 1] A diagram showing an example of the hardware configuration of an information processing apparatus according to an embodiment. [Figure 2] A diagram showing an example of the software configuration of an information processing apparatus according to an embodiment. [Figure 3] (a)-(c) Diagrams for explaining the detailed configuration of an information processing apparatus according to Embodiments 2, 3, and 4. [Figure 4] (a)-(f) Diagrams for explaining Embodiment 1. [Figure 5] (a) and (b) Flowcharts showing the procedure of image registration processing according to Embodiment 1. [Figure 6] A flowchart showing the procedure of learning processing according to an embodiment. [Figure 7] (a) and (b) Flowcharts showing the procedure of display screen generation processing according to Embodiment 1. [Figure 8] Diagram for explaining Embodiment 2. [Figure 9] (a)-(d) Flowcharts showing the procedure of display screen generation processing according to Embodiment 2. [Figure 10] Diagram for explaining Embodiment 3. [Figure 11] (a) and (b) Flowcharts showing the procedure of display screen generation processing according to Embodiment 3. [Figure 12] Diagram for explaining Embodiment 4. [Figure 13] (a) and (b) Flowcharts showing the procedure of output map confirmation processing according to Embodiment 4.

Modes for Carrying Out the Invention

[0013] The embodiments will be described in detail below with reference to the attached drawings. Note that the following embodiments do not limit the invention as defined in the claims. While the embodiments describe multiple features, not all of these features are essential to the invention, and the features may be combined in any way. Furthermore, in the attached drawings, identical or similar configurations are given the same reference numerals, and redundant descriptions are omitted.

[0014] (Embodiment 1: Label information is displayed superimposed) In this embodiment, a hidden image is displayed to the user during learning, and the original image is used for learning. For example, an example is described in which label information (face label, pupil label, mouth label, category label, gender label, orientation label, etc.) is displayed on or together with the hidden image.

[0015] <Hardware Configuration> Figure 1 shows an example of the hardware configuration of an information processing device according to one embodiment. The information processing device 200 includes a CPU (Central Processing Unit) 100, a ROM (Read Only Memory) 110, a RAM (Random Access Memory) 120, and an HDD 130. The information processing device 200 further includes an input unit 140, an information display unit 150, and a communication unit 160.

[0016] The CPU 100 is a central processing unit that performs calculations and logical decisions for various processes. The ROM 110 stores the control program. The RAM 120 is used as the main memory and temporary storage area of ​​the CPU 100, such as the work area. The HDD 130 is a hard disk for storing electronic data and programs according to this embodiment. An external storage device may be used to perform a similar role. Here, the external storage device can be realized, for example, with a media (recording medium) and an external storage drive to enable access to the media. Examples of such media include flexible disks (FD), CD-ROMs, DVDs, USB memory, MOs, and flash memory. The external storage device may also be a server device connected via a network.

[0017] The input unit 140 consists of a keyboard, touch panel, etc., and accepts input from the user. The information display unit 150 consists of a liquid crystal display, etc., and can display various data and processing results to the user. The information processing device 200 can also communicate with other devices via the communication unit 160. It may receive instructions from the user via the communication unit 160 from other devices, or it may output processing results to other devices.

[0018] <Software Configuration> Figure 2 shows an example of the software configuration of an information processing device according to one embodiment. The information processing device 200 includes a user interface unit 210 that performs input and output to the user via an input unit 140, an internal processing unit 220 that executes internal processing of the information processing device 200, and a data management unit 230 that registers and manages the input data. This is just an overview; details will be described later.

[0019] The user interface unit 210 includes an operation unit 211, a display unit 212, and a data input unit 213. The operation unit 211 receives mouse and keyboard input from the user and selects images to be registered in the data management unit 230 and images to be learned by the learning unit 223. The display unit 212 displays images to be registered in the data management unit 230 and images to be learned by the learning unit 223, allowing the user to visually confirm the data. The data input unit 213 receives input data such as captured image data. For example, images obtained from a camera or surveillance camera.

[0020] The internal processing unit 220 includes a confidentiality area determination unit 221, an image conversion unit 222, a learning unit 223, and a labeling unit 224.

[0021] The confidentiality area determination unit 221 determines the area in the image data input from the data input unit 213 to conceal personal information. The confidentiality area may be determined by the user via the operation unit 211. The image conversion unit 222 performs image conversion based on the confidentiality area determined by the confidentiality area determination unit 221 to conceal personal information. The learning unit 223 performs machine learning using the training data held by the data management unit 230. In this embodiment, a human face detector is used as an example of the target of machine learning. The detector may be a convolutional neural network or a ViT (Vision Transformer). Alternatively, it may be a Support Vector Machine (SVM) combined with a feature extractor, and various models can be used. This embodiment is not limited to the above format, but in this embodiment, the face detector will be described as a CNN. The labeling unit 224 assigns labels to the data input from the data input unit 213. The labeling may be determined by the user via the operation unit 211.

[0022] The data management unit 230 comprises a management unit 231 and management data 232. The management unit 231 registers data input from the data input unit 213 and data processed by the internal processing unit 220. The management data 232 represents the data set held by the data management unit 230, and includes, for example, learning data input from the data input unit 213, confidential area information determined by the confidential area determination unit 221, and learning models generated by the learning unit 223.

[0023] <Process Overview> In this embodiment, we will describe an example in which user A registers their own image with the data management unit 230, and then user B uses the image registered by user A to perform machine learning.

[0024] Figures 4(a)-4(f) show an example image in this embodiment, which includes a person and a dog. Figures 5(a)-5(b) are flowcharts illustrating the image registration process according to this embodiment. Figure 6 is a flowchart illustrating the learning process according to this embodiment. Figures 7(a)-7(b) are flowcharts illustrating the screen generation process displayed when checking the learned image according to this embodiment. Hereinafter, each process in the flowchart is realized by the CPU 100 executing a control program.

[0025] <Image registration process> First, the method for registering an image by user A will be explained with reference to the flowcharts in Figures 5(a) and 5(b). In S501, the data input unit 213 receives an image input from user A. For example, user A inputs the image 400 shown in Figure 4(a). In S502, the display unit 212 displays the image 400 to user A.

[0026] In S503, the confidential area determination unit 221 determines a confidential area for image 400 based on the mouse operation of user A. Assume that the confidential area 411 shown in image 410 of Figure 4(b) is determined by user A's operation. In this embodiment, an example is described in which user A performs an operation to specify an area for image 400 displayed on the display unit 212, but the embodiment is not limited to this. For example, the confidential area determination unit 221 may automatically determine the confidential area using a trained model of the data management unit 230. When setting the face area of ​​a person as the confidential area, as in Figure 4(b), a face detector can be used as the trained model.

[0027] In S504, the image conversion unit 222 performs personal information concealment processing. Here, Figure 5(b) is a flowchart showing the detailed procedure for personal information concealment processing. In S510, the image conversion unit 222 determines the method for concealing personal information. In this embodiment, the concealment method is determined to be a blackout process as specified by user A, but is not limited to this. For example, any image conversion process that can conceal personal information, such as blurring or conversion to a similar image, may be used.

[0028] In S511, the image conversion unit 222 performs image conversion of image 400 using the concealment method determined in S510. For example, as shown in Figure 4(c), a personal information concealment image 420 is generated, and it can be seen that the concealment area 421 is filled in. In this embodiment, only the concealment area is converted, but any range of area determined based on the concealment area may be converted. For example, an area that is a predetermined size of the concealment area may be filled in.

[0029] In S512, the display unit 212 displays the generated personal information concealment image 420. Then, in S513, the display unit 212 notifies user A to confirm whether personal information has been concealed by the personal information concealment image 420 (for example, by displaying a Yes / No button). If user A determines that personal information has been concealed, the personal information concealment process ends, for example, when user A presses the Yes button. On the other hand, if user A determines that personal information has not been concealed, the process returns to S510, for example, when user A presses the No button, and the concealment method is determined again. In this embodiment, the personal information concealment process ends, assuming that personal information has been concealed by the personal information concealment image 420. Then, the process proceeds to S505.

[0030] In S505, the labeling unit 224 performs labeling of the confidential area. In this embodiment, labeling is performed automatically, but is not limited to this; user A may manually label the confidential area using the operation unit 211, or label areas other than the confidential area may be labeled. Figures 4(d) and 4(e) show the applied label information. The labeled image 430 shown in Figure 4(d) shows the applied area labels, with face label 431, pupil labels 432, 433, and mouth label 434 applied. The label information 440 shown in Figure 4(e) shows the classification label of the confidential area, with face as the category label, female as the gender label, and 0 degrees as the orientation label. Here, a label is also applied to the confidential area 411, and a person's face label, which is the same type of label as face label 431, is applied.

[0031] In S506, the display unit 212 displays a confirmation screen. For example, the information shown in Figures 4(d) and 4(e) is displayed. Then, in S507, the display unit 212 asks user A to confirm whether the assigned label information is correct. Based on the user's response, if the label information is correct, the process proceeds to S508. On the other hand, if the label information is incorrect, the process returns to S505 and assigns the label again. In this embodiment, it is assumed that all the label information is correct, and the process proceeds to S508.

[0032] In S508, the management unit 231 saves the image 400, confidentiality area information indicating the confidentiality method and coordinates and size of the confidentiality area, personal information confidentiality image 420, label information, etc., as management data 232. With this, the image registration process by user A is completed.

[0033] Next, referring to the flowchart in Figure 6, we will explain how User B can learn using the original image while concealing personal information during the machine learning process.

[0034] In S601, the learning unit 223 selects a training dataset based on the operation of the operation unit 211 by user B. Here, the training dataset is an image dataset that is previously held as management data 232. The training dataset includes the image 400 and label information shown in Figure 4(a).

[0035] In S602, the display unit 212 displays the training images in order to allow the user to confirm the images to be used for training before executing machine learning. Here, Figure 7(a) shows a detailed flowchart of the image display when user B selects image 400 as the image to be displayed.

[0036] In S701, the learning unit 223 accepts an operation by user B on the operation unit 211 and selects the image 400 selected by user B as the display image. In S702, the learning unit 223 obtains information about image 400 from the management unit 231. In S703, the learning unit 223 checks whether or not an area for concealing personal information is assigned to image 400. If this step is Yes, the process proceeds to S704. On the other hand, if this step is No, the process proceeds to S707. In this embodiment, since an area for concealing personal information is assigned to image 400, the process proceeds to S704.

[0037] In S704, the display unit 212 displays the personal information concealment image 420 obtained from the management unit 231 on the display screen. In this embodiment, an example is shown in which the personal information concealment image 420 is obtained from the management unit 231, but this is not the only example. For example, the personal information concealment image 420 may not be stored in the management unit 231, but may be generated using the image 400 and the concealment area information.

[0038] In S705, the learning unit 223 determines whether label information has been assigned to image 400. If this step is Yes, the process proceeds to S706. On the other hand, if this step is No, the process ends. In this embodiment, since label information has been assigned, the process proceeds to S706. In S706, the learning unit 223 adds a label list to the display screen. Details of the process in S706 will be described later with reference to Figure 7(b). In S707, the display unit 212 displays the original image 400 because there is no area for concealing personal information. The above is the series of processes shown in Figure 7(a).

[0039] Next, we will explain the detailed process of S706 by referring to the flowchart in Figure 7(b).

[0040] In S710, the learning unit 223 selects one of the assigned label information based on the operation of the user B's operation unit 211. In this embodiment, the face label 431 is selected first.

[0041] In S711, the learning unit 223 determines whether the selected label information is a region label. If this step is Yes, the learning unit 223 proceeds to S712. On the other hand, if this step is No, the process proceeds to S716. The face label 431 selected in S710 is a label indicating a region in the image and contains coordinate information, so the process proceeds to S712.

[0042] In S712, the learning unit 223 superimposes the labels onto the personal information concealment image 420 using coordinate information. In S713, the learning unit 223 adds the label information to the region label list. Here, the region label list is a list of label information superimposed on the personal information concealment image 420, and is the information displayed on the screen when the display screen is updated in S715. A detailed explanation will follow later. Here, the label information for face label 431 is added to the region label list.

[0043] In S714, the learning unit 223 determines whether all labels assigned to image 400 have been processed. If this step is Yes, the process proceeds to S715. On the other hand, if this step is No, the process returns to S710. At this point, there are still unprocessed labels, so the process returns to S710 and continues.

[0044] In S710, the learning unit 223 assumes that a category label has been selected based on the operation of user B's operation unit 211. In that case, in the determination process of S711, the category label is a classification label assigned to the hidden area, and the selected label does not contain coordinate information, so the process moves to S716.

[0045] In S716, the learning unit 223 adds category label information to the classification label list. Here, the classification label list is a list of label information assigned to the confidential area, and is the information displayed on the screen when the display screen is updated in S715. A detailed explanation will be given later. After that, the process proceeds to S714, where it is determined whether all labels assigned to image 400 have been processed. Here, since there is still unprocessed label information remaining, the process returns to S710 and the same process continues.

[0046] By processing the labels assigned in the manner described above, the region labels are superimposed on the personal information concealment image 420 and added to the region label list, and the classification labels are added to the classification label list. After all labels have been processed, the process moves on to S715.

[0047] In S715, the display unit 212 updates the display screen. Here, Figure 4(f) shows an example of the display screen 450 that is displayed to user B after all processing is completed in this embodiment. Region labels are superimposed on the display image 451, and the region label list 452 contains information indicating the region labels. In the illustrated example, the dashed line indicates the face region, the solid line indicates the pupil region, and the dotted line indicates the mouth region. This allows for visual confirmation of which label is assigned to which location, even in the case of an image with personal information concealed, where the original image cannot be viewed for the protection of personal information. In addition, the classification label list 453 contains classification label information for the concealed region. In the illustrated example, it can be recognized that the category is face, the gender is female, and the orientation is 0 degrees (front). This allows for informing user B, who cannot view the original image, what kind of image it is. When the processing in S715 is completed, the processing in S706 is completed, and the processing in S602 is also completed.

[0048] Subsequently, in S603 of Figure 6, the learning unit 223 determines the training data based on the operation of the operation unit 211 by user B. In S604, the learning unit 223 executes machine learning. This completes the process shown in Figure 6.

[0049] As described above, in this embodiment, the management unit 231 manages the image data, the image 400 is used for machine learning, and the image 420 with concealed personal information is used when displaying it to user B who is learning.

[0050] This allows machine learning to be performed using the original image while concealing personal information. Furthermore, by superimposing labels onto the image with concealed personal information, user B, who does not know the original image, can visually recognize the information in the original image, which helps in deciding whether or not to use the image with concealed personal information for training.

[0051] (Embodiment 2: Display of similar images) In this embodiment, an example is described in which a hidden image is displayed to the user undergoing training, and the original image is used for training. More specifically, an example is described in which a similar image (i.e., a modified hidden image) similar to the hidden region is displayed.

[0052] In this embodiment, we will explain an example in which user A registers image 400, shown in Figure 4(a), which is an image of a person and a dog, in the data management unit 230, similar to Embodiment 1, and display an image similar to the confidential area 411 shown in Figure 4(b) on the display screen.

[0053] The hardware configuration of the information processing device 200 according to this embodiment is the same as that shown in Figure 1 of Embodiment 1, and the software configuration is also the same as that of Embodiment 1 for the user interface unit 210 and the data management unit 230, but the configuration of the internal processing unit is different.

[0054] Figure 3(a) shows an example of the internal processing unit 300 according to this embodiment. In this embodiment, in addition to the components shown in Embodiment 1, the internal processing unit 300 further includes a similar image search unit 301 for displaying similar images. The similar image search unit 301 includes an image similarity calculation unit 302, a label similarity calculation unit 303, and a similar image generation unit 304.

[0055] In this embodiment, we will describe an example in which, after the processing shown in Figures 7(a) and 7(b) described in Embodiment 1 has been completed and the display screen 450 shown in Figure 4(f) is displayed, user B confirms the images to be used for learning by displaying similar images.

[0056] Figure 8 shows an example of a display screen when similar images are displayed according to this embodiment. Display screen 800 is an example of a display screen after processing according to this embodiment has been performed. The search method selection box 801 allows you to select whether or not to display similar images, and if similar images are to be displayed, to select a similar image search method. In the illustrated example, the options listed are "Do not display similar images", "Display images that are similar", "Display images with similar labels", and "Generate similar images". First, we will explain the processing when "Display images that are similar" is selected in the search method selection box 801.

[0057] <Displaying similar images with high image similarity> Figures 9(a) and 9(b) are flowcharts showing the procedure for searching for similar images according to this embodiment.

[0058] In S901, the similar image search unit 301 determines the search method based on the input from user B. Here, user B has selected "Show similar images" in the search method selection box 801. Therefore, it is decided to use the image similarity calculation unit 302 to search for images with high image similarity (above a predetermined value).

[0059] Here, image similarity is calculated by quantifying the distribution of brightness and color within an image, the positional information of objects, edge information, etc., and expressing the similarity between images. In this embodiment, image feature extraction is performed using a trained model held by the data management unit 230, and the similarity is compared, but this is not the only method. For example, the images may be compared directly.

[0060] In S902, the similar image search unit 301 determines whether or not to display similar images. If this step is Yes, the process proceeds to S903. On the other hand, if this step is No, the process proceeds to S905. For example, if "Do not display similar images" is selected in the search method selection box 801, the process proceeds to S905 and similar images are not displayed. Here, assuming that "Display images that are similar" is selected, the process proceeds to S903.

[0061] In S903, the similar image search unit 301 searches for similar images. Here, the flowchart in Figure 9(b) shows the detailed processing of S903. In S911, the similar image search unit 301 acquires the original image. Here, it acquires the image 400 shown in Figure 4(a).

[0062] In S912, the similar image search unit 301 retrieves an image from the management data 232 held by the data management unit 230 as a candidate image to be displayed as a similar image, which has a person's face label assigned to it and does not contain a personal information concealment area. Here, as in Embodiment 1, a person's face label is assigned to the concealment area 411 (Figure 4(b)). Therefore, it can be seen that an image with a person's face label is suitable as an image to be displayed as a similar image. Also, in S912, in order to retrieve a candidate image to be displayed as a similar image, an image that does not contain a personal information concealment area and can be displayed to user B is retrieved.

[0063] In S913, the image similarity calculation unit 302 calculates the image similarity. For example, it calculates the image similarity between the concealed area 411 shown in Figure 4(b) and the face areas of each person in the image group acquired in S912. In this embodiment, the image similarity is calculated for the concealed area, but it is not limited to this. For example, the image similarity of the entire image or the image similarity of an area determined based on the concealed area may be used; any image similarity of any area may be used.

[0064] In S914, the similar image search unit 301 acquires images with a high similarity (above a predetermined value) calculated by the image similarity calculation unit 302. This completes the process shown in Figure 9(b). The process then proceeds to S904.

[0065] In S904, the display unit 212 displays a similar image. For example, an image with a face area similar to the concealed area of ​​image 400 shown in Figure 4(a) is displayed as a similar image in the similar image display unit 802 (Figure 8). In S905, the display unit 212 does not display a similar image. This completes the series of processes shown in Figure 9(a).

[0066] <Displaying similar images with high label similarity> Next, we will explain the process when "Show items with similar labels" is selected in the search method selection box 801 in Figure 8.

[0067] In S901, if user B has selected "Show images with similar labels" in the search method selection box 801 in Figure 8, it is decided to use the label similarity calculation unit 303 to search for images with high label similarity.

[0068] Here, label similarity is a numerical representation of how closely the label information within each image matches the images themselves; the more matching the labels, the higher the similarity. In this embodiment, the similarity is calculated such that a higher number of matching labels results in a higher similarity, but this is not the only method. For example, the similarity may be calculated by weighting each type of label.

[0069] In S902, it is determined whether or not to display similar images. Here, "Display images with similar labels" is selected, so the process moves to S903. Now, referring to the flowchart in Figure 9(c), the detailed procedure for S903 when "Display images with similar labels" is selected in the search method selection box 801 in Figure 8 will be explained.

[0070] In S921, the label similarity calculation unit 303 obtains the label of the original image. Here, it obtains the label of image 400 in Figure 4(a), which is the original image. In S922, the label similarity calculation unit 303 obtains candidate images to be displayed as similar images from the management data 232 held by the data management unit 230. Here, it obtains images that have a person's face label assigned to them and that do not contain an area for concealing personal information.

[0071] In S923, the label similarity calculation unit 303 calculates the label similarity. Here, the label similarity between the concealed area 411 shown in Figure 4(b) and the face areas of each person in the image group acquired in S922 is calculated. Here, the label similarity within the concealed area is calculated, but this is not the only way. For example, the label similarity of any area may be used, such as the label similarity of the entire image or the label similarity of an area determined based on the concealed area.

[0072] In S924, the similar image search unit 301 acquires images with high label similarity (above a predetermined value) based on the label similarity calculated by the label similarity calculation unit 303. This completes the process shown in Figure 9(c). The process then proceeds to S904.

[0073] In S904, the display unit 212 displays similar images. For example, an image with a face area of ​​a person that is similar to the concealed area and label of image 400 shown in Figure 4(a) is displayed as a similar image in the similar image display unit 802 (Figure 8).

[0074] <Generation and display processing of similar images> Finally, we will explain the process when "Generate similar images" is selected in the search method selection box 801.

[0075] In S901, if user B selects "Generate similar images" in the search method selection box 801 in Figure 8, it is decided to generate images with a high degree of similarity using the similar image generation unit 304. In this embodiment, the generation of similar images refers to the generation of similar images by, for example, a generative artificial intelligence.

[0076] In S902, it is determined whether or not to display similar images. Since "Generate similar images" is selected here, the process proceeds to S903. Now, referring to the flowchart in Figure 9(d), the detailed procedure for S903 when "Generate similar images" is selected in the search method selection box 801 in Figure 8 will be explained.

[0077] In S931, the similar image generation unit 304 converts the confidential region into a similar image. Here, the confidential region 411 shown in Figure 4(b) is converted into a similar image. Note that this is an example of converting a confidential region into a similar image, but it is not limited to this. For example, the conversion to a similar image may be performed on any region, such as the entire image region or a region determined based on the confidential region.

[0078] In S932, the image similarity calculation unit 302 determines whether the similarity between the generated similar image and the original image is greater than or equal to a first threshold and less than or equal to a second threshold. Focusing on the original purpose of concealing personal information, the lower the similarity between the generated similar image and the original image, the better the personal information is concealed. However, if dissimilar images are displayed as similar images, user B may mistakenly perceive the original image as a different image. Therefore, in this embodiment, the first threshold is set to 0.2 and the second threshold to 0.8, and the similarity between the generated similar image and the original image is determined by determining whether the similar image is not too similar. If the similarity is too low (similarity < first threshold) or too high (similarity > second threshold), the process in S931 is repeated. If an image with a similarity of greater than or equal to the first threshold and less than or equal to the second threshold is generated, the process proceeds to S933.

[0079] In S933, the similar image search unit 301 acquires the image generated by the similar image generation unit 304. This completes the process shown in Figure 9(d). The process then proceeds to S904.

[0080] In S904, the display unit 212 displays similar images. For example, the generated image is displayed as a similar image in the similar image display unit 802 (Figure 8).

[0081] As described above, in this embodiment, it is possible to display an image similar to the confidential area, providing visual support to user B who does not know the original image. It can also support the decision of whether or not to use the personal information confidential image for learning. Therefore, user convenience is improved.

[0082] (Embodiment 3: Display of Recommended Learning Use) This embodiment describes an example in which a hidden image is displayed to the user during learning, while the original image is used for learning. Furthermore, an example of displaying learning usage information and a learning usage recommendation score is also described.

[0083] In this embodiment, we will explain an example in which user A registers image 400, shown in Figure 4(a), which is an image of a person and a dog, in the data management unit 230, similar to Embodiment 1, and display the learning usage information and learning usage recommendation score of image 400 on the display screen.

[0084] Here, the training usage information is information associated with the original image and may include at least one of the following: the accuracy improvement rate for each task when other users used the image for training, the number of times it was used for training, and the number of high ratings. The number of high ratings is the number of users who used the target image for training and evaluated that using that image improved their training. The training usage recommendation score is a score calculated according to equation (1), and a higher score indicates that the image is recommended for training.

[0085]

number

[0086] Here, S is the recommended learning utilization score for a person's face, and R f is the average value of the accuracy improvement rate of the face detection rate of a person when other users use Image 400 for learning. Also, U f is the number of times other users have used Image 400 for learning face detection of a person, and G f is the number of high evaluations of Image 400. Also, W Rf , W Uf , W Gf represent the respective weight coefficients, and the weight coefficients can be determined for each image and each learning task. In this embodiment, W Rf = 0.02, W Uf = 0.00005, W Gf = 0.0002 will be used for the explanation.

[0087] The hardware configuration example of the information processing apparatus 200 according to this embodiment is the same as the configuration shown in FIG. 1 of Embodiment 1, and the software configuration is also the same as that of Embodiment 1 for the user interface unit 210 and the data management unit 230, but the configuration of the internal processing unit is different.

[0088] FIG. 3(b) shows an example of the internal processing unit 310 according to this embodiment. In this embodiment, the internal processing unit 310 further includes a recommendation degree calculation unit 311 for calculating the recommended learning utilization degree in addition to the components shown in Embodiment 1.

[0089] In this embodiment, from the state where the processes of FIGS. 7(a) and 7(b) described in Embodiment 1 are completed and the display screen 450 shown in FIG. 4(f) is displayed, an example in which User B displays the learning utilization information and the recommended learning utilization degree and checks the quality of the image used for learning will be described.

[0090] Figure 10 shows an example of a display screen when displaying learning usage information and learning usage recommendation level according to this embodiment. Display screen 1000 is an example of a display screen after performing the processing according to this embodiment. The learning usage information display selection box 1001 allows the user to choose whether or not to display learning usage information. In this embodiment, the processing when "Display learning usage information" is selected in the learning usage information display selection box 1001 will be described.

[0091] <Display processing of learning usage information> Figures 11(a) and 11(b) are flowcharts showing the procedure for displaying learning usage information and learning usage recommendation scores according to this embodiment. In S1101, the internal processing unit 310 obtains information indicating whether or not to display learning usage information based on input from user B. In S1102, the internal processing unit 310 determines whether or not to display learning usage information based on the information obtained in S1101. If this step is Yes, proceed to S1103. On the other hand, if this step is No, proceed to S1106. Specifically, if "Do not display learning usage information" is selected in the learning usage information display selection box 1001, the process moves to S1106 and the learning usage information is not displayed. Here, "Display learning usage information" is selected, so the process moves to S1103.

[0092] In S1103, the internal processing unit 310 determines whether or not a label has been assigned to the image (for example, image 400). If this step is Yes, the process proceeds to S1104. On the other hand, if this step is No, the process proceeds to S1106. Here, since a label has been assigned to image 400, the process proceeds to S1104.

[0093] In S1104, the recommendation calculation unit 311 calculates the learning usage recommendation score. Here, Figure 11(b) is a flowchart detailing the processing in S1104. In S1111, the recommendation calculation unit 311 selects one of the assigned labels. In this embodiment, the face label 431 is selected first. In S1112, the recommendation calculation unit 311 calculates the learning usage recommendation score. The recommendation calculation unit 311 calculates the learning usage recommendation score for the face label using, for example, the learning usage information 1002 in Figure 10 (accuracy improvement rate = 10, number of learning uses = 2000, number of high ratings = 500) and formula (1). Here, 10 × 0.02 + 2000 × 0.00005 + 500 × 0.0002 = 0.4. In S1113, the recommendation calculation unit 311 determines whether all labels assigned to the image (for example, image 400) have been processed. If this step is Yes, the process ends. On the other hand, if this step is No, the process returns to S1111. Here, since there are still unprocessed labels, the process returns to S1111 and continues.

[0094] As described above, by processing the assigned labels in order, it is possible to calculate the learning usage recommendation score for all labels. After processing all labels is complete, proceed to process S1105.

[0095] In S1105, the display unit 212 displays the learning usage information and the learning usage recommendation score. In Figure 10, the learning usage information 1002 shows, as an example, the learning usage information (accuracy improvement rate, number of learning uses, number of high ratings) and the learning usage recommendation score for face labels, pupil labels, and gender labels. For example, Figure 10 displays the score including 0.4 calculated for face labels. In S1106, the display unit 212 does not display the learning usage information or the learning usage recommendation score. This completes the process shown in Figure 11(a).

[0096] As described above, this embodiment makes it possible to display learning usage information and learning usage recommendation scores, thereby supporting user B who does not know the original image. It also supports the decision of whether or not to use images with concealed personal information for learning. Therefore, user convenience is improved.

[0097] (Embodiment 4: Display of learning status) According to embodiments 1 to 3 described above, it is possible to perform machine learning using the original image while concealing personal information, but it is difficult to check the learning progress of machine learning while concealing personal information when using the original image for machine learning. In contrast, this embodiment describes a method that makes it possible to conceal personal information even when checking the learning progress of machine learning using the original image.

[0098] The hardware configuration of the information processing device 200 according to this embodiment is the same as the configuration shown in Figure 1 of Embodiment 1, and the software configuration is also the same as in Embodiment 1 for the data management unit 230, but the configuration of the internal processing unit and the user interface unit is different.

[0099] Figure 3(c) shows an example of the internal processing unit 320 and user interface unit 330 according to this embodiment. In addition to the components of Embodiment 1, the internal processing unit 320 further includes a similar image search unit 301 and an output map conversion unit 321.

[0100] The similar image search unit 301 includes an image similarity calculation unit 302. The output map conversion unit 321 includes a model output map conversion unit 322 and a model output map conversion parameter determination unit 323. In addition, the user interface unit 330 further includes a model output map display unit 331 in addition to the components of Embodiment 1.

[0101] In this embodiment, after the determination of the training data in S603 in Figure 6 is completed, an example is described in which, when machine learning is performed in S604, the output map obtained during training is displayed while concealing personal information.

[0102] Figure 12 shows an image illustrating the method for checking the learning status according to this embodiment. Image 1200 shows a learning image used in this embodiment. It is an image of a person and a dog, with the person wearing a name tag with their name on it around their neck. Labeled image 1210 shows the assigned region labels, each region label being a face label 1211, pupil labels 1212 and 1213, mouth label 1214, and name tag label 1215. Personal information concealed image 1220 is an image of image 1200 registered by user A that has undergone personal information concealment processing. In the regions where the face label 1211 and name tag label 1215 are assigned, the personal information is concealed by filling in the image as concealed region 1221 and concealed region 1222.

[0103] <Checking learning progress> Figures 13(a) and 13(b) are flowcharts showing the procedure for checking the learning status according to this embodiment. In S1301, the internal processing unit 320 selects a learning image. For example, image 1200 is selected as the learning image. In S1302, the learning unit 223 performs machine learning. In this embodiment, the tasks to be learned are a human face detection task and a human pupil detection task. Therefore, machine learning is performed using face label 1211 and pupil labels 1212 and 1213.

[0104] In S1303, the learning unit 223 acquires the output map after the learning is complete. At this time, the output map is not yet displayed to user B. The output maps are, for example, the pupil output map 1230, the face output map 1240, and the model intermediate output map 1250. Here, the pupil output map and the face output map are maps output from the final layer of the model and are output at the same size as the training image. The model intermediate output map is a map output from the intermediate layer of the model and is output at the same size as the training image.

[0105] This embodiment describes an example in which these three maps are output, but it is not limited to this. For example, other maps such as a mouth map, a head map, or a category classification map may be output.

[0106] The pupil output map 1230 shows a response to the positions of both pupils, while the face output map 1240 shows a response to the center of the face. By displaying such output maps, it is possible to check whether learning is progressing correctly for each task being studied.

[0107] To more accurately verify the accuracy of the learning process, one method is to overlay the output map onto the image used for training and examine in detail where the map is responding on the image. However, since image 1200 is an image with personal information concealed, user B cannot overlay the output map onto image 1200 to check the map's response. Note that the size of the output map may differ from that of the training image if the model includes elements such as pooling layers or upsampling layers. In this case, it is best to resize either the map or the image so that their sizes are equal before overlaying them.

[0108] Furthermore, the intermediate output map 1250 of the model shows that many edge features from image 1200 remain. If such an output map is displayed to user B, user B will know what the original image was like, which could lead to the leakage of personal information.

[0109] From these points, it can be seen that the following problems occur when trying to check the progress of learning with images that conceal personal information: (1) The original image cannot be used as the image on which the output map is superimposed. (2) There is a possibility that personal information remains in the output map. Therefore, in this embodiment, support is provided so that user B can accurately know the response of the output map even when learning with images that conceal personal information.

[0110] In S1304, the learning unit 223 selects one of the output maps. In this embodiment, the pupil output map 1230 is selected first. In S1305, the learning unit 223 determines whether or not the image 1200 contains an area for concealing personal information among the selected output maps. If this step is Yes, the process proceeds to S1306. On the other hand, if this step is No, the process proceeds to S1310. In this embodiment, since the image 1200 has an area for concealing personal information, the process proceeds to S1306.

[0111] In S1306, the model output map conversion unit 322 and the image similarity calculation unit 302 perform image conversion on the output map. The reason for performing image conversion on the output map is that there is a possibility that personal information remains on the output map. A detailed flow of S1306 is shown in Figure 13(b).

[0112] In S1321, the image similarity calculation unit 302 calculates the image similarity between the source image and the output map. For example, it calculates the image similarity between the source image, image 1200, and the output map, pupil output map 1230. In this embodiment, the image similarity between image 1200 and pupil output map 1230 is calculated for each pupil region. Here, the pupil region refers to the region in the labeled image 1210 to which pupil labels 1212 and 1213 are assigned. Furthermore, the region used in calculating the similarity is not limited to the region to which labels are assigned; any region may be used, such as a concealed region or a region determined based on the concealed region. In this embodiment, the similarities of the regions to which pupil labels 1212 and 1213 are assigned are assumed to be 0.1 and 0.15, respectively.

[0113] In S1322, the model output map conversion parameter determination unit 323 obtains image conversion parameters corresponding to the similarity. In this embodiment, if the image similarity is 0.5 or higher, a fill process is performed, and if the image similarity is less than 0.5, no image conversion is performed. Furthermore, the method for determining the image conversion parameters is not limited to this, and blurring or other image conversions may be performed, or the intensity of the image conversion may be changed according to the similarity.

[0114] In S1323, the output map conversion unit 321 performs image conversion of the output map. Here, assume that the similarity of the regions to which pupil labels 1212 and 1213 are assigned is 0.1 and 0.15, respectively. In this case, since the similarity of both is less than 0.5, no image conversion is performed.

[0115] Subsequently, in S1324, the image similarity calculation unit 302 calculates the image similarity between the original image and the output map processed in S1323. Here, the image similarity between image 1200 and the pupil output map processed in S1323 is calculated. The image similarity calculation unit 302 then determines whether the calculated image similarity is below a threshold. If this step is Yes, the process ends. On the other hand, if this step is No, the process returns to S1322 to determine the image conversion parameters again. Here, since the calculated image similarity is low and below the threshold, the process ends and the process returns to the process in Figure 13(a).

[0116] In S1307 of Figure 13(a), the output map conversion unit 321 superimposes the image-converted output map onto the personal information concealment image (for example, the personal information concealment image 1220 in Figure 12).

[0117] In S1308, the output map conversion unit 321 superimposes labels onto the image on which the output map is superimposed. Here, it is assumed that the display image 1260 in Figure 12 is generated by superimposing pupil labels 1212 and 1213 onto the image generated in S1307. The pupil output map 1230 and pupil labels 1212 and 1213 are superimposed on the personal information concealment image 1220. In this embodiment, an example of superimposing label information corresponding to the output map being processed has been described, but this is not the only example. Any label information may be assigned, and if there is no label information, it does not need to be assigned.

[0118] In S1309, the output map conversion unit 321 determines whether all output maps have been processed. If this step is Yes, the process proceeds to S1311. On the other hand, if this step is No, the process returns to S1304 and continues.

[0119] In S1310, the learning unit 223 superimposes the output map onto the original image. In S1311, the model output map display unit 331 updates the display screen using the generated display image and displays the generated display image to user B. This completes the series of processes shown in Figure 13(a).

[0120] In this case, since there are still unprocessed output maps in S1309, the process returns to S1304 and continues. In S1304, the face output map 1240 is selected. Then, the process proceeds to S1305 and S1306, and the processing shown in Figure 13(b) is performed. Similar to the pupil output map, the similarity between image 1200 and face output map 1240 is low. As a result, the display image 1270 in Figure 12 is generated. Looking at the display image 1270, it can be seen that the face output map 1240 and face label 1211 are superimposed on the personal information concealed image 1220. Then, since there are still unprocessed output maps, the process returns to S1304 and continues.

[0121] In S1304, the model intermediate output map 1250 is selected. Then, the process proceeds to S1305 and S1306, and the processing shown in Figure 13(b) is performed. In S1321, the image similarity of each hidden region between image 1200 and model intermediate output map 1250 is calculated. Here, the image similarity calculation unit 302 can calculate multiple similarities, such as edge similarity and color similarity. Any one of the multiple similarities, or the result with the highest value, may be used as the image similarity. The similarity between image 1200 and model intermediate output map 1250 is low in terms of image color information, but high in terms of image edge information, and the calculated similarity is 0.8.

[0122] In S1322, image transformation parameters corresponding to the similarity are obtained. Here, since the image similarity (0.8) between image 1200 and the model intermediate output map 1250 is greater than or equal to the threshold (0.5), the process of filling in the concealed area is decided.

[0123] In S1323, the output map is filled in. The areas that are filled in are the areas on the output map that correspond to the areas on the personal information concealment image 1220 to which the concealment areas 1221 and 1222 are assigned. The subsequent processing is the same as in the explanation of the pupil output map, so the explanation is omitted. Here, we assume that the display image 1280 in Figure 12 has been generated. Looking at the display image 1280, we can see that the model intermediate output map 1250 is superimposed on the personal information concealment image 1220, and the concealment areas have been filled in.

[0124] Subsequently, in S1309, since the processing of all output maps is complete, the process proceeds to S1311, where the model output map display unit 331 displays the generated display image to user B.

[0125] As explained above, in this embodiment, personal information can be kept confidential even when checking the learning progress of machine learning, and user B, who does not know the original image, can be visually supported.

[0126] In this embodiment, it has been explained that all learned tasks, such as pupils and faces, are tasks that can be displayed on the display screen, but this is not limited to this. For example, the system may decide whether or not to display the output map on the display screen depending on the label information attached to the image. For example, the system may be configured to display the output map for pupil and face labels, but not for mouth labels. Furthermore, the tasks that can be displayed may be set for each image.

[0127] [Differentiation] Although the embodiments described above used people as examples, other categories such as dogs or landscapes may also be used. When applying the embodiments described above to other categories, it is possible to learn using the original images while concealing personal information.

[0128] Furthermore, although the above-described embodiment has been explained using the example of a single information processing device as shown in Figure 2, it is not limited to this. For example, user A and user B may each possess an operating terminal equipped with a user interface unit 210 and an internal processing unit 220, and perform operations using each terminal. Also, the data management unit 230 may be a server device connected via a network, and the internal processing unit 220 may also be part of an information processing device managed on the network.

[0129] The disclosures herein include the following information processing devices, methods for controlling the information processing devices, and programs.

[0130] (Item 1) A determination means for determining a confidential area in which personal information is concealed from the first image, A means for performing an encryption process on the first image based on the aforementioned encrypted area, A display means for displaying a second image, which is a confidential image on which the aforementioned confidentiality process has been performed, A learning means that performs learning using the first image described above, An information processing device characterized by comprising:

[0131] (Item 2) The system further comprises a means for assigning label information to the confidential area, The information processing apparatus according to item 1, characterized in that the display means displays the second image and the label information.

[0132] (Item 3) The information processing apparatus according to item 2, characterized in that the display means displays the label information superimposed on the second image.

[0133] (Item 4) The information processing apparatus according to item 3, characterized in that the label information includes at least one region label from among the face label, pupil label, and mouth label of a person included in the first image.

[0134] (Item 5) The information processing device according to item 2, characterized in that the label information includes at least one classification label from among the category label, gender label, and orientation label of the person included in the first image.

[0135] (Item 6) The information processing device according to any one of items 1 to 5, further comprising management means for managing a plurality of images including the first image and a plurality of confidential images including the second image.

[0136] (Item 7) The system further comprises an image similarity calculation means for calculating the image similarity between the confidential area of ​​the first image and the confidential area of ​​another image managed by the management means, The information processing apparatus according to item 6, characterized in that the display means acquires and displays an image whose image similarity is equal to or greater than a threshold from among the other images managed by the management means.

[0137] (Item 8) The system further comprises a label similarity calculation means for calculating the label similarity between the label information assigned to the confidential area of ​​the first image and the label information assigned to other images managed by the management means, The information processing apparatus according to item 6 or 7, characterized in that the display means acquires and displays an image from among the other images managed by the management means in which the label similarity is equal to or greater than a threshold.

[0138] (Item 9) The system further comprises a similar image generation means for converting the confidential area into a similar image, The information processing apparatus according to any one of items 1 to 8, characterized in that the display means displays the second image in which the confidential area has been converted into the similar image.

[0139] (Item 10) The system further comprises a calculation means for calculating the image similarity between the first image and the second image in which the concealed region has been converted into a similar image, The information processing apparatus according to item 9, characterized in that the display means displays the second image in which the concealed area has been converted into the similar image when the image similarity is greater than or equal to a first threshold and less than or equal to a second threshold greater than the first threshold.

[0140] (Item 11) The first image mentioned above is associated with learning usage information. The system further comprises a score calculation means for calculating a learning usage recommendation score based on the learning usage information of the first image, The information processing device according to any one of items 1 to 10, characterized in that the display means further displays the learning usage information and the learning usage recommendation score.

[0141] (Item 12) The information processing device according to item 11, characterized in that the learning utilization information includes at least one of the accuracy improvement rate when the first image was used for learning, the number of times the first image was used for learning, and the number of high ratings for the first image.

[0142] (Item 13) The information processing device according to any one of items 1 to 12, further comprising a map display means for displaying a model output map obtained as a result of inputting the first image into the model and performing training.

[0143] (Item 14) The system further comprises a conversion means for converting the model output map based on the confidential region of the first image, The information processing apparatus according to item 13, characterized in that the map display means displays the model output map converted by the conversion means.

[0144] (Item 15) A calculation means for calculating the similarity between the first image and the model output map, A parameter determination means that determines the conversion parameters used by the conversion means so that the similarity is below a threshold, The information processing device according to item 14, further comprising:

[0145] (Item 16) The information processing device according to item 15, characterized in that the map display means determines whether or not to display the model output map according to the label information assigned to the first image.

[0146] (Item 17) The information processing device according to any one of items 1 to 16, characterized in that the display means displays the second image to the user when the user performs learning using the first image.

[0147] (Item 18) A method for controlling an information processing device, A decision process for determining a confidential area to conceal personal information in the first image, A concealment step in which concealment processing is performed on the first image based on the aforementioned concealed area, A display step of displaying a second image, which is a confidential image on which the aforementioned confidentiality process has been performed, A learning process that performs learning using the first image described above, A control method for an information processing device, characterized by having the following features.

[0148] (Item 19) A program to cause a computer to function as an information processing device as described in any one of items 1 through 17.

[0149] (Other embodiments) The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0150] The invention is not limited to the embodiments described above, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, claims are attached to disclose the scope of the invention. [Explanation of Symbols]

[0151] 200: Information processing unit, 210: User interface unit, 211: Operation unit, 212: Display unit, 213: Data input unit, 220: Internal processing unit, 221: Confidential area determination unit, 222: Image conversion unit, 223: Learning unit, 224: Labeling unit, 230: Data management unit, 231: Management unit, 232: Management data

Claims

1. A determination means for determining a confidentiality area in which personal information is concealed from the first image, A means for performing a concealment process on the first image based on the concealment area, A means for assigning label information to the aforementioned confidential area, A display means that superimposes and displays the label information onto a second image, which is a concealed image on which the concealment process has been performed. A learning means that performs learning using the first image described above, An information processing device characterized by comprising:

2. The information processing apparatus according to claim 1, characterized in that the label information includes at least one region label from among the face label, pupil label, and mouth label of a person included in the first image.

3. The information processing apparatus according to claim 1, characterized in that the label information includes at least one classification label from among the category label, gender label, and orientation label of the person included in the first image.

4. The information processing apparatus according to claim 1, further comprising management means for managing a plurality of images including the first image and a plurality of confidential images including the second image.

5. The system further comprises an image similarity calculation means for calculating the image similarity between the confidential area of ​​the first image and the confidential area of ​​another image managed by the management means, The information processing apparatus according to claim 4, characterized in that the display means acquires and displays an image whose image similarity is equal to or greater than a threshold from among the other images managed by the management means.

6. The system further comprises a label similarity calculation means for calculating the label similarity between the label information assigned to the confidential area of ​​the first image and the label information assigned to other images managed by the management means, The information processing apparatus according to claim 4, characterized in that the display means acquires and displays an image from among the other images managed by the management means in which the label similarity is equal to or greater than a threshold.

7. The system further comprises a similar image generation means for converting the confidential area into a similar image, The information processing apparatus according to claim 1, characterized in that the display means displays the second image in which the confidential area has been converted into the similar image.

8. The system further comprises a calculation means for calculating the image similarity between the first image and the second image in which the concealed region has been converted into a similar image, The information processing apparatus according to claim 7, characterized in that the display means displays the second image in which the concealed area has been converted into the similar image when the image similarity is greater than or equal to a first threshold and less than or equal to a second threshold greater than the first threshold.

9. The first image mentioned above is associated with learning usage information. The system further comprises a score calculation means for calculating a learning usage recommendation score based on the learning usage information of the first image, The information processing apparatus according to claim 1, wherein the display means further displays the learning usage information and the learning usage recommendation score.

10. The information processing apparatus according to claim 9, characterized in that the learning utilization information includes at least one of the accuracy improvement rate when the first image was used for learning, the number of times the first image was used for learning, and the number of high ratings for the first image.

11. The information processing apparatus according to claim 1, further comprising a map display means for displaying a model output map obtained as a result of inputting the first image into the model and performing training.

12. The system further comprises a conversion means for converting the model output map into an image based on the confidential region of the first image, The information processing apparatus according to claim 11, characterized in that the map display means displays the model output map converted by the conversion means.

13. A calculation means for calculating the similarity between the first image and the model output map, A parameter determination means that determines the conversion parameters used by the conversion means so that the similarity is below a threshold, The information processing apparatus according to claim 12, further comprising:

14. The information processing apparatus according to claim 13, wherein the map display means determines whether or not to display the model output map according to the label information assigned to the first image.

15. The information processing apparatus according to claim 1, characterized in that the display means displays the second image to the user when the user performs learning using the first image.

16. A method for controlling an information processing device, A decision-making process for determining a confidentiality area to conceal personal information in the first image, A concealment step in which concealment processing is performed on the first image based on the aforementioned concealment area, A step of assigning label information to the confidential area, A display step of superimposing the label information onto a second image, which is a concealed image on which the concealment process has been performed, A learning process that performs learning using the first image described above, A control method for an information processing device, characterized by having the following features.

17. A program for causing a computer to function as an information processing device according to any one of claims 1 to 15.

Images