Learning living body data generation device and method for generating learning living body data

Abstract

To more efficiently generate learning data of a learning model which is used for biometric authentication.SOLUTION: A learning fingerprint data living body device comprises: a first generation part for generating pseudo living body data of a person who does not exist really, on the basis of living body data including biological information; a second generation part for generating pseudo shape data of the living body part of the person who does not really exist, on the basis of shape data indicating a three-dimensional shape of the living body part corresponding to the biological data; and an output part for generating pseudo living body part data of the person who does not exist really, on the basis of the pseudo living body data and the pseudo shape data, and generating and outputting learning living body data used for machine learning of the learning model, on the basis of the generated pseudo living body part data.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a learning biometric data generation device and a learning biometric data generation method. [Background technology]

[0002] Patent document 1 describes an identification system characterized by having a learning processing unit that uses machine learning to learn the relationship between categories and areas specified in areas of a training fingerprint image to generate a learning model, and an area classification unit that classifies areas of a target fingerprint image using the learning model generated by the learning processing unit. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-165911 A Summary of the Invention [Problem to be solved by the invention]

[0004] Generating a learning model for fingerprint authentication requires learning using a large amount of fingerprint data, but the large amount of fingerprint data used for learning is generated based on the fingerprints of many real people, so there is a problem that collecting fingerprint data takes time and money. Therefore, in Patent Document 1, a new learning fingerprint image is generated by duplicating an input learning fingerprint image and processing the copied learning fingerprint image. However, in Patent Document 1, since the fingerprint image of an actual person is acquired as the learning fingerprint image and then copied and processed, the large amount of collected fingerprint data must be handled with great care from the viewpoint of protecting the privacy of the person who provides the fingerprint.

[0005] The present disclosure has been devised in consideration of the above-described conventional circumstances, and aims to provide a training biometric data generation device and a training biometric data generation method that more efficiently generate training data for a learning model used in biometric authentication. [Means for solving the problem]

[0006] The present disclosure provides a training biometric data generation device including: a first generation unit that generates pseudo biometric data of a non-existent person based on biometric data including biometric information; a second generation unit that generates pseudo shape data of a biometric part of the non-existent person based on shape data indicating a three-dimensional shape of the biometric part corresponding to the biometric data; and an output unit that generates pseudo biometric part data of the non-existent person based on the pseudo biometric data and the pseudo shape data, and generates and outputs training biometric data to be used in machine learning of a learning model based on the generated pseudo biometric part data.

[0007] The present disclosure also provides a method for generating training biometric data performed by a terminal device, which generates pseudo-biometric data of a non-existent person based on biometric data including biometric information, generates pseudo-shape data of a biometric part of the non-existent person based on shape data indicating a three-dimensional shape of the biometric part corresponding to the biometric data, generates pseudo-biometric part data of the non-existent person based on the pseudo-biometric data and the pseudo-shape data, and generates and outputs training biometric data to be used in machine learning of a learning model based on the generated pseudo-biometric part data. Effect of the Invention

[0008] According to the present disclosure, learning data for a learning model used in fingerprint authentication can be generated more efficiently. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing an example of the internal configuration of a training fingerprint data generating device according to an embodiment; [Diagram 2] 1 is a flowchart showing an example of an operation procedure of a training fingerprint data generating device according to an embodiment. [Diagram 3] FIG. 1 is a diagram for explaining an example of a pseudo-fingerprint pattern generation process and a pseudo-finger 3D model generation process; [Figure 4] FIG. 1 is a diagram for explaining an example of a process for generating a 3D model of a finger with a pseudo fingerprint. [Diagram 5] FIG. 1 is a diagram for explaining an example of a pseudo-contact type fingerprint generation process and a pseudo-contactless type fingerprint generation process; [Figure 6] FIG. 1 shows an example of a pseudo fingerprint image data set. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, with reference to the drawings as appropriate, each embodiment that specifically discloses the configuration and operation of the learning biometric data generation device and learning biometric data generation method according to the present disclosure will be described in detail. However, more detailed explanation than necessary may be omitted. For example, detailed explanation of already well-known matters and duplicate explanation of substantially the same configuration may be omitted. This is to avoid the following explanation becoming unnecessarily redundant and to facilitate understanding by those skilled in the art. Note that the attached drawings and the following description are provided to enable those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter described in the claims.

[0011] The training biometric data generating device and training biometric data generating method according to the present disclosure described below will be described as an example of a technique for generating training data used in training biometric authentication using a fingerprint, but are not limited thereto. For example, the training biometric data generating device and training biometric data generating method according to the present disclosure may be applied to generating training data used in training other biometric authentication using other biometric information (e.g., palm print, etc.).

[0012] A training fingerprint data generating device P1 according to an embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the internal configuration of a training fingerprint data generating device P1 according to an embodiment.

[0013] The training fingerprint data generation device P1 (an example of a training biometric data generation device and a terminal device) accepts input of a fingerprint pattern image showing the fingerprint pattern (fingerprint) of a real or non-existent person, and generates a fingerprint pattern image of the non-existent person (hereinafter referred to as a "pseudo fingerprint pattern image") based on the input fingerprint pattern image. The training fingerprint data generation device P1 also accepts input of a 3D model of a finger (fingertip) of a real or non-existent person, and generates a 3D model of the finger of the non-existent person (hereinafter referred to as a "pseudo 3D finger model") based on the input 3D model of the finger. The training fingerprint data generation device P1 generates fingerprint data of a non-existent person who is different from the person who provided the fingerprint pattern image or the 3D finger model, based on the generated pseudo fingerprint pattern image and the 3D model of the pseudo finger.

[0014] The training fingerprint data generation device P1 is realized by a terminal device such as a personal computer (hereinafter, referred to as “PC”), a notebook PC, a tablet terminal, a smartphone, etc. The training fingerprint data generation device P1 includes an input unit 10, a control unit 11, a memory 12, and an output unit 13.

[0015] The input unit 10 (an example of an acquisition unit) is realized by a communication interface that enables data transmission and reception between, for example, a camera, a sensor, a storage medium, a PC, a notebook PC, a tablet terminal, a smartphone, etc. The input unit 10 accepts input of a fingerprint pattern image and a 3D model of a finger of an actual or non-actual person. The input unit 10 outputs the input fingerprint pattern image or 3D model of a finger of the actual person to the control unit 11.

[0016] The input unit 10 may be realized by, for example, a camera, a sensor, etc., and may be capable of acquiring a fingerprint pattern image and a 3D model of a finger of an actual person. The input unit 10 outputs the acquired fingerprint pattern image or 3D model of a finger of the actual person to the control unit 11.

[0017] Here, the fingerprint pattern image in this embodiment may be an image that shows at least a part of the entire fingerprint pattern. Also, the 3D model of the finger in this embodiment is a 3D model of the area where the fingerprint exists, which includes at least the first joint of the fingertip.

[0018] Furthermore, the fingerprint pattern image and the 3D model of the finger may be any of the index finger, middle finger, ring finger, or little finger. The fingerprint pattern image and the 3D model of the finger may be the fingerprint pattern image and the 3D model of the finger of the same person, or may be the fingerprint pattern image and the 3D model of the finger of different people. Furthermore, the 3D model of the finger is not limited to the 3D model (three-dimensional shape) of the finger of an actual person, but may be the 3D model (three-dimensional shape) of the finger of a non-actual person (i.e., a fictitious person).

[0019] The control unit 11 is configured using, for example, a central processing unit (CPU), a field programmable gate array (FPGA), or a graphics processing unit (GPU), and performs various processes and controls in cooperation with the memory 12. Specifically, the control unit 11 refers to the programs and data stored in the memory 12, and executes the programs to realize the functions of each unit of the control unit 11. Note that the respective units referred to here are a learning unit 111, an image generating unit 112, a simulation unit 113, and a labeling unit 114.

[0020] The learning unit 111 uses a fingerprint pattern image of a person to generate (learn) a pseudo fingerprint generation model for generating a fingerprint pattern image of a non-existent person, and uses a 3D model of a person's finger to generate (learn) a pseudo finger generation model for generating a 3D model of a non-existent person's finger.

[0021] The learning unit 111 is not essential and may be omitted. In such a case, the pseudo-fingerprint generation model and the pseudo-fingerprint generation model each acquire a trained pseudo-fingerprint generation model and a trained pseudo-fingerprint generation model from an external terminal via the input unit 10.

[0022] The pseudo-fingerprint generation model and the pseudo-finger generation model referred to here are a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), a Stable Diffusion, or the like. Note that the pseudo-fingerprint generation model is not limited to the above-mentioned examples, and may be any model capable of generating a pseudo-fingerprint pattern image of a non-existent person. Note that the pseudo-finger generation model is not limited to the above-mentioned examples, and may be any model capable of generating a 3D model of a pseudo-finger of a non-existent person.

[0023] The image generation unit 112 (an example of the first generation unit and the second generation unit) executes generation of a pseudo-fingerprint pattern image using the pseudo-fingerprint generation model, and generation of a 3D model of a pseudo finger using the pseudo-fingerprint generation model. The image generation unit 112 generates a 3D model of a finger with a pseudo-fingerprint by texture mapping the pseudo-fingerprint pattern image onto the 3D model of the pseudo finger. The image generation unit 112 outputs the generated 3D model of the finger with a pseudo-fingerprint to the simulation unit 113.

[0024] The simulation unit 113 (an example of an output unit) acquires each of the 3D models of the finger with a pseudo-fingerprint output from the image generation unit 112. The simulation unit 113 executes a simulation for acquiring (capturing or sensing) a pseudo fingerprint possessed by the 3D model of the finger with a pseudo-fingerprint under each fingerprint acquisition condition for each of the 3D models of the finger with a pseudo-fingerprint. The simulation unit 113 generates pseudo-fingerprint image data acquired under the fingerprint acquisition condition as a result of each simulation, and outputs the data to the labeling unit 114.

[0025] The fingerprint acquisition conditions referred to here are so-called fingerprint imaging conditions or fingerprint sensing conditions. The fingerprint acquisition conditions may be, for example, the pressure applied to the fingertip (fingerprint) when acquired in contact with the fingerprint imaging surface or the sensor surface that senses the fingerprint, the angle and attitude (yaw angle, roll angle, and pitch angle) of the fingertip (fingerprint) contacting the imaging surface or sensor surface, the position (coordinates) of the fingertip relative to the image sensor or sensor surface of the camera, the illuminance or brightness of illuminating the fingertip, the contact range of the fingertip contacting the contact surface, etc. Also, the fingerprint acquisition conditions may be, for example, the angle and attitude (yaw angle, roll angle, and pitch angle) of the fingertip (fingerprint) when acquired in a non-contact state, the position (coordinates, imaging distance) of the fingertip relative to the image sensor or angle of view of the camera, the illuminance or brightness of illuminating the fingertip, the imaging parameters of the camera that images the fingertip (for example, the focal length of the lens, the imaging distance, the exposure time, etc.), etc.

[0026] The labeling unit 114 (an example of an output unit) acquires the pseudo-fingerprint image data output from the simulation unit 113, and classifies each of the acquired pseudo-fingerprint image data into 3D models of fingers with pseudo-fingerprints before simulation (i.e., into each finger). The labeling unit 114 assigns a label indicating the name of a non-existent person or the like to each of the classified pseudo-fingerprint image data. The labeling unit 114 outputs each of the pseudo-fingerprint image data to which the same label has been assigned as one pseudo-fingerprint image data set to the pseudo-fingerprint image data set DB 13 in the memory 12, and stores the same therein. The labeling unit 114 may output the generated pseudo-fingerprint image data set to the output unit 13.

[0027] The memory 12 has, for example, a random access memory (hereinafter, referred to as "RAM") as a work memory used when executing each process of the control unit 11, and a read only memory (hereinafter, referred to as "ROM") for storing programs and data that define the operation of the control unit 11. The RAM temporarily stores data or information generated or acquired by the control unit 11. The ROM has written therein a program that defines the operation of the control unit 11.

[0028] The memory 12 also stores a fingerprint pattern data set DB11, a finger 3D model data set DB12, and a pseudo fingerprint image data set DB13. Each of the fingerprint pattern data set DB11, the finger 3D model data set DB12, and the pseudo fingerprint image data set DB13 may be realized by a storage medium or a storage medium device separate from the training fingerprint data generation device P1.

[0029] The fingerprint pattern dataset DB11 is a storage medium device such as a Hard Disk Drive (hereinafter, abbreviated as "HDD") or a Solid State Drive (hereinafter, abbreviated as "SSD") The fingerprint pattern dataset DB11 stores fingerprint pattern images for each person.

[0030] The finger 3D model dataset DB12 is a storage medium device such as an HDD or an SSD. The finger 3D model dataset DB12 stores a 3D model of a fingertip for each person.

[0031] The pseudo fingerprint image dataset DB13 is a storage medium device such as an HDD or SSD. The pseudo fingerprint image dataset DB13 stores the pseudo fingerprint image dataset after labeling for each label.

[0032] The output unit 13 is configured using, for example, a Liquid Crystal Display (LCD) or an organic electroluminescence (EL). The output unit 13 outputs and displays a screen (not shown) including the generated pseudo-fingerprint image data. The output unit 13 may also be realized as a storage medium. In such a case, the output unit 13 may store the pseudo-fingerprint image data output from the labeling unit 114.

[0033] Next, an example of the operation procedure of the training fingerprint data generation device P1 will be described with reference to Fig. 2 to Fig. 5. Fig. 2 is a flowchart showing an example of the operation procedure of the training fingerprint data generation device P1 according to an embodiment. Fig. 3 is a diagram explaining an example of a pseudo-fingerprint pattern generation process and a pseudo-fingerprint 3D model generation process. Fig. 4 is a diagram explaining an example of a pseudo-fingerprint-bearing finger 3D model generation process. Fig. 5 is a diagram explaining an example of a pseudo-contact-type fingerprint generation process and a pseudo-contactless fingerprint generation process.

[0034] Note that Figure 2 describes an example in which a fingerprint pattern image is obtained from a real person and a 3D model of a finger is obtained from a non-existent person, but the fingerprint pattern image and the 3D model of the finger may be a 3D model of a non-existent person's finger.

[0035] The input unit 10 accepts input of a fingerprint pattern image FP1 (an example of biometric data, see FIG. 3) of an actual person (St11A). The input unit 10 also accepts input of a 3D model FM1 (an example of shape data, see FIG. 3) of a finger of a fictitious person (St11B).

[0036] The learning unit 111 causes an image generation engine based on a deep learning network to learn (generate) a pseudo-fingerprint generation model using the input fingerprint pattern image FP1 (St12A). Also, the learning unit 111 causes a 3D model generation engine based on a deep learning network to learn (generate) a pseudo-finger generation model using the input 3D model FM1 of a finger (St12B).

[0037] Image generation unit 112 generates pseudo fingerprint pattern images FP11, FP12, FP13, ... (an example of pseudo biometric data) of at least one non-existent person from fingerprint pattern image FP1 using the pseudo fingerprint generation model generated by learning unit 111 (St13A). The pseudo fingerprint pattern images FP11, ... generated here have characteristics (individuality) that are determined to be different persons from the person corresponding to fingerprint pattern image FP1 in fingerprint authentication processing. In other words, image generation unit 112 can generate pseudo fingerprint pattern images FP11, ... of multiple non-existent persons from one fingerprint pattern image FP1.

[0038] The image generating unit 112 generates 3D models FM11, FM12, FM13, ... (an example of pseudo shape data) of pseudo fingers of at least one person who is a non-existent person from the 3D finger model FM1 using the pseudo finger generation model generated by the learning unit 111 (St13B). The pseudo finger 3D models FM11, ... generated here have characteristics (individuality) that are determined to be persons different from the person corresponding to the 3D finger model FM1 in the fingerprint authentication process. In other words, the image generating unit 112 can generate 3D models FM11, ... of pseudo fingers of multiple non-existent persons from the 3D finger model FM1 of one person.

[0039] The image generating unit 112 executes a process of texture mapping each of the multiple pseudo fingerprint pattern images FP11, ... onto each of the generated pseudo finger 3D models FM11, ... (St14), and generates 3D models MD11, ... of fingers with pseudo fingerprints (an example of pseudo body part data, see FIG. 4) in which each pseudo fingerprint pattern image FP11, ... is texture mapped onto the pseudo finger 3D models FM11, ... (St15). In other words, the 3D models MD11, ... of fingers with pseudo fingerprints are data obtained by texture mapping the fingerprint of a non-existent person onto the 3D model of the non-existent person's finger.

[0040] For example, the image generating unit 112 texture maps each of the pseudo fingerprint pattern images FP11, ... onto the pseudo finger 3D model FM11, texture maps each of the pseudo fingerprint pattern images FP11, ... onto the pseudo finger 3D model FM12, and texture maps each of the pseudo fingerprint pattern images FP11, ... onto the pseudo finger 3D model FM13. As a result, the number of 3D models of fingers with pseudo fingerprints generated by the image generating unit 112 is a number obtained by multiplying the number of 3D models of pseudo fingers generated by the number of pseudo fingerprint pattern images generated. As a result, the learning fingerprint data generating device P1 can obtain a large number of 3D models of fingers with pseudo fingerprints (an example of biometric data) by using fingerprint pattern images and finger 3D models collected from real people. The number of 3D models of fingers with pseudo fingerprints generated may be one.

[0041] The simulation unit 113 acquires each of the 3D models MD11, ... of fingers with pseudo fingerprints generated by the image generation unit 112. When acquiring a fingerprint by bringing a fingertip into contact with a contact surface (e.g., an imaging surface or a sensor surface, etc.), the simulation unit 113 executes a simulation (pressing simulation) of pressing a fingertip (fingerprint) against a predetermined contact surface with a plurality of different pressures for each of the 3D models MD11, ... of fingers with pseudo fingerprints (St16A).

[0042] The simulation unit 113 generates pseudo-contact fingerprint images IMG1, . . . (an example of learning biometric data, 2D data) based on the 3D models MD11, . . . of fingers with pseudo fingerprints on which the pressing simulations have been performed (St17A).

[0043] As a result, the number of pseudo-contact type fingerprint images generated by simulation unit 113 is the product of the number of 3D models of fingers with pseudo fingerprints multiplied by the number of pressing simulations (here, the number of pressure changes). Note that the pressing simulation may be performed by further changing the posture, position or contact range of the fingertip in contact with the contact surface as other fingerprint acquisition conditions. In such a case, the number of pseudo-contact type fingerprint images to be generated is determined based on the number of fingerprint acquisition conditions and the number of combinations of the fingerprint acquisition conditions.

[0044] For example, the pseudo-contact fingerprint images IMG11 and IMG13 shown in Fig. 6 are pseudo-contact fingerprint images acquired by performing a pressure simulation under fingerprint acquisition conditions in which the center position of the camera's image sensor is misaligned with the center position of the fingertip. Also, for example, the pseudo-contact fingerprint image IMG12 is acquired by performing a pressure simulation under fingerprint acquisition conditions in which the pressure applied to the fingertip pressed against the contact surface is not uniform and the pressure changes greatly from the center of the fingertip toward the periphery. In the pseudo-contact fingerprint image IMG12, the fingerprint becomes fuzzy from the center to the periphery of the fingertip.

[0045] When a fingertip (fingerprint) held up in free space is imaged by a camera, the simulation unit 113 performs a simulation (camera imaging simulation) in which the imaging parameters (e.g., exposure time, focal length, etc.) of the camera imaging the fingertip are changed for each 3D model MD11, ... of a finger with a pseudo fingerprint (St16B).

[0046] The simulation unit 113 generates pseudo contactless fingerprint images IMG2, . . . (an example of learning biometric data, 2D data) based on the 3D models MD11, . . . of the finger with the pseudo fingerprint on which the imaging simulation of each camera has been performed (St17B).

[0047] As a result, the number of pseudo-contactless fingerprint images generated by simulation unit 113 is the product of the number of 3D models of fingers with pseudo fingerprints multiplied by the number of camera imaging simulations (i.e., the number of combinations of camera imaging parameters). Note that the camera imaging simulation may be executed by further varying other fingerprint acquisition conditions, such as the illuminance or angle of light irradiated onto the fingertip, the imaging distance between the camera and the fingertip, etc. In such a case, the number of pseudo-contactless fingerprint images to be generated is determined based on the number of fingerprint acquisition conditions and the number of combinations of the fingerprint acquisition conditions.

[0048] For example, the pseudo-contactless fingerprint images IMG21 and IMG22 shown in Fig. 6 are pseudo-contactless fingerprint images acquired by simulating fingerprint acquisition conditions in which the fingertip postures relative to the image sensor of the camera are different. The fingerprints captured in the pseudo-contactless fingerprint images IMG21 and IMG22 are captured in different ways because the fingertip postures are different.

[0049] The labeling unit 114 performs a labeling process to label the name of a fictitious person for each 3D model MD11 of a finger with a pseudo fingerprint used to generate each of the pseudo-contact-type fingerprint images IMG1, ... and the pseudo-contactless-type fingerprint images IMG2, ... (St18).

[0050] When the labeling process is completed, the labeling unit 114 outputs the 3D model MD11 of the finger with the pseudo fingerprint labeled with the same label (for example, the name of a fictitious person) and at least one of the pseudo-contact fingerprint images IMG1, ... or the pseudo-contactless fingerprint images IMG2, ... to the memory 12 as a set of pseudo images (pseudo fingerprint image dataset) of one fictitious person (a non-existent person), and stores them in the pseudo fingerprint image dataset DB13 (St19).

[0051] As described above, the training fingerprint data generation device P1 in the embodiment can generate a pseudo-contact fingerprint image IMG1 or a pseudo-contactless fingerprint image IMG2 of a non-existent person that is more suitable for learning a learning model, using the fingerprint pattern image FP1 and the 3D finger model FM1. As a result, by generating the pseudo-contact fingerprint image IMG1, the pseudo-contactless fingerprint image IMG2, etc. of a non-existent person (i.e., training biometric data), the training fingerprint data generation device P1 does not need to consider the privacy of the generated training biometric data, and can increase the degree of freedom regarding the handling of the training biometric data.

[0052] As described above, the training fingerprint data generation device P1 in the embodiment can generate 3D models of many non-existent fingers with pseudo fingerprints by using a small number of fingerprint pattern images FP1 and a small number of finger 3D models FM1. As a result, the training fingerprint data generation device P1 can generate 3D models of many non-existent fingers with pseudo fingerprints by using a small number of fingerprint pattern images FP1 or a small number of finger 3D models FM1, and can therefore obtain a larger number of pseudo fingerprint images (e.g., pseudo-contact fingerprint images IMG1, ... and pseudo-non-contact fingerprint images IMG2, ...) that are training biometric data.

[0053] Moreover, the learning fingerprint data generation device P1 can execute a simulation corresponding to each fingerprint acquisition condition by using the generated 3D model of the finger with the pseudo fingerprint. By using the 3D model of the finger with the pseudo fingerprint in the simulation, the learning fingerprint data generation device P1 can reproduce with high accuracy how the fingerprint is captured based on the position or posture of the finger relative to the camera. By using the 3D model of the finger with the pseudo fingerprint in the simulation, the learning fingerprint data generation device P1 can reproduce with high accuracy how the fingerprint is captured for use in biometric authentication, such as when pressure is applied to the fingertip. This allows the learning fingerprint data generation device P1 to acquire a large number of highly accurate pseudo fingerprint images (for example, pseudo-contact type fingerprint images IMG1, ... and pseudo-non-contact type fingerprint images IMG2, ...) corresponding to each fingerprint acquisition condition.

[0054] Furthermore, since the 3D model of the pseudo fingerprint and the pseudo fingerprint image generated by the learning fingerprint data generation device P1 are fingerprint data of a fictitious person, the privacy of the person providing the fingerprint pattern image FP1 or the 3D finger model FM1 can be protected even if the fingerprint pattern image FP1 or the 3D finger model FM1 is provided by a real person.

[0055] Next, the pseudo-fingerprint image data set DB13 will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the pseudo-fingerprint image data set DB13. Note that the pseudo-fingerprint image data set DB13 shown in Fig. 6 is merely an example and is not limited thereto.

[0056] The pseudo fingerprint image dataset DB13 stores, for each label, a 3D model of a finger with a pseudo fingerprint, at least one pseudo-contact fingerprint image or at least one pseudo-contactless fingerprint image generated by a simulation using the 3D model of the finger with a pseudo fingerprint, and labels "Mr. A," "Mr. B," and "Mr. C" indicating non-existent persons corresponding to the 3D model of the finger with a pseudo fingerprint.

[0057] The pseudo-fingerprint image data set DB13 may record the fingerprint acquisition conditions under which the pseudo-contact-type fingerprint image or the pseudo-contactless-type fingerprint image was acquired in association with the pseudo-contact-type fingerprint image or the pseudo-contactless-type fingerprint image.

[0058] As a result, the training fingerprint data generation device P1 in the embodiment can treat the 3D model MD11 of a finger with a pseudo fingerprint as the finger of a non-existent person, and can manage the non-existent person in association with the training fingerprint data (pseudo-contact fingerprint image IMG1 or pseudo-non-contact fingerprint image IMG2) generated from the fingerprint of this non-existent person.

[0059] As described above, the training fingerprint data generation device P1 (an example of a training biometric data generation device and a terminal device) in the embodiment includes an image generation unit 112 (an example of a first generation unit) that generates pseudo biometric data (e.g., pseudo fingerprint pattern images FP11 to FP13, etc.) of a non-existent person based on biometric data including biometric information (e.g., fingerprint pattern image FP1, etc.), and generates pseudo shape data (e.g., a 3D model MD1 of a finger) of a biometric part of the non-existent person based on shape data indicating the three-dimensional shape of the biometric part (e.g., a finger, etc.) corresponding to the biometric data. The image generation unit 112 (an example of a second generation unit) generates 3D models of pseudo fingers (FM11 to FM13), and an output unit (an example of a simulation unit 113, a labeling unit 114) generates pseudo biological part data of a non-existent person (e.g., a 3D model MD11 of a finger with a pseudo fingerprint) based on the pseudo biological data and the pseudo shape data, and generates and outputs learning biological data (e.g., pseudo-contact fingerprint image IMG1, pseudo-non-contact fingerprint image IMG2, etc.) to be used in machine learning of a learning model based on the generated pseudo biological part data.

[0060] As a result, the training fingerprint data generation device P1 in the embodiment can use the fingerprint pattern image FP1 and the 3D finger model FM1 to generate a pseudo-contact fingerprint image IMG1 or a pseudo-contactless fingerprint image IMG2 of a non-existent person that is more suitable for learning a learning model. Also, by generating the pseudo-contact fingerprint image IMG1, pseudo-contactless fingerprint image IMG2, etc. of a non-existent person (i.e., training biometric data), the training fingerprint data generation device P1 does not have to consider the privacy of the generated training biometric data and can increase the degree of freedom regarding the handling of the training biometric data.

[0061] In the training fingerprint data generation device P1 in the embodiment, the biometric data (e.g., fingerprint pattern image FP1, etc.) is biometric information of an actual person. As a result, the training fingerprint data generation device P1 in the embodiment can generate a pseudo-contact type fingerprint image IMG1 or a pseudo-contactless type fingerprint image IMG2 that is closer to an actual person and is of a non-actual person, and can generate a pseudo-contact type fingerprint image IMG1 or a pseudo-contactless type fingerprint image IMG2 that is more suitable for learning a learning model.

[0062] Furthermore, in the training fingerprint data generation device P1 in the embodiment, the shape data (for example, the 3D model MD1 of a finger) is data that indicates the three-dimensional shape of a biological part of an actual person. As a result, the training fingerprint data generation device P1 in the embodiment can generate a pseudo-contact type fingerprint image IMG1 or a pseudo-contactless type fingerprint image IMG2 that is closer to an actual person and is of a non-actual person, and can generate a pseudo-contact type fingerprint image IMG1 or a pseudo-contactless type fingerprint image IMG2 that is more suitable for learning a learning model.

[0063] Furthermore, the simulation unit 113 of the training fingerprint data generation device P1 in the embodiment performs texture mapping of the pseudo biometric data (pseudo fingerprint pattern images FP11, ...) onto the pseudo shape data (3D model FM11 of a pseudo finger) to generate pseudo biometric part data (for example, 3D model MD11 of a finger with a pseudo fingerprint). As a result, the training fingerprint data generation device P1 in the embodiment can generate a three-dimensional fingertip with a fingerprint, which is the fingertip of a fictitious person, and can generate a pseudo-contact-type fingerprint image IMG1 or a pseudo-non-contact-type fingerprint image IMG2 that is more suitable for learning a learning model.

[0064] Furthermore, the labeling unit 114 of the training fingerprint data generation device P1 in the embodiment outputs the pseudo shape data (3D model MD11 of a finger with a pseudo fingerprint) in association with the training biometric data (pseudo-contact type fingerprint image IMG1 or pseudo-non-contact type fingerprint image IMG2) generated based on the pseudo shape data (3D model MD11 of a finger with a pseudo fingerprint). This allows the training fingerprint data generation device P1 in the embodiment to manage the 3D model MD11 of a finger with a pseudo fingerprint (i.e., one finger) in association with the training fingerprint data (pseudo-contact type fingerprint image IMG1 or pseudo-non-contact type fingerprint image IMG2) generated from the fingerprint of the 3D model MD11 of a finger with a pseudo fingerprint.

[0065] Furthermore, the labeling unit 114 of the training fingerprint data generation device P1 in the embodiment assigns the same label to the pseudo shape data (3D model MD11 of a finger with a pseudo fingerprint) and the training biometric data (pseudo-contact-type fingerprint image IMG1 or pseudo-contactless fingerprint image IMG2) generated based on this pseudo shape data (3D model MD11 of a finger with a pseudo fingerprint), and outputs them. This allows the training fingerprint data generation device P1 in the embodiment to handle the 3D model MD11 of a finger with a pseudo fingerprint as the finger of a non-existent person, and to manage the non-existent person in association with the training fingerprint data (pseudo-contact-type fingerprint image IMG1 or pseudo-contactless fingerprint image IMG2) generated from the fingerprint of this non-existent person.

[0066] Furthermore, the simulation unit 113 of the training fingerprint data generation device P1 in the embodiment executes a simulation (e.g., pressing simulation, camera imaging simulation, etc.) corresponding to the acquisition conditions of the biometric data on the pseudo biometric part data (3D model MD11 of a finger with a pseudo fingerprint), and generates training biometric data (pseudo-contact type fingerprint image IMG1 or pseudo-contactless type fingerprint image IMG2) corresponding to the acquisition conditions based on the simulated pseudo biometric part data (3D model MD11 of a finger with a pseudo fingerprint). As a result, the training fingerprint data generation device P1 in the embodiment can acquire training fingerprint data (pseudo-contact type fingerprint image IMG1 or pseudo-contactless type fingerprint image IMG2) under each fingerprint acquisition condition for the 3D model MD11 of a finger with a pseudo fingerprint (i.e., one finger).

[0067] In the training fingerprint data generation device P1 in the embodiment, the condition for acquiring biometric data is whether or not to acquire biometric data with a biological part (e.g., a finger) in contact with a predetermined surface. This allows the training fingerprint data generation device P1 in the embodiment to acquire training fingerprint data in a contact state (pseudo-contact-type fingerprint image IMG1) or training fingerprint data in a non-contact state (pseudo-non-contact-type fingerprint image IMG2) for the 3D model MD11 of a finger with a pseudo fingerprint (i.e., one finger).

[0068] Moreover, the conditions for acquiring biometric data in the training fingerprint data generation device P1 in the embodiment are conditions related to the imaging parameters of a camera that images a biological part (e.g., a finger). This allows the training fingerprint data generation device P1 in the embodiment to acquire training fingerprint data for the 3D model MD11 of a finger with a pseudo fingerprint (i.e., one finger) when the imaging parameters of the camera (e.g., exposure time, focal length, etc.) are changed.

[0069] In the training fingerprint data generation device P1 in the embodiment, the acquisition conditions for biometric data are conditions related to the posture parameters or position parameters of a biometric part (e.g., a finger) relative to a camera that captures an image of the biometric part (e.g., a finger). This allows the training fingerprint data generation device P1 in the embodiment to acquire training fingerprint data when parameters such as the posture, position, and imaging distance of the 3D model MD11 of a finger with a pseudo fingerprint (i.e., one finger) relative to the camera are changed.

[0070] Although various embodiments have been described above with reference to the accompanying drawings, the present disclosure is not limited to such examples. It is clear that a person skilled in the art can conceive of various modifications, corrections, substitutions, additions, deletions, and equivalents within the scope of the claims, and it is understood that these also belong to the technical scope of the present disclosure. In addition, the components in the various embodiments described above may be arbitrarily combined within the scope of the invention. [Industrial Applicability]

[0071] INDUSTRIAL APPLICABILITY The present disclosure is useful as providing a training biometric data generation device and a training biometric data generation method that more efficiently generate training data for a learning model used in biometric authentication. [Explanation of symbols]

[0072] 10 Input section 11 Control section 12. Memory 13 Output section 111 Learning Department 112 Image Generation Unit 113 Simulation Department 114 Labeling section DB11 Fingerprint Pattern Dataset DB12 Finger 3D Model Dataset DB13 Pseudo Fingerprint Image Dataset FM1 Finger 3D Model FM11, FM12, FM13 3D model of artificial finger FP1 Fingerprint pattern image FP11, FP12, FP13 Pseudo fingerprint pattern images IMG1,IMG11,IMG12,IMG13 Pseudo-contact fingerprint image IMG2,IMG21,IMG22 Pseudo non-contact fingerprint image MD11 3D model of a finger with a fake fingerprint P1 Fingerprint data generator for learning

Claims

1. A first generation unit that generates pseudo biometric data of a non-existent person based on biometric data including biometric information; a second generating unit that generates pseudo shape data of a body part of a non-existent person based on shape data indicating a three-dimensional shape of the body part corresponding to the biometric data; an output unit that generates pseudo body part data of the non-existent person based on the pseudo biometric data and the pseudo shape data, and generates and outputs learning biometric data to be used in machine learning of a learning model based on the generated pseudo body part data. A device for generating biological data for learning purposes.

2. The biometric data is biometric information of a real person. The learning biometric data generating device according to claim 1 .

3. The shape data is data indicating a three-dimensional shape of the body part of an actual person.

3. The learning biometric data generating device according to claim 1 or 2.

4. The output unit is generating the pseudo body part data by texture mapping the pseudo biological data onto the pseudo shape data; The learning biometric data generating device according to claim 1 .

5. The output unit is outputting the simulated body part data in association with learning biometric data generated based on the simulated body part data; The learning biometric data generating device according to claim 1 .

6. The output unit is assigning the same label to the simulated body part data and the learning biometric data generated based on the simulated body part data, and outputting the resulting data; The learning biometric data generating device according to claim 1 .

7. The output unit is performing a simulation corresponding to an acquisition condition of biometric data on the pseudo body part data, and generating learning biometric data corresponding to the acquisition condition based on the simulated pseudo body part data; The learning biometric data generating device according to claim 1 .

8. the condition for acquiring the biometric data is a condition as to whether or not the biometric data is acquired in a state in which the body part is in contact with a predetermined surface; The learning biometric data generating device according to claim 7.

9. The biological data acquisition conditions are conditions related to imaging parameters of a camera that images the biological part. The learning biometric data generating device according to claim 7.

10. The acquisition condition of the biological data is a condition related to a posture parameter or a position parameter of the biological part with respect to a camera that captures an image of the biological part. The learning biometric data generating device according to claim 7.

11. A method for generating learning biometric data performed by a terminal device, comprising: Generate pseudo biometric data of a non-existent person based on biometric data including biometric information; generating pseudo shape data of the body part of a non-existent person based on shape data indicating a three-dimensional shape of the body part corresponding to the biometric data; generating pseudo body part data of the non-existent person based on the pseudo body data and the pseudo shape data; Based on the generated pseudo body part data, learning biological data to be used in machine learning of a learning model is generated. Generate and output A method for generating biometric data for training.

Images