Detection device, detection method, and detection program

Abstract

To detect impersonation of an object person with sufficient accuracy regardless of a distance to the object person.SOLUTION: A detection device that detects impersonation of a person includes a storage unit that holds a plurality of models learned for impersonation detection of a face of a person, each of the plurality of models corresponding to each of a plurality of sensors having different sensing methods, a communication unit that receives a plurality of sensor images in which the face of the person is imaged by each of the plurality of sensors, and a processor, wherein the processor specifies distance information to the person using at least one of the plurality of sensor images, and determines a model to be used for impersonation detection of the person among the plurality of models based on the distance information. The sensor image of the person corresponding to the model is input to the determined model, and it is determined whether or not the person is spoofing.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a detection device, a detection method, and a detection program. [Background technology]

[0002] As one of the conventional authentication technologies for verifying identity, a facial recognition technology has been known in which the face of the person to be authenticated is photographed and the photographed facial image is compared with a pre-registered facial image of the person to be authenticated to verify the identity. In this facial recognition technology, when photographing the face of the person to be authenticated, it is necessary to prevent so-called "impersonation," in which a photograph of the person's face is taken to impersonate the person. The photograph of the person's face may be a photographic paper on which a photographed image of the person is printed, or may be a photographed image of the person displayed on a screen.

[0003] Patent Document 1 discloses a face recognition device that detects impersonation using a photograph by obtaining the distance from a predetermined position to the face of the person to be authenticated before matching the face image, and determining the authenticity of the person's face based on the obtained distance. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 4734980 Summary of the Invention [Problem to be solved by the invention]

[0005] However, with the method of Patent Document 1, if the distance to the subject is far, the accuracy of the distance between the projections and recesses of the subject's face, such as the nose and mouth, may decrease, which may reduce the accuracy of determining the authenticity of the subject's face.

[0006] Therefore, an object of the present disclosure is to provide a technology that can determine the authenticity of a target person's face (that is, whether it is an impersonation or not) with sufficient accuracy regardless of the distance from the target person. [Means for solving the problem]

[0007] One aspect of the present disclosure provides a detection device for detecting impersonation of a person, the detection device comprising: a memory unit that stores multiple models that have been trained for detecting impersonation of a person's face, each corresponding to a multiple number of sensors that have different sensing methods; a communication unit that receives multiple sensor images of the person's face taken by each of the multiple sensors; and a processor, wherein the processor uses at least one of the multiple sensor images to identify distance information to the person, and based on the distance information, determines a model from the multiple models to be used to detect impersonation of the person, inputs the sensor image of the person corresponding to the model into the determined model, and determines whether the person is an impersonation.

[0008] One aspect of the present disclosure provides a detection device for detecting impersonation of a person, the detection device comprising: a memory unit that holds a plurality of quality models that belong to model groups corresponding to a plurality of sensors each having a different sensing method and that have been trained for impersonation detection for each quality of a person's face; a communication unit that receives a plurality of sensor images obtained by each of the plurality of sensors capturing an image of the person's face in RGB; and a processor, wherein one of the plurality of sensor images is an RGB face image of the person's face captured in RGB, the processor identifies quality information indicating the quality of the person's face based on the RGB face image, identifies distance information to the person using at least one of the plurality of sensor images, determines a model group from the plurality of model groups to be used for detecting impersonation of the person based on the identified distance information, determines the quality model to be used for detecting impersonation of the person that belongs to the determined model group based on the identified quality information, and inputs the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determines whether the person is an impersonator.

[0009] One aspect of the present disclosure provides a detection device for detecting impersonation of a person, the detection device comprising: a memory unit that holds a plurality of quality models that belong to a model group corresponding to each of a plurality of sensors with different sensing methods and that have been trained for impersonation detection for each quality of a person's face; a communication unit that receives a plurality of sensor images obtained by each of the plurality of sensors capturing an image of the person's face in RGB; and a processor, wherein one of the plurality of sensor images is an RGB face image of the person's face captured in RGB; the processor identifies quality information indicating the quality of the person's face based on the RGB face image; identifies distance information to the person using at least one of the plurality of sensor images; determines a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information; inputs the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determines whether the person is an impersonator.

[0010] One aspect of the present disclosure provides a detection method for detecting impersonation of a person, the detection method comprising: storing in a memory unit a plurality of quality models trained for impersonation detection for each quality of a person's face, the quality models belonging to model groups corresponding to a plurality of sensors each having a different sensing method; receiving a plurality of sensor images of the person's face, each of the plurality of sensors, one of the plurality of sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the plurality of sensor images; determining a model group from the plurality of model groups to be used for detecting impersonation of the person based on the identified distance information; determining the quality model to be used for detecting impersonation of the person, belonging to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator.

[0011] One aspect of the present disclosure provides a detection program for detecting impersonation of a person, the detection program causing a computer to execute the following: storing in a memory unit multiple quality models trained for impersonation detection for each quality of a person's face, the multiple quality models belonging to model groups corresponding to multiple sensors each having different sensing methods; receiving multiple sensor images of the person's face, each of the multiple sensors, one of the multiple sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the multiple sensor images; determining a model group from the multiple model groups to be used for detecting impersonation of the person based on the identified distance information; determining the quality model to be used for detecting impersonation of the person, belonging to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator.

[0012] One aspect of the present disclosure provides a detection method for detecting impersonation of a person, the detection method comprising: storing, in a memory unit, a plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for impersonation detection for each quality of a person's face; receiving a plurality of sensor images in which each of the plurality of sensors captures the person's face; one of the plurality of sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the plurality of sensor images; determining a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator.

[0013] One aspect of the present disclosure provides a detection program for detecting impersonation of a person, the detection program causing a computer to execute the following: storing, in a memory unit, multiple quality models that belong to model groups corresponding to multiple sensors each having a different sensing method and that have been trained for impersonation detection for each quality of a person's face; receiving multiple sensor images of the person's face from each of the multiple sensors, one of the multiple sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the multiple sensor images; determining a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator.

[0014] These comprehensive or specific aspects may be realized as a system, an apparatus, a method, an integrated circuit, a computer program, or a recording medium, or may be realized as any combination of a system, an apparatus, a method, an integrated circuit, a computer program, and a recording medium. [Effects of the Invention]

[0015] According to the present disclosure, it is possible to determine with sufficient accuracy whether the face of a target person is genuine (that is, whether it is an impersonation or not) regardless of the distance from the target person. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a block diagram showing a configuration example of a face authentication system according to a first embodiment. [Figure 2] 1 is a flowchart showing an example of face authentication processing according to the first embodiment; [Figure 3] 1 is a flowchart showing an example of a spoofing determination process according to the first embodiment. [Figure 4] FIG. 10 is a block diagram showing a configuration example of a face authentication system according to a second embodiment. [Figure 5] 10 is a flowchart showing an example of face authentication processing according to the second embodiment. [Figure 6] 10 is a flowchart showing a first example of a spoofing determination process according to the second embodiment. [Figure 7] 10 is a flowchart showing a second example of the spoofing determination process according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, embodiments of the present disclosure will be described in detail with appropriate reference to the drawings. However, more detailed description than necessary may be omitted. For example, detailed descriptions of well-known matters and redundant descriptions of substantially identical configurations may be omitted. This is to avoid unnecessary redundancy in the following description and to facilitate understanding by those skilled in the art. Note that the accompanying 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.

[0018] The functions of one configuration shown in this embodiment may be realized by two or more physical configurations, or the functions of two or more configurations may be realized by, for example, one physical configuration.

[0019] (Embodiment 1) FIG. 1 is a block diagram showing an example of the configuration of a face authentication system 10 according to the first embodiment.

[0020] The facial recognition system 10 is a system for performing facial recognition of a person entering, for example, a station ticket gate, a facility, a room, etc. For example, if the facial recognition system 10 is successful in authenticating a person's face, it allows the person to enter (for example, by opening a gate, unlocking a door, etc.), and if it is unsuccessful in authenticating a person's face, it does not allow the person to enter (for example, by closing a gate, locking a door, etc.).

[0021] Facial recognition must prevent so-called "impersonation," in which a person who does not have the right to enter uses photographic paper with the face image of the person who has the right to enter printed on it or a screen with that face image displayed on it.

[0022] As a method for preventing impersonation, as disclosed in Patent Document 1, a method is known in which the distance from a predetermined position to the face of the person to be authenticated is obtained before matching the face image, and the authenticity of the face of the person to be authenticated is determined based on the obtained distance, thereby detecting impersonation using a photograph.

[0023] However, in the method of Patent Document 1, when the distance to the target person is far, the accuracy of the distance between the projections and recesses of the target person's face, such as the nose and mouth, may decrease, and the accuracy of determining the authenticity of the target person's face may decrease. Therefore, this embodiment provides a technology that can determine the authenticity of the target person's face (i.e., whether it is an impersonation) with sufficient accuracy regardless of the distance to the target person. This will be described in detail below.

[0024] The face authentication system 10 includes an RGB sensor 11, a depth sensor 12, a liveness detection device 20, a face authentication server 30, and a management device 40. The RGB sensor 11 and the depth sensor 12 may be integrated with the liveness detection device 20.

[0025] The RGB sensor 11 is installed in a position where it can capture an image of a person attempting to enter, and captures RGB images continuously over time. RGB stands for Red, Green, and Blue. An RGB image is an image having color information for each pixel. The RGB sensor 11 includes, for example, a CMOS sensor or a CCD sensor. The RGB sensor 11 may also be read as an RGB camera.

[0026] The depth sensor 12 is installed at a position where it can capture an image of a person attempting to enter, and captures depth images continuously over time. A depth image is an image having depth (distance) information for each pixel. The depth sensor includes, for example, a Time Of Flight (ToF) sensor or a stereo camera. The depth sensor may also be referred to as a depth camera.

[0027] The living body detection device 20 is a device that detects whether an object is a living body or not. The living body detection device 20 includes a processor 21, a memory 22, a storage 23, a communication unit 24, an input unit 25, and an output unit 26.

[0028] The processor 21 realizes the functions of the living body detection device 20 by reading and processing programs, data, etc. stored in the memory 22. The processor 21 may be interpreted as a processing unit, a central processing unit (CPU), a microprocessor unit (MPU), a controller, etc. The processor may also include a neural processing unit (NPU) and / or a graphics processing unit (GPU). Details of the functions of the living body detection device 20 will be explained as appropriate.

[0029] The memory 22 stores programs, data, and the like for realizing the functions of the biological detection device 20. The memory 22 may be configured by a volatile storage medium and / or a non-volatile storage medium.

[0030] The storage 23 holds programs, data, and the like for realizing the functions of the biological detection device 20. The storage 23 may be configured with a non-volatile storage medium such as a hard disk drive (HDD), a solid state drive (SSD), or a flash memory. Note that at least one of the memory 22 and the storage 23 may be interpreted as a storage unit.

[0031] The communication unit 24 receives an RGB image from the RGB sensor 11 and a depth image from the depth sensor 12 through a communication network 51. Examples of the communication network 51 include a wired LAN, a wireless LAN, Bluetooth (registered trademark), a mobile communication network, the Internet, a USB cable, etc.

[0032] Furthermore, the communication unit 24 transmits and receives information to and from the face authentication server 30 and the management device 40 through a communication network 52. Examples of the communication network 52 include a wired LAN, a wireless LAN, Bluetooth, a mobile communication network, the Internet, etc. The communication network 52 may be the same as the communication network 51, or may be different.

[0033] The input unit 25 is a device that accepts input from a user. Examples of the input unit 25 include a keyboard, a mouse, a touch panel, and a microphone.

[0034] The output unit 26 is a device that outputs information to the user. Examples of the output unit 26 include a display and a speaker.

[0035] The face authentication server 30 authenticates a face image. For example, the face authentication server 30 receives an RGB face image captured by the RGB sensor 11 from the liveness detection device 20, and compares the received RGB face image with a face image of a person registered in advance. The face image comparison may be performed using a known technique. If the comparison is successful, the face authentication server 30 may transmit "authentication successful" as the face authentication result to the liveness detection device 20, and if the comparison is unsuccessful, the face authentication server 30 may transmit "authentication failed" as the face authentication result to the liveness detection device 20. This function of the face authentication server 30 may be included in the liveness detection device 20.

[0036] The management device 40 is installed, for example, near the entrance and is used by a security guard (or a monitor, etc.). For example, when the living body detection device 20 detects a person who is not permitted to enter among the people entering, it transmits an RGB facial image of that person to the management device 40. The management device 40 displays the received RGB facial image of the person on a display. The security guard can recognize a person attempting to enter illegally by looking at the RGB facial image displayed on the management device 40.

[0037] <How to detect spoofing> Next, a method for determining whether or not an imitation has occurred using an RGB image and a method for determining whether or not an imitation has occurred using a depth image according to this embodiment will be described.

[0038] As shown in FIG. 1, the storage 23 stores an RGB model 60 and a depth model 70 in advance.

[0039] The RGB model 60 is a learning model that has been trained to determine whether an RGB facial image input to the RGB model 60 is a "living body" or an "impersonation" by using as learning data an image of a person's face photographed by the RGB sensor 11 and an image of an impersonator's face photographed by the RGB sensor 11. When an RGB facial image photographed by the RGB sensor 11 is input, the RGB model 60 outputs as an impersonation determination result whether the RGB facial image is a "living body" or an "impersonation" score. Note that the impersonation determination result may also be a "living body" score or an "impersonation" score.

[0040] The depth model 70 is a learning model that has been trained to determine whether a depth image input to the depth model 70 is a "living body" or an "impersonation" using as learning data an image of a face of a person who is not an impersonator captured by the depth sensor 12 and an image of a face of an impersonator captured by the depth sensor 12. When a depth face image captured by the depth sensor 12 is input, the depth model 70 outputs as an impersonation determination result whether the depth face image is a "living body" or an "impersonation" score. Note that the impersonation determination result may be a "living body" score or an "impersonation" score.

[0041] As a method of determining impersonation using an RGB image, processor 21 inputs an RGB facial image into RGB model 60, and based on the impersonation determination result output from RGB model 60, identifies whether the RGB facial image is a "biometric" or "impersonation."

[0042] As a method of determining impersonation using a depth image, the processor 21 inputs a depth face image into a depth model 70, and based on the impersonation determination result output from the depth model 70, identifies whether the depth face image is a "living body" or an "impersonation."

[0043] 2 is a flowchart showing an example of face authentication processing according to Embodiment 1. The living body detection device 20 executes the face authentication processing shown in FIG.

[0044] The processor 21 receives an RGB image from the RGB sensor 11 and a depth image from the depth sensor 12 via the communication unit 24 (S101).

[0045] The processor 21 identifies a person to be authenticated (hereinafter referred to as a target person) from the RGB image and the depth image (S102).

[0046] The processor 21 extracts an RGB face image including the face of the target person from the RGB image, and extracts a depth face image including the face of the target person from the depth image (S103).

[0047] The processor 21 performs face recognition processing on the RGB face image (S104). For example, the processor 21 transmits the RGB face image to the face recognition server 30 and receives a face recognition result (authentication success or authentication failure) from the face recognition server 30.

[0048] The processor 21 also performs an impersonation determination process for the target person using the RGB face image and / or the depth face image (S105). The details of the impersonation determination process will be described later (see FIG. 3), but the impersonation determination process provides an impersonation determination result for the target person, either "biometric" or "impersonation."

[0049] After obtaining the face authentication result for the target person in step S104 and the impersonation determination result in step S105, the processor 21 proceeds to the next step S107 (S106).

[0050] The processor 21 determines whether the facial authentication result of the target person is "authentication successful" and whether the impersonation determination result of the target person is "biometric" (S107).

[0051] If the facial authentication result of the target person is "authentication successful" and the impersonation determination result of the target person is "biometric" (S107: YES), the processor 21, for example, permits the target person to enter (S108) and ends this process (or returns the process to step S101). Note that when the target person who has been permitted to enter approaches the entrance, the processor 21 may, for example, open the gate or unlock the door.

[0052] If the facial authentication result of the target person is "authentication failed" or the impersonation determination result of the target person is "impersonation" (S107: NO), the processor 21, for example, denies the target person entry (S109) and ends this process (or returns the process to step S101). Note that when the target person whose entry has been denied approaches the entrance, the processor 21 may, for example, close the gate or lock the door. In this case, the processor 21 may also transmit an RGB facial image of the target person to the management device 40.

[0053] Through the above process, the living body detection device 20 can permit entry to a target person who has been successfully authenticated and who has been determined to be not an impersonator (i.e., a living body), and can deny entry to a target person whose face authentication has failed or who has been determined to be an impersonator. In other words, the living body detection device 20 can prevent unauthorized entry by an impersonator.

[0054] 3 is a flowchart showing an example of the spoofing determination process according to Embodiment 1. This process corresponds to a detailed description of the process in step S105 in FIG.

[0055] The processor 21 calculates person distance information indicating the distance to the target person based on the depth face image of the target person (S201). However, the processor 21 may calculate the person distance information using other methods. For example, the processor may calculate the person distance information based on the size of the RGB face image (i.e., the size of the target person's face in the RGB image). Specifically, the larger the size of the RGB face image, the shorter the distance to the target person, and the smaller the size of the RGB face image, the longer the distance to the target person.

[0056] The processor 21 determines whether the person distance information is equal to or smaller than a predetermined threshold value (S202).

[0057] If the person distance information is equal to or less than the predetermined threshold (S202: YES), the processor 21 performs an impersonation determination process using the depth model 70 (S203). That is, the processor 21 inputs the depth face image of the target person into the depth model 70 and obtains an impersonation determination result. Then, this process ends (that is, the process of step S105 in FIG. 2 ends).

[0058] If the distance information is greater than the predetermined threshold (S202: NO), the processor 21 performs the spoofing determination process using the depth model 70, as in step S203 (S204). Additionally, the processor 21 performs the spoofing determination process using the RGB model 60 (S205). That is, the processor 21 inputs the RGB facial image of the target person into the RGB model 60 and obtains the spoofing determination result. Note that the processes of steps S204 and S205 may be executed first, or may be executed in parallel. Furthermore, even if the determination result of step S204 is "spoofing," if the determination result of step S205 is "biological," the processor 21 may determine the spoofing determination result as "biological." This is because the accuracy of spoofing determination using an RGB facial image is relatively high when the distance to the target person is relatively long. Alternatively, the spoofing determination process using the depth model 70 in step S204 may be omitted. Then, this process ends (that is, the process of step S105 in FIG. 2 ends).

[0059] In the case of spoofing detection using a depth face image, the accuracy of spoofing detection is relatively high when the distance to the target person is relatively close, but is relatively low when the distance to the target person is relatively far. On the other hand, in the case of spoofing detection using an RGB face image, the accuracy of spoofing detection does not decrease significantly even when the distance to the target person is relatively far. Therefore, as shown in FIG. 3, when the distance to the target person is relatively close (i.e., when the person distance information is equal to or less than a predetermined threshold), the spoofing detection process is performed using the depth model 70 as the main model and the RGB model 60 as the support model. This allows spoofing detection to be performed with sufficient accuracy regardless of the distance to the target person. Note that when the distance to the target person is relatively far, the spoofing detection process may be performed using the RGB model 60 as the main model and the depth model 70 as the support model, or alternatively, the RGB model 60 may be used without using the depth model 70.

[0060] (Embodiment 2) When a person is wearing a mask, glasses, or accessories, or when the face of the person detected by the sensor is not facing forward, the accuracy of the spoofing determination may be reduced. In the second embodiment, a technology is described that can perform the spoofing determination with sufficient accuracy even when the person is wearing a mask, glasses, or accessories, or when the face of the person detected by the sensor is not facing forward. Note that in the second embodiment, the description of the components already described in the first embodiment may be omitted.

[0061] FIG. 4 is a block diagram showing an example of the configuration of a face authentication system 10 according to the second embodiment.

[0062] The face authentication system 10 includes an RGB sensor 11, a depth sensor 12, an IR sensor 13, a thermal sensor 14, a living body detection device 20, a face authentication server 30, and a management device 40. Note that the face authentication system 10 does not necessarily need to include all of the depth sensor 12, the IR sensor 13, and the thermal sensor 14, and may include only sensors that are suitable for the environment in which the sensors are installed.

[0063] The RGB sensor 11, the depth sensor 12, the face authentication server 30, and the management device 40 are as described in the first embodiment.

[0064] The IR sensor 13 is installed in a position where it can capture an image of a person attempting to enter, and emits infrared light to capture IR images continuously over time. An IR image is an image in which each pixel has an amount of infrared light reflected. The IR sensor 13 may be interpreted as an IR camera.

[0065] The thermal sensor 14 is installed in a position where it can capture images of people attempting to enter, detects far-infrared rays emitted from people, and captures thermal images continuously over time. A thermal image is an image that has temperature information for each pixel. The thermal sensor 14 may also be referred to as a thermal camera.

[0066] The RGB sensor 11, the depth sensor 12, the IR sensor 13, and the thermal sensor 14 may be integrated with the living body detection device 20.

[0067] As shown in FIG. 4, the storage 23 stores a normal RGB model 61, a glasses RGB model 62, and a mask RGB model 63 as an RGB model group.

[0068] The normal RGB model 61 is similar to the RGB model 60 described in the first embodiment.

[0069] The glasses RGB model 62 is a learning model that has been trained to determine whether an RGB facial image wearing glasses input to the glasses RGB model 62 is a "living body" or an "impersonation" by using as learning data an image of the face of a person wearing glasses that has been photographed by the RGB sensor 11 and an image of a face wearing impersonation glasses that has been photographed by the RGB sensor 11. When an RGB facial image wearing glasses is input, the glasses RGB model 62 outputs as a determination result whether the RGB facial image wearing glasses is a "living body" or an "impersonation" image.

[0070] The mask RGB model 63 is a learning model that has been trained to determine whether an RGB face image wearing a mask input to the mask RGB model 63 is a "living body" or an "impersonation" by using as learning data a face image of a person wearing a non-impersonation mask photographed by the RGB sensor 11 and an image of a face wearing an impersonation mask photographed by the RGB sensor 11. When an RGB face image wearing a mask is input to the mask RGB model 63, the mask RGB model 63 outputs as a determination result whether the RGB face image wearing the mask is a "living body" or an "impersonation"

[0071] As shown in FIG. 4, the storage 23 stores a normal depth model 71, a glasses depth model 72, and a mask depth model 73 as a depth model group.

[0072] The normal depth model 71 is similar to the depth model 70 described in the first embodiment.

[0073] The glasses depth model 72 is a learning model that has been trained to determine whether a depth face image wearing glasses input to the glasses depth model 72 is a "living body" or an "impersonation" using as learning data an image of the face of a person wearing glasses that has been captured by the depth sensor 12 and an image of the face wearing impersonation glasses that has been captured by the depth sensor 12. When a depth face image wearing glasses is input, the glasses depth model 72 outputs as a determination result whether the depth face image wearing glasses is a "living body" or an "impersonation"

[0074] The mask depth model 73 is a learning model that has been trained to determine whether a depth face image wearing a mask input to the mask depth model 73 is a "living body" or an "impersonation" using as learning data a face image of a person wearing a non-impersonation mask photographed by the depth sensor 12 and an image of a face wearing an impersonation mask photographed by the depth sensor 12. When a depth face image wearing a mask is input, the mask depth model 73 outputs as a determination result whether the depth face image wearing the mask is a "living body" or an "impersonation"

[0075] As shown in FIG. 4, the storage 23 stores a normal IR model 81, a glasses IR model 82, and a mask IR model 83 as an IR model group.

[0076] The normal IR model 81 is a learning model that has been trained to determine whether an IR facial image input to the normal IR model 81 is a "living body" or an "impersonation" by using as learning data a facial image of a person who is not an impersonator photographed by the IR sensor 13 and an image of an impersonator's face photographed by the IR sensor 13. When an IR facial image is input to the normal IR model 81, the normal IR model 81 outputs as a determination result whether the IR facial image is a "living body" or an "impersonation" image.

[0077] Glasses IR model 82 is a learning model that has been trained to determine whether an IR face image wearing glasses input to glasses IR model 82 is "biological" or "spoofed," using as learning data face images of a person wearing glasses that have been captured by IR sensor 13 and images of a face wearing spoofed glasses that have been captured by IR sensor 13. When an IR face image wearing glasses is input, glasses IR model 82 outputs as a determination result whether the IR face image wearing glasses is "biological" or "spoofed."

[0078] The mask IR model 83 is a learning model that has been trained to determine whether an IR face image wearing a mask input to the mask IR model 83 is a "living body" or an "impersonation" by using as learning data a face image of a person wearing a non-impersonation mask photographed by the IR sensor 13 and an image of a face wearing an impersonation mask photographed by the IR sensor 13. When an IR face image wearing a mask is input to the mask IR model 83, the mask IR model 83 outputs as a determination result whether the IR face image wearing the mask is a "living body" or an "impersonation"

[0079] As shown in FIG. 4, the storage 23 stores a normal thermal model 91, an eyeglasses thermal model 92, and a mask thermal model 93 as a group of thermal models.

[0080] The normal thermal model 91 is a learning model that has been trained to determine whether a thermal facial image input to the normal thermal model 91 is a "living body" or an "impersonation" by using as learning data a facial image of a person who is not an impersonator photographed by the thermal sensor 14 and an image of an impersonator's face photographed by the thermal sensor 14. When a thermal facial image is input, the normal thermal model 91 outputs as a determination result whether the thermal facial image is a "living body" or an "impersonation" image.

[0081] The glasses thermal model 92 is a learning model that has been trained to determine whether a thermal face image wearing glasses input into the glasses thermal model 92 is a "living body" or an "impersonation" by using as learning data face images of a person wearing glasses that have not been photographed by the thermal sensor 14 and images of a face wearing impersonation glasses that have been photographed by the thermal sensor 14. When a thermal face image wearing glasses is input, the glasses thermal model 92 outputs as a determination result whether the thermal face image wearing glasses is a "living body" or an "impersonation"

[0082] The mask thermal model 93 is a learning model that has been trained to determine whether a thermal face image wearing a mask input to the mask thermal model 93 is a "living body" or an "impersonation" by using as learning data face images of a person wearing a non-impersonation mask photographed by the thermal sensor 14 and images of a face wearing an impersonation mask photographed by the thermal sensor 14. When a thermal face image wearing a mask is input, the mask thermal model 93 outputs as a determination result whether the thermal face image wearing the mask is a "living body" or an "impersonation"

[0083] The models such as "normal," "glasses," and "mask" included in the above-mentioned model group are examples of quality models. In addition, quality models for whether or not an accessory is worn, or quality models for different facial orientations of a person, etc. may be stored in the storage 23.

[0084] 5 is a flowchart showing an example of face authentication processing according to Embodiment 2. The living body detection device 20 executes the face authentication processing shown in FIG.

[0085] The processor 21 receives an RGB image from the RGB sensor 11, a depth image from the depth sensor 12, an IR image from the IR sensor 13, and a thermal image from the thermal sensor 14 through the communication unit 24 (S301).

[0086] The processor 21 identifies a person to be authenticated (hereinafter referred to as a target person) from the RGB image, the depth image, the IR image, and the thermal image (S302).

[0087] The processor 21 extracts an RGB facial image including the face of the target person from the RGB image, extracts a depth facial image including the face of the target person from the depth image, extracts an IR facial image including the face of the target person from the IR image, and extracts a thermal facial image including the face of the target person from the thermal image (S303).

[0088] The processor performs face recognition processing on the RGB face image (S304), similar to step S104.

[0089] The processor 21 also performs an impersonation determination process for the target person using at least one of the RGB face image, the depth face image, the IR face image, and the thermal face image (S305). The details of the impersonation determination process will be described later (see FIG. 6 or FIG. 7), but the impersonation determination process provides a determination result of either "biometric" or "impersonation" for the target person.

[0090] The processing of steps S306 to S309 shown in FIG. 5 is the same as steps S106 to S109 described with reference to FIG. 2, and therefore a description thereof will be omitted.

[0091] 6 is a flowchart showing a first example of the spoofing determination process according to Embodiment 2. This process corresponds to a detailed description of the process in step S305 in FIG.

[0092] The processor 21 identifies quality information of the target person based on the RGB facial image (S401). The quality information is information indicating whether the target person is wearing a mask and / or whether the target person is wearing glasses. The quality information may further include information indicating whether the target person is wearing accessories and / or the orientation of the target person's face. The process of identifying the quality information of the target person from the RGB facial image may be realized using a known image recognition technology.

[0093] Processor 21 determines a quality model to be used for spoofing determination based on the quality information (S402). For example, if the quality information indicates that a mask is being worn, processor 21 determines to use the quality model of "mask" for spoofing determination, if the quality information indicates that glasses are being worn, processor 21 determines to use the quality model of "glasses" for spoofing determination, and if the quality information indicates that neither a mask nor glasses is being worn, processor 21 determines to use the quality model of "normal" for spoofing determination.

[0094] The processor 21 calculates distance information indicating the distance to the target person based on the depth face image (S403), similar to the first embodiment. However, the processor 21 may calculate the distance information to the target person using another method, similar to the first embodiment.

[0095] The processor determines whether the distance information is equal to or less than a predetermined threshold (S404).

[0096] If the distance information is equal to or less than the threshold (S404: YES), processor 21 performs spoofing determination processing using the quality model determined in step S402 from the depth model group (S405). For example, if the quality model of "mask" is determined in step S402, processor 21 performs spoofing determination using mask depth model 73. For example, if the quality model of "glasses" is determined in step S402, processor 21 performs spoofing determination using glasses depth model 72. For example, if the quality model of "normal" is determined in step S402, processor 21 performs spoofing determination using normal depth model 71. Then, this processing ends (i.e., the processing of step S305 in FIG. 5 ends).

[0097] If the distance information is greater than the threshold (S404: NO), the processor 21 performs spoofing determination processing using the quality model determined in step S402 from the depth model group, as in step S405 (S406). In addition, the processor 21 performs spoofing determination using the quality model determined in step S402 from the RGB model group (S407). For example, if the quality model of "mask" is determined in step S402, the processor 21 performs spoofing determination using the mask RGB model 63. For example, if the quality model of "glasses" is determined in step S402, the processor 21 performs spoofing determination using the glasses RGB model 62. For example, if the quality model of "normal" is determined in step S402, the processor 21 performs spoofing determination using the normal RGB model 61. Then, this processing ends (i.e., the processing of step S305 in FIG. 5 ends).

[0098] This allows for sufficiently accurate spoofing determination even when the target person is wearing a mask, glasses, etc. In addition, as described in the first embodiment, it is possible to perform sufficiently accurate spoofing determination regardless of the distance to the target person.

[0099] Fig. 7 is a flowchart showing a second example of the spoofing determination process according to embodiment 2. This process corresponds to a detailed description of the process of step S305 in Fig. 5. In embodiment 2, the process in Fig. 7 may be executed instead of the process in Fig. 6 described above.

[0100] As in step S401, processor 21 identifies quality information of the target person based on the RGB face image (S501).

[0101] The processor 21 determines a model group and a quality model to be used for spoofing determination based on the quality information (S502). For example, if the quality information indicates that a person is wearing a mask, the processor 21 determines to use the "IR model group" as the sensor model group and "mask" as the quality model. If the quality information indicates that a person is wearing glasses, the processor 21 determines to use the "thermal model group" as the sensor model group and "glasses" as the quality model. If the quality information indicates that a person is not wearing glasses, the processor 21 determines to use the "depth model group" as the sensor model group and "normal" as the quality model. The relationship between the quality information and the model group and the quality model to be used for spoofing determination may be determined in advance according to the characteristics of each sensor. For example, since the RGB sensor 11 or the IR sensor 13 can more easily obtain facial features of a person wearing a mask than the thermal sensor 14 or the depth sensor 12, the processor 21 may determine to use the RGB sensor group or the IR sensor group when the quality information indicates that a person is wearing a mask.

[0102] The processor 21 calculates distance information indicating the distance to the target person based on the depth face image (S503), similar to the first embodiment. However, the processor 21 may calculate the distance information to the target person using another method, similar to the first embodiment.

[0103] The processor 21 determines whether the distance information is equal to or smaller than a predetermined threshold value (S504).

[0104] If the distance information is equal to or less than the threshold (S504: YES), the processor 21 performs spoofing determination processing using the sensor model group and the quality model determined in step S502 (S505). For example, if the "IR model group" is determined as the sensor model group and "mask" is determined as the quality model in step S502, the processor performs spoofing determination using the mask IR model 83 of the IR model group. For example, if the "thermal model group" is determined as the sensor model group and "glasses" is determined as the quality model in step S502, the processor 21 performs spoofing determination using the glasses thermal model 92. For example, if the "depth model" is determined as the sensor model group and "normal" is determined as the quality model in step S502, the processor 21 performs spoofing determination using the normal depth model 71. Then, this processing ends (i.e., the processing of step S305 in FIG. 5 ends).

[0105] If the distance information is greater than the threshold (S504: NO), the processor 21 performs spoofing determination processing using the model group and the quality model determined in step S502, as in step S405 (S506). In addition, the processor 21 performs spoofing determination using the quality model determined in step S502 from the RGB model group (S507). For example, if it is determined in step S502 that "mask" is to be used as the quality model, the processor 21 performs spoofing determination using the mask RGB model 63. For example, if it is determined in step S502 that "glasses" is to be used as the quality model, the processor 21 performs spoofing determination using the glasses RGB model 62. For example, if it is determined in step S502 that "normal" is to be used as the quality model, the processor 21 performs spoofing determination using the normal RGB model 61. Then, this processing ends (i.e., the processing of step S305 in FIG. 5 ends).

[0106] This allows for sufficiently accurate spoofing determination even when the target person is wearing a mask, glasses, etc. In addition, as described in the first embodiment, it is possible to perform sufficiently accurate spoofing determination regardless of the distance to the target person.

[0107] In step S502, the processor 21 may determine the priority of the model group used for spoofing determination based on the quality information. Then, in step S505 or step S506, the processor 21 may calculate a score for spoofing determination using each model group, weight each score according to the priority, and determine the result as "living body" if the sum of the weighted scores is greater than a predetermined threshold, or determine the result as "spoofing" if the sum is less than the threshold.

[0108] Furthermore, only the model group and quality model to be used may be stored in the storage 23. For example, quality models that are not used in determining whether a person is wearing a mask or whether a person is wearing glasses may not need to be stored in the storage 23. This allows the amount of data stored in the storage 23 to be reduced.

[0109] Summary of the Disclosure The above description of the first and second embodiments discloses the following techniques. <Technology 1> A detection device (e.g., a living body detection device 20) that detects impersonation of a person includes a memory unit (e.g., a memory 22, a storage 23) that holds multiple models (e.g., an RGB model 60, a depth model 70) that have been trained for detecting impersonation of a person's face, each corresponding to multiple sensors (e.g., an RGB sensor 11, a depth sensor 12) that have different sensing methods, a communication unit (24) that receives multiple sensor images of the person's face captured by each of the multiple sensors, and a processor (21).The processor uses at least one of the multiple sensor images to identify distance information to the person, and based on the distance information, determines which of the multiple models to use for detecting impersonation of the person, inputs the sensor image of the person corresponding to that model into the determined model, and determines whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person, so that spoofing can be judged with sufficient accuracy.

[0110] <Technology 2> In the detection device described in Technology 1, one of the plurality of sensor images is an RGB face image obtained by capturing the face of the person in RGB, and another of the plurality of sensor images is a depth face image obtained by capturing the depth of the face of the person. As a result, spoofing is judged using an appropriate model according to the distance information to the person, so that spoofing can be judged with sufficient accuracy.

[0111] <Technology 3> In the detection device described in Technology 2, if the distance information is equal to or less than a predetermined threshold, the processor determines to use a model corresponding to the depth face image for detecting impersonation of the person, and if the distance information is greater than the threshold, determines to use a model corresponding to the RGB face image for detecting impersonation of the person. This allows for sufficiently accurate impersonation determination even when the distance to the person is relatively long.

[0112] <Technology 4> A detection device (e.g., a living body detection device 20) for detecting impersonation of a person includes a storage unit (e.g., a memory 22, a storage 23) for storing a plurality of quality models (e.g., a normal model, a glasses model, a mask model, etc.) that belong to model groups (e.g., an RGB model group, a depth model group, an IR model group, a thermal model group, etc.) corresponding to a plurality of sensors (e.g., an RGB sensor 11, a depth sensor 12, an IR sensor 13, a thermal sensor 14) with different sensing methods, and that have been trained for impersonation detection for each quality of a person's face; a communication unit (24) for receiving a plurality of sensor images obtained by each of the plurality of sensors capturing the face of the person; and a processor (21), One of the sensor images is an RGB facial image of the person's face captured in RGB, and the processor identifies quality information indicating the quality of the person's face based on the RGB facial image, identifies distance information to the person using at least one of the plurality of sensor images, determines a model group from the plurality of model groups to be used for detecting impersonation of the person based on the identified distance information, determines the quality model to be used for detecting impersonation of the person, belonging to the determined model group, based on the identified quality information, and inputs the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determines whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0113] <Technology 5> A detection device (e.g., a living body detection device 20) for detecting impersonation of a person includes a storage unit (e.g., a memory 22, a storage 23) that holds a plurality of quality models that belong to model groups (e.g., an RGB model group, a depth model group, an IR model group, a thermal model group, etc.) corresponding to a plurality of sensors (e.g., an RGB sensor 11, a depth sensor 12, an IR sensor 13, and a thermal sensor 14) that have different sensing methods, and that have been trained for impersonation detection for each quality of a person's face; a communication unit (24) that receives a plurality of sensor images of the person's face captured by each of the plurality of sensors; and a processor (21). ), wherein one of the plurality of sensor images is an RGB face image of the person's face captured in RGB, and the processor identifies quality information indicating a quality of the person's face based on the RGB face image, identifies distance information to the person using at least one of the plurality of sensor images, determines a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information, inputs the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determines whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0114] <Technology 6> In the detection device according to Technique 4 or 5, the quality information is information indicating at least one of whether the person is wearing a mask or not and whether the person is wearing glasses or not. This allows for sufficiently accurate impersonation determination even for people wearing masks or glasses.

[0115] <Technology 7> In the detection device described in any one of Techniques 4 to 6, another of the plurality of sensor images is a depth face image capturing the depth of the person's face, and yet another of the plurality of sensor images is an IR image captured by irradiating infrared light onto the person's face. This allows spoofing to be determined with sufficient accuracy in a variety of environments.

[0116] <Technology 8> A detection method for detecting person impersonation includes storing in a memory unit a plurality of models that have been trained for detecting person impersonation, each corresponding to a plurality of sensors with different sensing methods; receiving a plurality of sensor images of the person's face, each of which is captured by one of the plurality of sensors; identifying distance information to the person using at least one of the plurality of sensor images; determining a model from the plurality of models to be used to detect person impersonation based on the distance information; inputting the sensor image of the person corresponding to that model into the determined model; and determining whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person, so that spoofing can be judged with sufficient accuracy.

[0117] <Technology 9> A detection program for detecting person impersonation causes a computer to perform the following steps: store in a memory unit multiple models that have been trained for detecting person impersonation, each corresponding to a plurality of sensors with different sensing methods; receive multiple sensor images of the person's face taken by each of the multiple sensors; identify distance information to the person using at least one of the multiple sensor images; determine a model from the multiple models to be used to detect person impersonation based on the distance information; input the sensor image of the person corresponding to the model into the determined model; and determine whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person, so that spoofing can be judged with sufficient accuracy.

[0118] <Technology 10> A detection method for detecting person impersonation includes storing in a memory unit a plurality of quality models trained for impersonation detection for each quality of a person's face, each of which belongs to a model group corresponding to a plurality of sensors with different sensing methods; receiving a plurality of sensor images of the person's face using each of the plurality of sensors; one of the plurality of sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the plurality of sensor images; determining a model group from the plurality of model groups to be used for detecting person impersonation based on the identified distance information; determining the quality model to be used for person impersonation detection, which belongs to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0119] <Technology 11> A detection program for detecting person impersonation causes a computer to execute the following: storing in a memory unit multiple quality models that belong to model groups corresponding to multiple sensors with different sensing methods and that have been trained for impersonation detection for each quality of a person's face; receiving multiple sensor images of the person's face using each of the multiple sensors, one of which is an RGB face image of the person's face captured in RGB; identifying quality information that indicates the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the multiple sensor images; determining a model group from the multiple model groups to be used for detecting person impersonation based on the identified distance information; determining the quality model that belongs to the determined model group to be used for detecting person impersonation based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0120] <Technology 12> A detection method for detecting impersonation of a person includes storing in a memory unit a plurality of quality models trained for impersonation detection for each quality of a person's face, each of which belongs to a model group corresponding to a plurality of sensors with different sensing methods; receiving a plurality of sensor images of the person's face using each of the plurality of sensors; one of the plurality of sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the plurality of sensor images; determining a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0121] <Technology 13> A detection program for detecting impersonation of a person causes a computer to execute the following: storing in a memory unit a plurality of quality models trained for impersonation detection for each quality of a person's face, the quality models belonging to a model group corresponding to each of a plurality of sensors with different sensing methods; receiving a plurality of sensor images of the person's face by each of the plurality of sensors; one of the plurality of sensor images being an RGB face image of the person's face captured in RGB; identifying quality information indicating the quality of the person's face based on the RGB face image; identifying distance information to the person using at least one of the plurality of sensor images; determining a model group and a quality model to be used for detecting impersonation of the person based on the identified distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group; and determining whether the person is an impersonator. As a result, spoofing is judged using an appropriate model according to the distance information to the person and the quality information, so that spoofing can be judged with sufficient accuracy.

[0122] Although the 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, alterations, substitutions, additions, deletions, and equivalents within the scope of the claims, and it is understood that these also fall within the technical scope of the present disclosure. Furthermore, the components in the above-described embodiments may be combined in any manner without departing from the spirit of the invention. [Industrial Applicability]

[0123] The techniques of the present disclosure are useful for detecting impersonation of a person. [Explanation of symbols]

[0124] 10. Facial Recognition System 11 RGB sensor 12 Depth Sensor 13 IR sensor 14 Thermal Sensor 20. Biodetection device 21 processors 22 Memory 23 Storage 24 Communications Department 25 Input section 26 Output section 30 Face Recognition Server 40 Management device 51,52 Communication Network 60 RGB models 61 Normal RGB model 62 Eyeglasses RGB Model 63 Mask RGB Model 70 Depth Model 71 Normal depth model 72 Eyeglass Depth Model 73 Mask Depth Model 81 Normal IR model 82 Glasses IR Model 83 Mask IR Model 91 Normal thermal model 92 Glasses Thermal Model 93 Mask Thermal Model

Claims

1. A detection device for detecting impersonation of a person, comprising: a storage unit that stores a plurality of trained models for detecting impersonation of a person's face, each of which corresponds to a plurality of sensors that have different sensing methods; a communication unit that receives a plurality of sensor images obtained by capturing images of the face of the person using the plurality of sensors; a processor, The processor: Identifying distance information to the person using at least one of the plurality of sensor images; determining a model to be used for detecting impersonation of the person from among the plurality of models based on the distance information; inputting the sensor image of the person corresponding to the model into the determined model, and determining whether the person is an impersonator; Detection device.

2. one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; Another of the plurality of sensor images is a depth face image capturing the depth of the face of the person. The detection device according to claim 1 .

3. The processor: If the distance information is equal to or less than a predetermined threshold, determining that a model corresponding to the depth face image is used for detecting impersonation of the person; If the distance information is greater than the threshold, determining that a model corresponding to the RGB face image is used for detecting impersonation of the person; The detection device according to claim 2 .

4. A detection device for detecting impersonation of a person, comprising: a storage unit that stores a plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face; a communication unit that receives a plurality of sensor images obtained by capturing images of the face of the person using the plurality of sensors; a processor, one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; The processor: Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a group of models to be used for detecting impersonation of the person from among the plurality of groups of models based on the specified distance information; determining the quality model to be used for detecting impersonation of the person, the quality model belonging to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; Detection device.

5. A detection device for detecting impersonation of a person, comprising: a storage unit that stores a plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face; a communication unit that receives a plurality of sensor images obtained by capturing images of the face of the person using the plurality of sensors; a processor, one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; The processor: Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a set of models and a quality model to be used for detecting impersonation of the person based on the determined distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; Detection device.

6. The quality information is information indicating at least one of whether or not the person is wearing a mask and whether or not the person is wearing glasses.

6. The detection device according to claim 4 or 5.

7. Another one of the plurality of sensor images is a depth face image capturing the depth of the face of the person, Yet another one of the plurality of sensor images is an IR image captured by irradiating infrared light onto the face of the person.

6. The detection device according to claim 4 or 5.

8. A detection method for detecting impersonation of a person, comprising: a storage unit that stores a plurality of trained models for detecting impersonation of a person's face, the models corresponding to the plurality of sensors having different sensing methods; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; Identifying distance information to the person using at least one of the plurality of sensor images; determining a model to be used for detecting impersonation of the person from among the plurality of models based on the distance information; inputting the sensor image of the person corresponding to the model into the determined model, and determining whether the person is an impersonator; Detection method.

9. A detection program for detecting impersonation of a person, a storage unit that stores a plurality of trained models for detecting impersonation of a person's face, the models corresponding to the plurality of sensors having different sensing methods; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; Identifying distance information to the person using at least one of the plurality of sensor images; determining a model to be used for detecting impersonation of the person from among the plurality of models based on the distance information; inputting the sensor image of the person corresponding to the model into the determined model, and determining whether the person is an impersonator; A detection program that causes a computer to perform the following:

10. A detection method for detecting impersonation of a person, comprising: A plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face are stored in a storage unit; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a group of models to be used for detecting impersonation of the person from among the plurality of groups of models based on the specified distance information; determining the quality model to be used for detecting impersonation of the person, the quality model belonging to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; Detection method.

11. A detection program for detecting impersonation of a person, A plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face are stored in a storage unit; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a group of models to be used for detecting impersonation of the person from among the plurality of groups of models based on the specified distance information; determining the quality model to be used for detecting impersonation of the person, the quality model belonging to the determined model group, based on the identified quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; A detection program that causes a computer to perform the following:

12. A detection method for detecting impersonation of a person, comprising: A plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face are stored in a storage unit; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a set of models and a quality model to be used for detecting impersonation of the person based on the determined distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; Detection method.

13. A detection program for detecting impersonation of a person, A plurality of quality models that belong to a model group corresponding to each of a plurality of sensors having different sensing methods and that have been trained for spoofing detection for each quality of a person's face are stored in a storage unit; receiving a plurality of sensor images in which the plurality of sensors have captured images of the face of the person; one of the plurality of sensor images is an RGB face image obtained by capturing a face of the person in RGB; Identifying quality information indicating a quality of the person's face based on the RGB face image; Identifying distance information to the person using at least one of the plurality of sensor images; determining a set of models and a quality model to be used for detecting impersonation of the person based on the determined distance information and quality information; inputting the sensor image of the person corresponding to the determined model group into the quality model of the determined model group, and determining whether the person is an impersonator; A detection program that causes a computer to perform the following:

Images