Facial recognition device

Abstract

The present invention provides a facial recognition device that has a low risk of misidentifying and registering parts of a captured image other than the face as a face. [Solution] The face recognition device 1 includes a shooting unit 15 that takes images for face recognition, a face detection means 22 that detects faces from the captured images, a face feature extraction means 24 that extracts feature quantities from the face image, which is the region of the detected face, a storage means 12 that stores non-face feature quantities, which are feature quantities of non-face images, which are images other than human faces, and a determination means 25 that determines whether or not to register the face image or the feature quantities extracted from the face image as data for face recognition based on the similarity between the feature quantities extracted from the face image and the non-face feature quantities.

Description

Technical Field

[0001] The present invention relates to a face authentication device.

Background Art

[0002] A collation device is known that collates registered images, which are a plurality of registered face images, with a face image, which is a captured image of the person to be collated, and determines whether the person to be collated is a person registered in advance in association with the registered images (see, for example, Patent Document 1). The collation device described in Patent Document 1 determines whether the person to be collated is a pre-registered person based on the similarity between the feature quantity of the captured face image and the feature quantity of the registered image.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0004] However, the collation device described in Patent Document 1 may erroneously recognize parts other than the human face, such as human body parts including ears and hands contained in the captured image, and articles reflected in the background, as a human face when capturing a registered image, and register them as an image for face authentication.

[0005] The present invention is made to solve such a problem, and an object thereof is to provide a face authentication device with a low risk of erroneously recognizing a part other than a human face contained in a captured image as a human face and registering it as an image for face authentication.

Means for Solving the Problems

[0006] The facial recognition device according to the present invention comprises: an imaging unit for capturing images for facial recognition; a face detection means for detecting faces from the captured images; a face feature extraction means for extracting feature quantities from the face image, which is the region of the detected face; a storage means for storing non-face feature quantities, which are feature quantities of non-face images, which are images other than human faces; and a determination means for determining whether or not to register the face image or the feature quantities extracted from the face image as data for facial recognition, based on the similarity between the feature quantities extracted from the face image and the non-face feature quantities.

[0007] Furthermore, the facial recognition device according to the present invention includes: an imaging unit for capturing images for facial recognition; a facial detection means for detecting faces from the captured images; a facial feature extraction means for extracting feature quantities from the facial images of the detected faces; a storage means for storing non-facial feature quantities, which are feature quantities of non-facial images that are images other than human faces; a determination means for determining whether or not to use the facial image for facial recognition based on the similarity between the feature quantities extracted from the facial image and the non-facial feature quantities; and a facial recognition means for performing facial recognition using the facial image based on the result of the determination.

[0008] Furthermore, in the facial recognition device according to the present invention, it is preferable that the non-face features are features extracted from non-face images detected by the face detection means.

[0009] Furthermore, in the facial recognition device according to the present invention, it is preferable that the face detection means includes a first learning model that outputs a face image which is a face region included in the input image, or a face feature point included in the face image, and the face feature means includes a second learning model that takes a face image or a face feature point included in the face image as input and outputs a face feature amount included in the face image.

[0010] Furthermore, in the facial recognition device according to the present invention, it is preferable that the second learning model is learned by inputting images of a person's face region. [Effects of the Invention]

[0011] The facial recognition device according to the present invention can reduce the risk of misidentifying parts of a captured image other than a person's face as a person's face and registering them as images for facial recognition. [Brief explanation of the drawing]

[0012] [Figure 1] (a) is a diagram showing an example of an image for facial recognition captured by the facial recognition device according to the embodiment, and (b) is a diagram showing an example of a feature quantity calculated from the region detected as a face in the image captured for facial recognition. [Figure 2] This figure shows the configuration of a facial recognition image registration system including a facial recognition device according to an embodiment. [Figure 3] Figure 2 is a flowchart showing the operation of the facial recognition image registration process performed by the facial recognition device. [Modes for carrying out the invention]

[0013] The facial recognition device according to the present invention will be described below with reference to the drawings. However, it should be noted that the technical scope of the present invention is not limited to those embodiments, but extends to the invention described in the claims and its equivalents.

[0014] Figure 1(a) shows an example of an image for facial recognition captured by the facial recognition device according to the embodiment, and Figure 1(b) shows an example of feature quantities calculated from the region detected as a human face in the image captured for facial recognition.

[0015] The facial recognition device according to the embodiment captures an image I1 for facial recognition and detects the facial region, which is the area where a person H1's face exists, and five organ points P1 to P5, which are located at both eyes, the nose, and the ends of the mouth, within the facial region A1, using a first learning model. Next, the facial recognition device according to the embodiment extracts feature quantities from the facial image, including the detected face, and the coordinates of organ points P1 to P5, using a second learning model. The second learning model is trained using an image of the facial region to output feature quantities that are highly similar to the user's own face and less similar to the faces of other people. Next, the facial recognition device according to the embodiment compares one or more non-facial feature quantities obtained from the second learning model when a non-facial image other than a human face is input with the feature quantities extracted using the second learning model. If the facial recognition device according to the embodiment determines that the feature quantities extracted using the second learning model are similar to the non-facial feature quantities, it decides not to register the captured image I1 as an image for facial recognition.

[0016] The facial recognition device according to this embodiment prevents objects appearing in the background of person H1 from being mistakenly recognized as a person's face and registered as an image for facial recognition by not registering image I1 which has features determined to be similar to non-face features. The first learning model used by the facial recognition device according to this embodiment may mistakenly recognize object G1 appearing in the background of person H1 as a person's face, detect object region A1' as a face region, and mistakenly detect object points P1'~P5' as organ points. When a second learning model trained using images of a person's face is input, a non-face image extracted based on points other than organ points P1~P5, such as object points P1'~P5', outputs features that are approximately the same as those of other non-face images.

[0017] In Figure 1(b), features S11, S12, S21, and S22 represent features obtained when facial feature points of a human face are input to the second learning model. Features S31 to S36 represent features obtained when images of human body parts other than human faces, or images of objects other than human body parts, are input to the second learning model.

[0018] The second learning model, upon input of a pair of facial images of the first person H11 with different facial expressions and organ points detected from the facial images, outputs a pair of features S11 and S12 with very high similarity. The second learning model, upon input of a pair of facial images of a second person H12, who is different from the first person H11, with different facial expressions and organ points detected from the facial images, outputs a pair of features S21 and S22 that have low similarity to features S11 and S12 but very high similarity to each other.

[0019] On the other hand, when the second learning model receives a non-face image that is not a face image, or feature points detected from a non-face image, it outputs approximately the same feature quantities regardless of what the non-face image is other than a human face. Feature quantity S31 is output by the second learning model when an image of the backs of both hands, or feature points detected from an image of the backs of both hands, is input. Feature quantity S32 is output by the second learning model when an image of the back of the head including one ear, or feature points detected from an image of the back of the head, is input. Feature quantity S33 is output by the second learning model when an image of one ear, or feature points detected from an image of one ear, is input. Feature quantity S34 is output by the second learning model when an image of crossed fingers, or feature points detected from an image of fingers, is input. Feature quantity S35 is output by the second learning model when an image of the back room, or feature points detected from an image of the back room, is input. Feature quantity S36 is output by the second learning model when an image of checkered clothing, or feature points detected from an image of checkered clothing, is input.

[0020] Here, the second learning model is trained such that it receives only face images as input, the feature quantities of the same person are close to each other, and the feature quantities of different persons are distant from each other. Since training is not performed on anything other than face images, the feature quantity output when an input other than a face image is input is originally indefinite. However, it has been experimentally found that features indicating non-face match, differ significantly from an original face, and the similarity between non-faces that are not human faces may become extremely high. Furthermore, since non-faces tend to aggregate into several clusters in the feature space, non-face determination can be performed using representative feature vectors of the clusters. Therefore, although the images input to the second learning model for G11 to G16 are completely different, the feature quantities S31 to S36 output from the second learning model have extremely high similarity.

[0021] The face authentication device according to the embodiment refers to one or more pieces of feature quantity data indicating feature quantities acquired by inputting non-face images into the second learning model, and determines whether or not to register the image I1 as an image for face authentication. The face authentication device according to the embodiment determines whether or not to register the image I1 as an image for face authentication using the non-face feature quantity, thereby preventing registration of an image for authentication when a human non-face is erroneously recognized as a face. The face authentication device according to the embodiment prevents registration of an image for authentication when a non-face is erroneously recognized as a human face, so that erroneous detection of a person other than a registrant caused by erroneous detection of a non-face image can be prevented.

[0022] Figure 2 is a diagram showing a configuration of a face authentication image registration system including the face authentication device according to the embodiment.

[0023] A face authentication image registration system 100 includes a server device 101 and a face authentication device 1. The number of each of the server device 101 and the face authentication device 1 is not limited to one, and may be plural. The face authentication devices 1 are connected to be communicable with each other via a network N. The network N is a wide area communication network such as an intranet and the Internet.

[0024] The server device 101 stores the first learning model 121 and the second learning model 122, generates desired information according to a request from the facial recognition device 1, and transmits the generated information to the facial recognition device 1. The server device 101 comprises a server communication unit 111, a server storage unit 112, a server operation unit 113, a server display unit 114, and a server control unit 115.

[0025] The server communication unit 111 has a communication interface circuit conforming to wired / wireless communication standards such as Ethernet® and IEEE802.11, and communicates with the facial recognition device 1 via the network N to send and receive various information.

[0026] The server storage unit 112 includes semiconductor memory, magnetic storage media, and / or optical storage media. The server storage unit 112 stores the code and various data of the computer program executed by the server control unit 115 to control the operation of the server device 101. The computer program is installed in the server storage unit 112 by known methods via a computer-readable storage medium such as a CD-ROM or DVD-ROM, or via a communication line. The computer program may also be distributed from a server or the like and installed in the server storage unit 112.

[0027] The server memory unit 112 stores the first learning model 121 and the second learning model 122. The first learning model 121 is trained to detect a rectangular face region containing a human face in an image when an image containing a human face is input, detect five organ points (both eyes, the tip of the nose, and both ends of the mouth) included in the detected face region, and output the coordinates of the detected face region and organ points. The first learning model 121 is a face recognition model such as CenterFace that enables high-speed processing by using a detection method called CenterNet, which defines a rectangular face region, also called a BoundingBox, at the center point of the BoundingBox. The first learning model 121 is trained using a dataset containing images of multiple faces. The first learning model 121 may also be trained to detect the coordinates of organ points other than both eyes, the tip of the nose, and both ends of the mouth. The second learning model 122 is trained to output a multi-dimensional feature vector when a face image, which is an image of a human face, or the coordinates of organ points, which are feature points of a human face in the face image, are input. The second learning model 122 is intentionally trained to output feature vectors such that, when the feature vectors are placed in the feature space, images of the same person are close together and images of different people are far apart, by inputting only images of human face regions. The second learning model 122 is, for example, an image classification model such as ArcFace that has been trained to output 512-dimensional feature vectors. The first learning model 121 may output face images, which are the face regions contained in the input image, or the feature points of faces contained in the face images, for the input image.

[0028] The server operation unit 113 has an operation device such as a keyboard, mouse, or touch panel, and an interface circuit that acquires signals from the operation device. It accepts operations from the user and outputs a signal corresponding to the accepted operation to the server control unit 115.

[0029] The server display unit 114 has a display device such as a liquid crystal display or an organic EL display, and an interface circuit that outputs images to the display device, and displays various information such as images and text according to instructions from the server control unit 115.

[0030] The server control unit 115 includes a processor such as a CPU or multiprocessor and its peripheral circuits. The processor controls the operation of the server device 101 by executing computer programs stored in the server storage unit 112. The server control unit 115 includes, for example, a CPU (Central Processing Unit). The processing unit 20 may include a DSP (Digital Signal Processor), LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), etc.

[0031] The facial recognition device 1 comprises a communication unit 11, a storage unit 12, an operation unit 13, a display unit 14, and a processing unit 20. The facial recognition device 1 is an electronic computer such as a personal computer, a notebook personal computer, a tablet computer, and a smartphone.

[0032] The communication unit 11 has a communication interface circuit that conforms to wired / wireless communication standards such as Ethernet and IEEE 802.11, and communicates with the server device 101 via the network N to send and receive various information.

[0033] The storage unit 12 is an example of a storage means and includes memory devices such as RAM and ROM, fixed disk devices such as hard disks, or portable storage devices such as flexible disks and optical disks. The storage unit 12 stores operating system programs, driver programs, application programs, data, etc., used for processing in the processing unit 20. For example, the storage unit 12 stores a facial recognition image registration program that causes the processing unit 20 to perform facial recognition image registration. The facial recognition image registration program may be installed in the storage unit 12 from a computer-readable portable recording medium such as a CD-ROM or DVD-ROM using a known setup program. Alternatively, the facial recognition image registration program may be distributed from a server or the like and installed in the storage unit 12.

[0034] Furthermore, the memory unit 12 stores one or more non-face feature data 120. The non-face feature data 120, also called dummy template data, represents the non-face features obtained from the second learning model 122 when an image other than a human face (non-face image) or corresponding feature points is input to the second learning model 122 stored in the server memory unit 112 of the server device 101. The non-face features may be each of the multiple features obtained from the second learning model 122 when multiple non-face images or corresponding feature points are input to the second learning model 122, or they may be representative values ​​that represent multiple features, such as the mean and median of the multiple features. For example, the non-face feature vectors are clustered, and the representative feature vector (mean, or nearest neighbor vector of the mean, etc.) of each cluster is calculated, and these representative vectors are stored as non-face feature data 120 (dummy template). Non-face features are very different from and dissimilar to the features obtained from the second learning model 122 when an image of a human face or corresponding feature points is input to the second learning model 122.

[0035] Non-face feature data 120 is obtained by inputting a non-face image from which five feature points can be extracted by the first learning model 121, or feature points contained in a non-face image, into the second learning model 122. Non-face images used when obtaining non-face feature data 120 include images that include parts of the human body captured by the imaging unit 15 when the user instructs the face recognition device 1 via the operation unit 13 to take a picture with the imaging unit 15. Non-face images used when obtaining non-face feature data 120 include images that include ears, images that include the back of a person's head with ears, images of multiple fingers, images of palms, and images of the back of hands, etc. Non-face images used when obtaining non-face feature data 120 also include images that include items other than parts of the human body captured by the imaging unit 15 when the user instructs the face recognition device 1 via the operation unit 13. Non-face images used when obtaining non-face feature data 120 include images of items that may be placed in the background of the user being photographed, images of the user's clothing being photographed, etc. Furthermore, when acquiring non-face feature data 120, it is desirable to use non-face images that have been pre-verified to not contain human faces.

[0036] The operation unit 13 has input devices such as a keyboard and mouse, and an interface circuit that acquires signals from the input devices, and outputs signals to the processing unit 20 according to the operations of the user, such as a system developer.

[0037] The display unit 14 has a display including liquid crystal, organic EL, etc., and an interface circuit that outputs image data to the display, and displays various information on the display according to instructions from the processing unit 20.

[0038] The processing unit 20 is configured to comprehensively control the operation of the facial recognition device 1 and comprises one or more processors and their peripheral circuits. The processing unit 20 may also include a DSP, LSI, ASIC, FPGA, etc. The processing unit 20 performs processing based on the program stored in the storage unit 12. The processing unit 20 also controls the operation of each component of the facial recognition device 1 so that processing is performed appropriately.

[0039] The processing unit 20 includes, as functional blocks, an acquisition means 21, a face detection means 22, a normalization means 23, a face feature extraction means 24, a determination means 25, a registration means 26, and an output means 27. These functional blocks are functional modules realized based on a program performed by the processing unit 20. These functional blocks may also be firmware implemented in the processing unit 20. The face detection means 22 may also include a first learning model 121 that, for an input image, outputs a face image which is a face region contained in the input image, or face feature points contained in the face image. The face feature extraction means 24 may also include a second learning model 122 that takes a face image or face feature points contained in the face image as input and outputs face feature quantities contained in the face image.

[0040] Figure 3 is a flowchart showing the operation of the facial recognition image registration process performed by the facial recognition device 1. The facial recognition image registration process shown in Figure 3 is performed mainly by the processing unit 20 in cooperation with each element of the facial recognition image registration system 100, based on a facial recognition image registration program stored in the storage unit 12 beforehand. The facial recognition image registration process shown in Figure 3 is performed when a user registers an image for authentication with the facial recognition device 1.

[0041] First, the acquisition means 21 acquires image data representing an image taken by the user (S101). First, the acquisition means 21 receives a shooting instruction from the user via the operation unit 13. Next, the acquisition means 21 instructs the shooting unit 15 to take a picture, and the shooting unit 15, having been instructed to take a picture, takes an image that is presumed to include the user's face and outputs a captured image signal indicating the captured image to the processing unit 20. Next, in response to the input of the captured image signal from the shooting unit 15, the system acquires captured image data representing the image taken by the shooting unit 15 and stores the acquired captured image data in the storage unit 12.

[0042] Next, the face detection means 22 detects a face from the image corresponding to the captured image data acquired by the acquisition means 21 (S102). First, the face detection means 22 acquires the captured image data stored in the storage unit 12. Next, the face detection means 22 outputs a captured image signal indicating the image corresponding to the acquired captured image data, and a face extraction instruction signal indicating the extraction of the face image region and five organ points from the image corresponding to the captured image data to the server device 101. Next, the server control unit 115 inputs the image corresponding to the captured image signal to the first learning model 121 in response to the input of the captured image signal and the face extraction instruction signal, and acquires face extraction data including face region data indicating the face region, which is a rectangular area containing a face, and organ point coordinate data indicating the coordinates of the organ points, which are output from the first learning model 121. Next, the server control unit 115 outputs a face extraction signal indicating the coordinates of the face region and organ points corresponding to the acquired face extraction data to the face authentication device 1. The face detection means 22 acquires face extraction data indicating the coordinates of the face region and organ points corresponding to the face extraction signal when a face extraction signal is input, and stores the acquired face extraction data in the storage unit 12.

[0043] Next, the normalization means 23 normalizes the face image including the face detected in the process shown in S102 (S103). The normalization means 23 generates normalized face image data, which is a normalized face image, by performing a similarity transformation and an affine transformation on the face region including the face detected in the process shown in S102 using the coordinates of organ points. The normalization means 23 generates normalized face image data by expanding or shrinking the face region so that the size of the face included in the normalized face image becomes a predetermined size, and by moving and rotating the face region so that the position and orientation of the face included in the normalized face image becomes a predetermined position and orientation. The normalization means 23 stores the generated normalized face image data in the storage unit 12. The normalization means 23 also modifies the organ point coordinate data indicating the coordinates of organ points according to the normalization process and stores the modified organ point coordinate data in the storage unit 12.

[0044] Next, the face feature extraction means 24 extracts the facial features detected in the process shown in S102 (S104). First, the face feature extraction means 24 acquires normalized face image data stored in the storage unit 12. Next, the face feature extraction means 24 outputs a face image signal indicating the normalized face image data and a feature extraction instruction signal indicating the extraction of features from the normalized face image to the server device 101. Next, the server control unit 115 inputs the normalized face image to the second learning model 122 in response to the input of the face image signal and the feature extraction instruction signal, and acquires feature data indicating the features of the normalized face image output from the second learning model 122. Next, the server control unit 115 outputs a feature signal corresponding to the acquired feature data to the face recognition device 1. The face feature extraction means 24 acquires feature data indicating the features corresponding to the feature signal and stores the acquired feature data in the storage unit 12.

[0045] Next, the determination means 25 determines whether or not to register the image acquired in the process shown in S101 as an image for face recognition, based on the similarity between the feature quantities extracted in the process shown in S104 and the feature quantities of the non-face image. First, the determination means 25 determines whether or not the feature quantities extracted in the process shown in S104 and the feature quantities of the non-face image are similar (S105). The determination means 25 calculates the similarity between the feature quantities extracted in the process shown in S104 and the non-face feature quantities corresponding to the non-face feature quantity data 120 stored in the storage unit 12, based on a similarity calculation algorithm such as the dot product or cosine similarity. Next, the determination means 25 determines whether or not the calculated similarity is equal to or greater than a predetermined similarity threshold. If the calculated similarity is equal to or greater than a predetermined similarity threshold, the determination means 25 determines that the feature quantities extracted in the process shown in S104 and the feature quantities of the non-face image are similar (S105-YES). The determination means 25 determines that the features extracted in the process shown in S104 and the features of the non-face image are not similar when the calculated similarity is less than a predetermined similarity threshold (S105-NO).

[0046] If the determination means 25 determines that the features extracted by the process shown in S104 are similar to the features of the non-face image (S105-YES), the registration means 26 does not register the image as a face recognition image. The output means 27 outputs an alarm signal to the display unit 14 indicating that the image acquired by the process shown in S101 will not be registered as a face recognition image (S107). In response to the alarm signal being input, the display unit 14 displays a warning image indicating that the image acquired by the process shown in S101 will not be registered as a face recognition image. The warning image may also include a shooting instruction indicating that the user's face should be photographed again.

[0047] If the determination means 25 determines that the feature quantities extracted in the process shown in S104 are not similar to the feature quantities of the non-face image (S105-NO), the registration means 26 registers the image acquired in the process shown in S101 as an image for face recognition (S106). The registration means 26 stores the face recognition image data representing the image acquired in the process shown in S101 in the storage unit 12, associating it with the registrant ID, registrant name, notification destination, etc.

[0048] When the facial recognition device 1 determines that a captured image is similar to a non-face image, it does not register the captured image as an image for facial recognition. Therefore, when a non-face is mistakenly recognized as a human face, the registration of an image for recognition is prevented. By preventing the registration of an image for recognition when something that is not a face is mistakenly recognized as a face, the facial recognition device 1 can prevent the misidentification of persons other than registered persons caused by the misdetection of non-face images.

[0049] Furthermore, the face recognition device 1 uses non-face features extracted from non-face images detected by the face detection means to determine whether or not a face image included in a captured image is a non-face image, thus enabling high-accuracy determination of whether or not it is a non-face image. The face recognition device 1 extracts non-face features from detected non-face images using the first learning model 121, similar to the process of extracting features from face images in the face recognition image registration process performed by the authentication image registration device 1. The face recognition device 1 uses the extracted non-face features, similar to the process of extracting features from face images in the face recognition image registration process performed by the authentication image registration device 1, to determine whether or not a face image included in a captured image is a non-face image, thereby enabling high-accuracy determination of whether or not it is a non-face image.

[0050] Furthermore, in the face recognition device 1, the face detection means 22 uses a first learning model 121 to extract face regions and facial feature points included in the captured image. In addition, the face feature extraction means 24 uses a second learning model 122, which is different from the first learning model 121, to extract features that represent the face image. By performing the processes of extracting face regions and feature points and the processes of extracting features using different learning models, the face recognition device 1 can determine with high accuracy whether or not to register an image as an image for face recognition.

[0051] Furthermore, the facial recognition device 1 uses a second learning model 122, which is trained by inputting images of people's faces, to extract facial features from the captured image, thereby enabling highly accurate facial recognition of faces in the captured image.

[0052] In the facial recognition image registration system 100, the first learning model 121 and the second learning model 122 are stored in the server storage unit 112. However, in the facial recognition device according to the embodiment, the first learning model 121 and the second learning model 122 may be stored in the storage unit 12. By storing the first learning model 121 and the second learning model 122 in the storage unit 12, the facial recognition device according to the embodiment can perform facial recognition image registration processing without communicating with the server device 101.

[0053] Furthermore, in the face recognition device 1, non-face feature data 120 is stored in the storage unit 12. However, in the face recognition device according to this embodiment, it is sufficient that at least one data set, such as non-face image data representing a non-face image and non-face feature data, is stored in the storage unit 12. When non-face image data is stored in the storage unit 12, the face recognition device according to this embodiment inputs the non-face image corresponding to the non-face image data and the corresponding feature points into the second learning model 122 to obtain non-face features. Alternatively, in the face recognition device according to this embodiment, the server device 101 may store non-face image data and non-face feature data without the device storing these data itself.

[0054] Furthermore, while the facial recognition device 1 performs a process to normalize the facial image, the facial recognition device according to the embodiment does not need to perform the process to normalize the facial image if the size of the facial region included in the facial image and the orientation of the face satisfy predetermined conditions.

[0055] Furthermore, while the facial recognition device 1 registers the image acquired in the process shown in S101 as data for facial recognition, the facial recognition device according to the embodiment may also register the feature quantities extracted from the facial image as data for facial recognition.

[0056] Furthermore, the facial recognition device 1 is an electronic computer such as a personal computer. However, the facial recognition device according to this embodiment may be a security device that is fixedly placed inside a monitoring area such as a residence, monitors for intrusion into the monitoring area using sensors, and notifies a security center device if an intrusion occurs. The facial recognition device, which is a security device, is communicated via a network to a security center device and a communication terminal (not shown), and is also communicated with a plurality of monitoring sensors that detect abnormalities in the monitoring area that are suspected to be intrusions into the monitoring area. The facial recognition device, which is a security device, also includes a shooting unit whose shooting direction is fixed in a predetermined direction, and captures the user's face with the shooting unit according to the user's instructions.

[0057] The facial recognition device, which is a security device, is set to either an alert mode, which monitors for intrusion into the monitored area, or a deactivation mode, which does not monitor for intrusion into the monitored area, according to the user's instructions. When the security mode is set to alert mode, the facial recognition device, which is a security device, takes a picture of the user's face according to the user's instructions and determines whether the face in the captured image matches the face in the image corresponding to the facial recognition image data stored in the memory unit 12. When the facial recognition device, which is a security device, determines that the face in the captured image matches the face in the image corresponding to the facial recognition image data, it authenticates the user, and the user authentication is successful. After successful user authentication, the facial recognition device, which is a security device, sets the security mode to deactivation mode according to the user's instructions.

[0058] In a facial recognition security device, the shooting direction of the shooting unit 15 is fixed, and inside a residence or other building where the facial recognition device is installed, objects in the background of the user are more likely to be captured when the shooting unit 15 captures the user's face image. In a facial recognition security device, since objects in the background of the user are more likely to be captured when the shooting unit 15 captures the user's face image, there is a higher risk that the first learning model 121 will misidentify non-face areas in the face image as faces. By preventing the registration of authentication images when the first learning model 121 misidentifies non-faces as faces, the facial recognition security device can prevent intruders other than registered persons from entering the facility by preventing non-faces from being misidentified as faces and registered as authentication images.

[0059] Furthermore, while we have described the use of non-face features when registering images for facial recognition, they may also be used during recognition, not just registration. That is, in facial recognition, the system determines whether or not to use a facial image for recognition based on the similarity between the features of the input facial image and the non-face features. If the similarity is above a threshold, the system does not perform facial recognition using the facial image because there is a high probability that it is a non-face. On the other hand, if the similarity is below the threshold, the system performs facial recognition using the facial image because there is a high probability that it is a face. This prevents unnecessary recognition using non-face images by excluding non-face images when they are input as input images for facial recognition. [Explanation of Symbols]

[0060] 1. Facial recognition device 21 Acquisition method 22 Face detection means 23 Normalization means 24. Facial feature extraction method 25 Judgment means 26. Registration Method 27 Output means

Claims

1. The camera unit takes images for facial recognition, A face detection means for detecting faces from the captured image, A face feature extraction means for extracting feature quantities from the face image, which is the detected region of the face, A memory means for storing non-face features, which are the feature quantities of non-face images, which are images other than human faces. A determination means for determining whether or not to register the face image or the features extracted from the face image as data for face recognition, based on the similarity between the features extracted from the face image and the non-face features, A facial recognition device characterized by comprising the following features.

2. The camera unit takes images for facial recognition, A face detection means for detecting faces from the captured image, A face feature extraction means for extracting feature quantities from the face image of the detected face, A memory means for storing non-face features, which are the feature quantities of non-face images, which are images other than human faces. A determination means for determining whether or not to use the face image for face recognition based on the similarity between the feature quantities extracted from the face image and the non-face feature quantities, Based on the result of the above determination, a facial recognition means that performs the facial recognition using the facial image, A facial recognition device characterized by comprising the following features.

3. The face recognition device according to claim 1 or 2, wherein the non-face feature quantity is a feature quantity extracted from the non-face image detected by the face detection means.

4. The face detection means includes a first learning model that outputs a face image which is a face region included in the input image, or a face feature point included in the face image, for the input image. The facial feature extraction means includes a second learning model that takes the facial image or facial feature points contained in the facial image as input and outputs the facial feature quantities contained in the facial image. The facial recognition device according to claim 1 or 2.

5. The facial recognition device according to claim 4, wherein the second learning model is learned by inputting images of a human face region.

Images