KR102651303B1 - Apparatus and method for de-identifying and restoring face image - Google Patents

Abstract

The present invention relates to a facial image de-identification and restoration device and method, which includes a de-identification unit that objectifies a face image to select a target image, anonymizes the target image to create an anonymized image, and an anonymized unique identifier that matches the target image. An anonymization unique identifier generator that generates anonymization, a de-anonymization request receiver that receives a de-anonymization request for an anonymized image and verifies the legitimacy of the de-anonymization request, and a data restoration unit that restores the anonymized image and generates restoration data if the de-anonymization request is justified. May include wealth.

Description

Apparatus and method for de-identifying and restoring face images {APPARATUS AND METHOD FOR DE-IDENTIFYING AND RESTORING FACE IMAGE}

The present invention relates to a facial image de-identification and restoration device and method, and more specifically, to a de-identification unit that objectifies a face image, selects a target image, anonymizes the target image, and generates an anonymized image, the target. An anonymization unique identifier generator that generates an anonymization unique identifier matching the image, a de-anonymization request receiver that receives a de-anonymization request for the anonymized image and verifies the legitimacy of the de-anonymization request, and, if the de-anonymization request is justified, the de-anonymization request receiver. It may include a data restoration unit that restores the anonymized image and generates restored data.

Recently, image data collection has been rapidly increasing to develop and advance deep learning technology to imitate human cognitive abilities. Global data platform companies are collecting large amounts of artificial intelligence learning data and are making efforts to link existing cloud services with artificial intelligence learning support services. Because the data collected in this way includes personal information, using or storing it without anonymization and a separate management system may cause privacy issues.

Video data is a representative form of unstructured data and is used for deep learning in various industrial fields such as manufacturing, medical care, autonomous driving, agriculture, and logistics. However, with the revision of the Personal Information Protection Act and the enforcement of the 3 Data Act, it is essential to de-identify faces other than the information subject when using personal information videos or viewing video information processing devices, so there are restrictions in terms of data use. there is. De-identification refers to processing so that a specific individual cannot be identified through pseudonymization or anonymization. De-identification of faces in video data generally involves masking or encryption such as mosaic or blur. it means. However, masking and encryption methods that change the shape of an object have the problem of making it difficult to preserve the usefulness of image data. Therefore, in order to preserve the identity of deep learning results, there is a problem that a lot of time is required for unmasking or decoding before use.

Therefore, research on methods to generate fake faces using generative models has been actively conducted recently. Among them, feature conversion that changes some features of the face through a generative model based on GAN (Generative Adversarial Network), style transfer that changes facial features, face shape, and hair to a desired style, or transfer of the original face to the face of another target. There are many research cases where face shifting, etc., is applied to face de-identification.

Most de-identification studies using GAN generate images by simply mapping based on noise without any constraints. Therefore, there is a problem that image quality may decrease. Additionally, since most studies only focus on creating fake faces, there is a problem that data cannot be restored as needed. Additionally, as learning data, there is a need to control the anonymization strength depending on the purpose of use or the algorithm to be developed, but this is rarely considered in existing studies.

Korean Patent No. 10-2215056 (2021.02.10) relates to a medical image de-identification system, method, and computer-readable recording medium, which receives medical images containing images that can identify the patient's facial area. A medical image receiver, a facial area separator that separates the facial area from the medical image, a request receiver that receives a request from a client to provide data on at least one of the eyes, nose, and mouth included in the facial area, and the requested part. It includes a de-identification unit that performs de-identification on the remaining facial areas except for.

Korean Patent No. 10-2215056 (2021.02.10)

An embodiment of the present invention relates to a facial image de-identification and restoration device and method that can restore an anonymized image by anonymizing a person's facial image without damaging it.

One embodiment of the present invention is a facial image de-identification and restoration device that generates a unique identifier while anonymizing the image, determines whether the restoration request for the anonymized image is justified, and can restore the anonymized image only if the restoration request is justified. and methods.

An apparatus for de-identifying and restoring a face image according to an embodiment of the present invention, comprising: a de-identification unit that objectsizes a face image, selects a target image, anonymizes the target image, and generates an anonymized image; an anonymized unique identifier generator that generates an anonymized unique identifier matching the target image; a de-anonymization request receiving unit that receives a de-anonymization request for the anonymized image and verifies the legitimacy of the de-anonymization request; and a data restoration unit that restores the anonymized image and generates restored data when the de-anonymization request is justified.

The de-identification unit includes a de-identification encoder including a personal identity encoder for generating a personal identity latent vector and an attribute information encoder for generating an attribute information latent vector; and a de-identification decoder that generates a random vector matching the anonymized unique identifier.

The de-identification decoder calculates a summed latent vector including the personal identity latent vector and the attribute information latent vector, and projects the random vector onto the summed latent vector to generate a noise vector that is linearly independent from the summed latent vector. You can.

The de-identification decoder may project the random vector onto the personal identity latent vector to generate a noise vector that is linearly independent from the personal identity latent vector.

The non-identification decoder can generate the noise vector according to [Equation 1].

The de-identification unit may generate the anonymized image based on [Equation 2] according to a preset de-identification level.

The de-identification unit may generate the anonymized image by combining the modulation vector generated according to [Equation 2] and the attribute information latent vector.

The de-anonymization request receiving unit may determine that the de-anonymization request is legitimate when the unique identification key included in the de-anonymization request corresponds to the anonymization unique identifier.

The data restoration unit includes a restoration encoder including a personal identity restoration encoder for generating a non-identifying personal identity latent vector for the anonymized image and an attribute information restoration encoder for generating a non-identifying attribute information latent vector for the anonymized image; and

It may include a reconstruction decoder that calculates a non-identifying sum vector including the non-identification personal identity latent vector and the non-identification attribute information latent vector, and generates the reconstruction data by applying a reconstruction vector to the sum vector.

The data restoration unit may generate the restoration vector according to [Equation 3].

The disclosed technology can have the following effects. However, since it does not mean that a specific embodiment must include all of the following effects or only the following effects, the scope of rights of the disclosed technology should not be understood as being limited thereby.

The facial image de-identification and restoration device and method according to an embodiment of the present invention can anonymize a person's face image without damaging it, thereby restoring the anonymized image.

The device and method for de-identifying and restoring facial images according to an embodiment of the present invention generates a unique identifier while anonymizing the image, determines whether the restoration request for the anonymized image is justified, and restores the anonymized image only if the restoration request is justified. It can be restored.

1 is a diagram illustrating a facial image de-identification and restoration system according to an embodiment of the present invention.
Figure 2 is a diagram explaining the physical configuration of a facial image de-identification and restoration device according to an embodiment of the present invention.
Figure 3 is a diagram explaining the functional configuration of a facial image de-identification and restoration device according to an embodiment of the present invention.
Figure 4 is a diagram illustrating the order in which the facial image de-identification and restoration method is performed according to an embodiment of the present invention.
Figure 5 is a diagram illustrating a process by which a facial image de-identification and restoration device according to an embodiment of the present invention creates an anonymized image and restores it.
FIG. 6 is a diagram illustrating an anonymized image generated by a facial image de-identification and restoration device according to a de-identification level according to an embodiment of the present invention.

Since the description of the present invention is only an example for structural or functional explanation, the scope of the present invention should not be construed as limited by the examples described in the text. In other words, since the embodiments can be modified in various ways and can have various forms, the scope of rights of the present invention should be understood to include equivalents that can realize the technical idea. In addition, the purpose or effect presented in the present invention does not mean that a specific embodiment must include all or only such effects, so the scope of the present invention should not be understood as limited thereby.

Meanwhile, the meaning of the terms described in this application should be understood as follows.

Terms such as “first” and “second” are used to distinguish one component from another component, and the scope of rights should not be limited by these terms. For example, a first component may be named a second component, and similarly, the second component may also be named a first component.

When a component is referred to as being “connected” to another component, it should be understood that it may be directly connected to the other component, but that other components may exist in between. On the other hand, when a component is referred to as being “directly connected” to another component, it should be understood that there are no other components in between. Meanwhile, other expressions that describe the relationship between components, such as "between" and "immediately between" or "neighboring" and "directly neighboring" should be interpreted similarly.

Singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as “comprise” or “have” refer to implemented features, numbers, steps, operations, components, parts, or them. It is intended to specify the existence of a combination, and should be understood as not excluding in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

For each step, identification codes (e.g., a, b, c, etc.) are used for convenience of explanation. The identification codes do not explain the order of each step, and each step clearly follows a specific order in context. Unless specified, events may occur differently from the specified order. That is, each step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the opposite order.

The present invention can be implemented as computer-readable code on a computer-readable recording medium, and the computer-readable recording medium includes all types of recording devices that store data that can be read by a computer system. . Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices, and are also implemented in the form of a carrier wave (e.g., transmission via the Internet). It also includes Additionally, the computer-readable recording medium can be distributed across computer systems connected to a network, so that computer-readable code can be stored and executed in a distributed manner.

All terms used herein, unless otherwise defined, have the same meaning as commonly understood by a person of ordinary skill in the field to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as consistent with the meaning they have in the context of the related technology, and cannot be interpreted as having an ideal or excessively formal meaning unless clearly defined in the present application.

Figure 1 is a diagram illustrating a facial image de-identification and restoration system 100 according to an embodiment of the present invention.

Referring to FIG. 1 , the facial image de-identification and restoration system 100 may include a user terminal 110, a facial image de-identification and restoration device 130, and a database 150.

The user terminal 110 requests anonymization of images collected through the facial image de-identification and restoration device 130, verifies the anonymization unique identifier, sends and receives the de-anonymization request, and verifies the restored data generated through this. It can be implemented in smartphones and wearable devices that can check the data generated through the above process and the metadata analyzed, but is not necessarily limited to this and can also be implemented in various devices such as tablet PCs. there is. The user terminal 110 may be connected to the facial image de-identification and restoration device 130 through a network, and a plurality of user terminals 110 may be simultaneously connected to the facial image de-identification and restoration device 130.

The facial image de-identification and restoration device 130 anonymizes the target image to generate an anonymized image, generates an anonymized unique identifier matching the target image, and, if the de-anonymization request is justified, restores the anonymized image to generate restored data. It can be implemented as a server corresponding to a computer or program that sequentially performs the following operations. The facial image de-identification and restoration device 130 can be wirelessly connected to the user terminal 110 through Bluetooth, WiFi, a communication network, etc., and can exchange data with the user terminal 110 through the network.

The database 150 receives the user's profile, performs payment for order data based on the selected and selected payment method, and generates transaction aggregate data created through processing for the approved payment. It may correspond to a storage device that stores information.

FIG. 2 is a diagram illustrating the physical configuration of a facial image de-identification and restoration device 130 according to an embodiment.

Referring to FIG. 2, the facial image de-identification and restoration device 130 may be implemented including a processor 210, a memory 230, a user input/output unit 250, and a network input/output unit 270.

The processor 210 anonymizes the target image to create an anonymized image, generates an anonymized unique identifier that matches the target image, and, if the de-anonymization request is justified, restores the anonymized image to generate restored data. You can execute, manage the memory 230 that is read or written throughout the process, and schedule the synchronization time between the volatile memory and the non-volatile memory in the memory 230. The processor 210 can control the overall operation of the facial image de-identification and restoration device 130, and is electrically connected to the memory 230, the user input/output unit 250, and the network input/output unit 270 to communicate between them. Data flow can be controlled. The processor 210 may be implemented as a central processing unit (CPU) of the facial image de-identification and restoration device 130.

The memory 230 is implemented as a non-volatile memory such as a solid state drive (SSD) or a hard disk drive (HDD) and is an auxiliary memory used to store all data required for the facial image de-identification and restoration device 130. It may include a main memory implemented as volatile memory such as RAM (Random Access Memory).

The user input/output unit 250 may include an environment for receiving user input and an environment for outputting specific information to the user. For example, the user input/output unit 250 may include an input device including an adapter such as a touch pad, touch screen, on-screen keyboard, or pointing device, and an output device including an adapter such as a monitor or touch screen. In one embodiment, the user input/output unit 250 may correspond to a computing device connected through a remote connection, and in such case, the facial image de-identification and restoration device 130 may be performed as a server.

The network input/output unit 270 includes an environment for connecting with external devices or systems through a network, for example, Local Area Network (LAN), Metropolitan Area Network (MAN), Wide Area Network (WAN), and VAN ( It may include an adapter for communication such as a Value Added Network).

Figure 3 is a diagram explaining the functional configuration of a facial image de-identification and restoration device according to an embodiment of the present invention. Referring to FIG. 3, the facial image de-identification and restoration device 130 includes a de-identification unit 310, an anonymization unique identifier generation unit 330, a de-anonymization request reception unit 350, and a data restoration unit 370. can do.

The de-identification unit 310 may objectify a face image to select a target image and anonymize the target image to create an anonymized image. For example, the de-identification unit 310 may specify a face image arranged on the original image, objectify the face image, and select it as a target image to be anonymized. For example, when several people's faces and a background are present in a photo, the de-identification unit 310 may select each face as the target image. For another example, when there are multiple people's faces in a photo, the de-identification unit 310 may select each person's face as the target image, and through this, the unique identifier generator may create a separate anonymized unique identifier for each person. It has the effect of creating . In addition, when there are multiple people's faces in a photo, the de-identification unit 310 can select each person's face as one target image, and through this, the unique identifier generator can generate an anonymized unique identifier for one target image. can be created. In other words, the de-identification unit 310 can determine whether to set a single target image for input images containing multiple people's faces or to set a target image for each of multiple human faces according to the controller's design. . Here, if the target image is set to a single target image, the anonymized image can be restored through one solution through the data restoration unit, and if the target image is set to multiple, the anonymized image can be restored through a different solution for each target image. It can be restored.

In one embodiment, the de-identification unit 310 includes a de-identification encoder and an anonymization unique encoder including a personal identity encoder for generating a personal identity latent vector and an attribute information encoder for generating an attribute information latent vector. It may include a non-identifying decoder that generates a random vector matching the identifier. That is, the de-identification unit 310 can generate an anonymized image through a de-identification encoder and a de-identification decoder. Here, the de-identification encoder and de-identification decoder are not only the hierarchical structure of the widely used encoder and decoder, but also the process of converting the encoded data from the original image to the anonymized image to input it to the de-identification decoder. It can play a role in processing the data conversion process. Here, the de-identification unit 310 can separately encode the personal identity latent vector and the attribute information latent vector in the process of encoding the target image, and the process of extracting the personal identity latent vector and the attribute information latent vector is Keras implementation. Alternatively, it may be performed through a previously learned function such as Face Net, but is not limited to this.

As described above, the de-identification unit 310 may include generating a random vector matching the anonymized unique identifier through a de-identification decoder. Here, the random vector may consist of the same dimensions as the individual identity latent vector or the summed latent vector described below. For example, the de-identification unit 310 may generate different random vectors according to the anonymized unique identifiers given to each of several target images. In more detail, the de-identification unit 310 may generate a random vector matching the anonymized unique identifier given to A in the process of de-identifying the target image of person A, and the target image identified as person A. Even if there are multiple A, the same random vector can be generated according to the anonymized unique identifier given to A. For another example, in the process of de-identifying the target image of person B, if there are multiple target images, the de-identification unit 310 generates a random vector that matches each target image according to an anonymization unique identifier separately given. can do. That is, the de-identification unit 310 can generate a random vector for the target image according to the given anonymization unique identifier according to the design of the controller, and the process of generating an anonymization image according to the random vector can be encrypted, as described below. As explained, the function used in the process of restoring an anonymized image is dependent on the anonymized unique identifier, so if there is no unique identification key matching the anonymized unique identifier, the anonymized image cannot be restored. This will be explained in detail below.

In one embodiment, the de-identification decoder may produce a summed latent vector containing a personal identity latent vector and an attribute information latent vector, and project a random vector onto the summed latent vector to generate a noise vector that is linearly independent of the summed latent vector. there is. Here, the process of including the personal identity latent vector and the attribute information latent vector can be a combination of vector matrices, and the combination of vector matrices can be a widely known method such as simple sum, merge, or concatenate. A method of projecting an arbitrary vector onto a summed potential vector can be done by using a method such as Gram-Schmidt to obtain orthogonal components of the target vector.

In one embodiment, the de-identification decoder may project a random vector onto the personal identity latent vector to generate a noise vector that is linearly independent of the personal identity latent vector. As described above, the de-identification decoder can generate an orthogonal noise vector that is linearly independent of the personal identity latent vector.

In one embodiment, the de-identification decoder may generate a noise vector according to [Equation 1].

[Equation 1]

(From here, is the noise vector, is a random vector, is the summed potential vector, Is , is an anonymized unique identifier.)

To explain in more detail, Is It is composed of a summed latent vector consisting of components orthogonal to or a component orthogonal to the personal identity latent vector. Here, the random vector ( ) can be determined according to the anonymized unique identifier as described above. also, is described as a summed potential vector, but depending on the controller's design, it can be a personal identity potential vector.

In one embodiment, the de-identification unit 310 may generate an anonymized image based on [Equation 2] according to a preset de-identification level.

[Equation 2]

(From here, is the modulation vector, is the summed potential vector, is the noise vector, is the de-identification level, is the similarity parameter.)

Here, the preset de-identification level may be determined by the controller and may have a 2π periodicity. Here also, is described as a summed potential vector, but depending on the controller's design, it can be a personal identity potential vector. Referring to FIG. 6, the de-identification unit 310 may generate different anonymized images depending on the de-identification level.

In one embodiment, the de-identification unit 310 may generate an anonymized image by combining the modulation vector generated according to Equation 2 and the attribute information latent vector. For example, the de-identification unit 310 If is a personal identity latent vector, the generated modulation vector and attribute information latent vector can be combined through well-known methods such as simple sum, merge, or concatenate, and the combined data can be decoded to create an anonymized image.

The anonymized unique identifier generator 330 may generate an anonymized unique identifier that matches the target image. As described above, the anonymized unique identifier generator 330 can generate an anonymized unique identifier that matches each target image according to the controller's design. Referring to Figure 5, the anonymized unique identifier can be used when creating an anonymous image and can be used to verify the legitimacy of a de-anonymization request to restore data.

The de-anonymization request receiving unit 350 may receive a de-anonymization request for an anonymized image and verify the legitimacy of the de-anonymization request. As described below, the de-anonymization request receiver 350 can verify the legitimacy of the information included in the de-anonymization request, and this legitimacy is determined by the fact that the subject of the de-anonymization request has a unique identification key corresponding to the anonymization unique identifier. It can be verified by whether or not.

In one embodiment, the de-anonymization request receiver 350 may determine that the de-anonymization request is justified when the unique identification key included in the de-anonymization request corresponds to the anonymization unique identifier. For example, the de-anonymization request receiver 350 may request the data restoration unit 370 to restore an anonymized image only when the unique identification key corresponds to the anonymized unique identifier. For another example, the de-anonymization request receiver 350 may request the data restoration unit 370 not to restore the anonymized image if the unique identification key does not correspond to the anonymized unique identifier, and if a third party forcibly arbitrarily Even if the anonymized image is restored through the unique identification key, the random vector used for anonymization cannot be calculated unless the anonymized unique identifier is accurately specified, so restored data corresponding to the target image cannot be generated.

If the request for de-anonymization is justified, the data restoration unit 370 can restore the anonymized image and generate restored data. The data restoration unit 370 may restore the anonymized image and generate restored data if the de-anonymization request is justified according to the de-anonymization request reception unit 350's determination of the legitimacy of the de-anonymization request.

In one embodiment, the data recovery unit 370 includes a personal identity recovery encoder for generating a non-identifying personal identity latent vector for an anonymized image and an attribute information recovery encoder for generating a non-identifying attribute information latent vector for the anonymized image. It may include a restoration encoder and a restoration decoder that calculates a de-identification sum vector including a de-identification personal identity latent vector and a de-identification attribute information latent vector, and generates restored data by applying the restoration vector to the de-identification sum vector. That is, the data restoration unit 370 may include a restoration encoder and a restoration decoder, and can extract characteristics of the anonymized image through the restoration encoder and convert the extracted characteristic data into a form of data that can be decoded. As explained above, this process can be performed if the de-anonymization request is justified, and even if the anonymized data is compulsorily de-anonymized, if the anonymized unique identifier cannot be accurately specified, restored data matching the target image corresponding to the original data is generated. Can not.

In another embodiment, the data restoration unit 370 may generate restored data by combining the vector value calculated by applying the restored vector to the non-identified personal identity latent vector and the attribute information latent vector. As described above, the data recovery unit 370 may selectively perform an operation of generating a de-identified summed latent vector according to the controller's design.

In one embodiment, the data restoration unit 370 may generate the restoration vector according to [Equation 3].

[Equation 3]

(From here, is the restoration vector)

That is, the data restoration unit 370 can transform the data into a form that can decode the vector value extracted from the anonymized image through the restoration vector, and decode the transformed data to generate restored data.

Figure 4 is a diagram illustrating the order in which the facial image de-identification and restoration method is performed according to an embodiment of the present invention. Referring to FIG. 4, the facial image de-identification and restoration method can objectify the face image through the de-identification unit 310, select a target image, and anonymize the target image to create an anonymized image (S410). .

The facial image de-identification and restoration method can generate an anonymized unique identifier matching the target image through the anonymized unique identifier generator 330 (S430).

The facial image de-identification and restoration method can receive a de-anonymization request for an anonymized image through the de-anonymization request receiver 350 and verify the legitimacy of the de-anonymization request (S450).

The facial image de-identification and restoration method can generate restored data by restoring the anonymized image if the request for de-anonymization is justified (S470).

Although the present invention has been described above with reference to preferred embodiments, those skilled in the art may make various modifications and changes to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. You will understand that you can do it.

100: Facial image de-identification and restoration system
110: user terminal
130: Facial image de-identification and restoration device
150: database
210: Processor 230: Memory
250: user input/output unit 270: network input/output unit
310: De-identification unit 330: Anonymization unique identifier generation unit
350: De-anonymization request receiving unit 370: Data restoration unit

Claims (11)

a de-identification unit that objectifies a face image, selects a target image, and anonymizes the target image to create an anonymized image;
an anonymized unique identifier generator that generates an anonymized unique identifier matching the target image;
a de-anonymization request receiving unit that receives a de-anonymization request for the anonymized image and verifies the legitimacy of the de-anonymization request; and
A data restoration unit that restores the anonymized image and generates restored data when the de-anonymization request is justified,
The de-identification unit,
a non-identifying encoder including a personal identity encoder for generating a personal identity latent vector and an attribute information encoder for generating an attribute information latent vector; and
a de-identifying decoder that generates a random vector matching the anonymized unique identifier,
The non-identification decoder is,
Calculating a summed latent vector including the personal identity latent vector and the attribute information latent vector, and projecting the random vector onto the summed latent vector to generate a noise vector that is linearly independent from the summed latent vector.
Facial image de-identification and restoration device. delete delete According to paragraph 1,
The non-identification decoder is,
Facial image de-identification and restoration device, characterized in that the random vector is projected onto the personal identity latent vector to generate a noise vector that is linearly independent from the personal identity latent vector. According to paragraph 1,
The non-identification decoder is,
A facial image de-identification and restoration device characterized by generating the noise vector according to [Equation 1].
[Equation 1]

(From here, is the noise vector, is a random vector, is the summed potential vector, Is , is an anonymized unique identifier.) According to paragraph 1,
The de-identification unit,
A facial image de-identification and restoration device characterized in that it generates the anonymized image based on [Equation 2] according to a preset de-identification level.
[Equation 2]

(From here, is the modulation vector, is the summed potential vector, is the noise vector, is the de-identification level, is the similarity parameter.) According to clause 6,
The de-identification unit,
Facial image de-identification and restoration device, characterized in that the anonymized image is generated by combining the modulation vector generated according to Equation 2 and the attribute information latent vector.
According to paragraph 1,
The de-anonymization request receiving unit,
Facial image de-identification and restoration device, characterized in that if the unique identification key included in the de-anonymization request corresponds to the anonymization unique identifier, the de-anonymization request is determined to be legitimate.
According to paragraph 1,
The data recovery unit,
a recovery encoder including a personal identity recovery encoder for generating a non-identifying personal identity latent vector for the anonymized image and an attribute information recovery encoder for generating a non-identifying attribute information latent vector for the anonymized image; and
and a reconstruction decoder that calculates a non-identification sum vector including the non-identification personal identity latent vector and the non-identification attribute information latent vector, and generates the reconstruction data by applying a reconstruction vector to the non-identification sum vector. Facial image de-identification and restoration device.
delete objectifying a face image to select a target image and anonymizing the target image to create an anonymized image;
generating an anonymized unique identifier matching the target image;
Receiving a de-anonymization request for the anonymized image and verifying the legitimacy of the de-anonymization request; and
If the de-anonymization request is justified, restoring the anonymized image to generate restored data,
The step of generating the anonymized image is,
generating a personal identity latent vector and an attribute information latent vector; and
Generating a random vector matching the anonymized unique identifier,
The step of generating the random vector is,
Calculating a summed latent vector including the personal identity latent vector and the attribute information latent vector, and projecting the random vector onto the summed latent vector to generate a noise vector that is linearly independent from the summed latent vector.
Face image de-identification and restoration method.