Face data processing method, device, electronic device and storage medium

Abstract

The embodiments of this specification disclose a facial data processing method, device, electronic device and storage medium: the current facial attack data intercepted by the facial liveness detection model is added to the attack data pool, which contains historical facial attack data; the attack data pool is clustered according to the attack similarity to obtain the attack cluster where the current facial attack data is located; if the attack cluster where the current facial attack data is located reaches the attack count threshold, the current facial attack data is added to the attack blacklist to form a new facial attack record in the attack blacklist.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, device, electronic device and storage medium for processing face data. Background Art

[0002] In the eKYC (electronic Know-Your-Customer) authentication scenario or face-swiping login / payment scenario, it is necessary not only to verify whether the collected face and the comparison source belong to the same face, but also to perform face liveness detection on the collected face to determine whether the collected face is a face attack data (such as a photocopy, print, or mask) or live face data, in order to ensure the system security in the eKYC scenario and face-swiping login / payment scenario. The interception of face attack data has been evolving and improving in the constant confrontation between attack and defense. Summary of the invention

[0003] The embodiments of this specification provide a facial data processing method, device, electronic device and storage medium to more effectively intercept facial attack data.

[0004] In a first aspect, an embodiment of the present specification provides a facial data processing method, comprising: adding current facial attack data intercepted by a facial liveness detection model to an attack data pool, wherein the attack data pool contains historical facial attack data; clustering the attack data pool according to attack similarity to obtain an attack cluster where the current facial attack data is located; if the attack cluster where the current facial attack data is located reaches an attack count threshold, adding the current facial attack data to an attack blacklist to form a new facial attack record in the attack blacklist.

[0005] In a second aspect, an embodiment of the present specification provides a facial data processing method, comprising: performing facial liveness detection on the collected current pen facial data through a facial liveness detection model; if it is detected that the current pen facial data is pen facial liveness data, then matching the current pen facial data with historical facial attack data for attack similarity; if historical facial attack data with a similarity less than a preset attack similarity threshold between the current pen facial data is matched, then setting the current pen facial data as new facial attack data; and intercepting the new facial attack data.

[0006] In a third aspect, the embodiment of the present specification provides a face data processing device, including: an attack adding unit, used to add the current pen face attack data intercepted by the face liveness detection model to the attack data pool, wherein the attack data pool contains historical face attack data; an attack clustering unit, used to cluster the attack data pool according to attack similarity to obtain the attack cluster where the current pen face attack data is located;

[0007] The blacklist adding unit is used to add the current face attack data to the attack blacklist if the attack cluster where the current face attack data is located reaches the attack count threshold, so as to form a new face attack record in the attack blacklist.

[0008] In a fourth aspect, an embodiment of the present specification provides a face data processing device, comprising: an algorithm detection unit, configured to perform face liveness detection on the collected current pen face data through a face liveness detection model; an attack matching unit, configured to match the attack similarity of the current pen face data with historical face attack data if it is detected that the current pen face data is pen face liveness data; a matching result processing unit, configured to set the current pen face data as new face attack data if historical face attack data with a similarity less than a preset attack similarity threshold between the current pen face data is matched; and an attack interception unit, configured to intercept the new face attack data.

[0009] In a fifth aspect, an embodiment of the present specification provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect, or the steps of the method described in the second aspect.

[0010] In a sixth aspect, an embodiment of the present specification provides an electronic device comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors to execute operating instructions contained in the one or more programs for performing the method described in the first aspect or the method described in the second aspect.

[0011] The above one or more technical solutions in the embodiments of this specification have at least the following technical effects:

[0012] By adding the face attack data intercepted by the face liveness detection model to the attack data pool, face attack data is continuously accumulated; and the attack data pool is clustered according to the attack similarity. If the attack cluster where the current face attack data is located reaches the attack number threshold, the current face attack data is added to the attack blacklist, thereby combining historical attack data to mine the same attack to form a face attack record in the attack blacklist. Then, the formed attack blacklist can be used to intercept face attack data, instead of relying solely on the face liveness detection algorithm to intercept face attacks, thereby improving the ability to intercept face attacks and achieving more effective interception of face attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of this specification, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0014] Figure 1 A schematic diagram of a scenario of a face data processing method provided in an embodiment of this specification;

[0015] Figure 2 A flowchart of a method for processing face data provided in accordance with the first aspect of the embodiments of this specification;

[0016] Figure 3 A flowchart of a method for processing face data provided in accordance with the second aspect of the embodiments of this specification;

[0017] Figure 4 A functional module diagram of a face data processing device provided in the third aspect of the embodiments of this specification;

[0018] Figure 5 A functional module diagram of a face data processing device provided in the fourth aspect of the embodiments of this specification;

[0019] Figure 6 A schematic diagram of an electronic device provided in accordance with the fifth aspect of the embodiments of this specification. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the embodiments of this specification more clear, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0021] refer to Figure 1 As shown, Figure 1 The scenario diagram of the face data processing method provided in the embodiment of this specification includes: a client 10, and a function provided by the client 10 to the user to collect face data, so that the client 10 collects face data and sends the collected face data to the server 20.

[0022] The server 20 performs face liveness detection on the received face data based on the face liveness detection model to determine whether the face data is face attack data or live face data from the algorithm level. If it is face attack data, it is intercepted, and if it is live face data, it is let go, so as to intercept each face attack data through the face liveness detection model. The current face attack data intercepted by the face liveness detection model is added to the attack data pool to accumulate face attack data in the attack data pool; the attack data pool is clustered according to the attack similarity to obtain the attack cluster where the current face attack data is located; if the attack cluster where the current face attack data is located reaches the attack count threshold, the current face attack data is added to the attack blacklist to form a new face attack record in the attack blacklist, so as to mine the same attack in combination with the historical face attack data, and then produce the attack blacklist.

[0023] For live face data missed by the live face detection model, the server 20 performs attack similarity matching through historical attack data, and effectively retrieves face attack data missed by the live face detection model.

[0024] The main implementation principles, specific implementation methods and corresponding beneficial effects of the technical solutions of the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0025] On the first aspect, the embodiments of this specification provide a facial data processing method that can be applied to any business scenario that requires face scanning, for example, it can be applied to business scenarios such as face scanning login, face scanning payment, eKYC authentication, etc.

[0026] Combination Figure 1 and Figure 2 As shown, Figure 2 This is a flowchart of the face data processing method provided in the embodiment of this specification. The client collects face data and uploads the collected current face data to the server. The server performs face liveness detection on the received current face data through the following steps to more effectively detect and intercept face attack data:

[0027] S200: adding the current pen face attack data intercepted by the face liveness detection model to the attack data pool.

[0028] It should be noted that if the face data processing method provided in the embodiment of this specification has been running online for a period of time, the attack data pool includes not only the current face attack data newly added to the attack data pool, but also multiple historical face attack data that have been added to the attack data pool before. That is, the face attack data intercepted by the face liveness detection model is continuously added to the attack data pool to continuously enrich the historical face attack data in the attack data pool.

[0029] In the specific application process, after the client starts face scanning, facial data will be collected to obtain the current pen facial data. In order to ensure the accuracy of face liveness detection performed by the server on the current pen facial data, the current pen facial data collected by the client should meet the pre-set image requirements, such as: only containing a single face, and the facial angle, clarity, jitter, etc. meet the pre-set image requirements.

[0030] Specifically, the face liveness detection model is used to perform face liveness detection on the current pen face data collected, so as to determine from the algorithm level whether the current pen face data is face attack data or live face data.

[0031] Since the face liveness detection model is trained with a large number of attack samples, the trained face liveness detection model can only recall face attack data similar to the training samples. Therefore, some face attack data (such as injection attacks) that are very different from the attack samples used to train the face liveness detection model cannot be successfully intercepted by the face liveness detection model. The technical solution provided in the embodiments of this specification is to make up for the deficiency of the face liveness detection model in intercepting face attack data, so as to achieve more effective interception of face attack data and improve the system security level in the corresponding business scenario.

[0032] In specific implementation, the face liveness detection model can be a silent liveness algorithm, a colorful and random multi-action face liveness detection algorithm, etc. All of them set a certain pass rate to determine the degree of interception of attacks.

[0033] Specifically, different processing methods are adopted according to different detection results of the face liveness detection model for the current face data, which are described in detail below:

[0034] 1. If the face liveness detection model detects that the current pen face data is face attack data, the current pen face data is intercepted, thereby obtaining the face attack data of the current pen intercepted by the face liveness detection model. Then, for the current pen face data intercepted by the face liveness detection model, the following steps S202 to S204 are executed:

[0035] S202: Cluster the attack data pool according to attack similarity to obtain the attack cluster where the current pen face attack data is located.

[0036] Specifically, step S202 includes: extracting attack similarity features of face attack data in the attack data pool (including current face attack data and each historical face attack data); clustering the attack data pool according to attack similarity to obtain the attack cluster where the current face attack data is located.

[0037] Among them, the attack similarity features include: attack face features and attack medium features. Based on this, the implementation process of step S202 includes: extracting the attack medium features and attack face features of the face attack data in the attack data pool; clustering the attack data pool according to the attack medium features of the face attack data in the attack data pool to obtain attack clusters of attacks with the same medium, and clustering the attack data pool according to the attack face features of the face attack data in the attack data pool to obtain attack clusters of attacks with the same face; and determining the attack cluster where the current pen face attack data is located from the clustered attack clusters.

[0038] It should be noted that the attack cluster of the same-medium attack refers to: multiple attacks using the same attack medium, such as multiple attacks using the same display, the same mask, or the same mobile phone. In specific implementation, the pooling layer features of the silent live model can be used, but are not limited to, to describe the attack medium features of the same-medium attack. Because the training images used by the silent live model are full-image images, and the silent live model is a two-classification model, the pooling layer features of the silent live model can represent the background information of the image, so the pooling layer features of the silent live model can well describe the attack medium features. The attack cluster of the same-face attack refers to: multiple attacks using the same face. Therefore, the extracted facial features can well describe the attacking facial features.

[0039] Specifically, the attack medium features of the current face data and each historical face attack data in the attack data pool can be extracted through the FC (fully connected layers) layer in the CNN for face-anti-spoofing model, but are not limited to the FC (fully connected layers) layer in the CNN for face-anti-spoofing model. The CNN for face-anti-spoofing model is a face liveness detection model based on CNN (convolutional neural network).

[0040] Specifically, the attacking face features of the current face data and each historical face attack data in the attack data pool can be extracted through the FC layer in the Tensorflow-implementation-of-LCNN model, but are not limited to the FC layer in the Tensorflow-implementation-of-LCNN model. The Tensorflow-implementation-of-LCNN model is a LCNN (Lookup-based Convolutional Neural Network) implemented based on TensorFlow, and TensorFlow is an open source deep learning framework.

[0041] S204: If the attack cluster where the current facial attack data is located reaches the attack count threshold, the current facial attack data is added to the attack blacklist to form a new facial attack record in the attack blacklist.

[0042] Specifically, the face attack records in the attack blacklist are based on the account dimension, and the attacking face images are saved in the attack blacklist. For example, for the current face attack data, the account information and face image that initiated the current face attack data are associated and added to the attack blacklist to form a new face attack record.

[0043] After obtaining a new face attack record in the attack blacklist through the above steps S200 to S204, potential risk mining will be performed in any one or more of the following implementation modes:

[0044] Implementation method 1: Directly use the attack blacklist to intercept face attacks:

[0045] For any facial data collected, before performing face liveness detection on the facial data through the face liveness detection model, determine whether the facial data belongs to face attack data according to the attack blacklist; if yes, intercept the facial data, and no longer perform face liveness detection on the intercepted facial data, or perform face liveness detection on the intercepted facial data according to a new face liveness detection model; if not, perform face liveness detection on the collected facial data based on the original face liveness detection model, wherein the new face liveness detection model is better than the original face liveness detection model.

[0046] Implementation method 2: Mining out potential risk accounts based on the attack blacklist, for example, mining out risk accounts with gang attack behaviors. For example, it is possible to mine out known accounts that have passed eKYC verification but have multiple attack behaviors in history. Specifically, the new face attack record in the attack blacklist is associated with the known account; the known account that is successfully associated with the new face attack record is regarded as a potential risk account.

[0047] In the specific implementation, any dimensional information existing in the face attack record in the attack blacklist can be used to associate with known accounts, such as face information, contact number, address information, etc.

[0048] Specifically, new face attack records can be associated with known accounts based on facial information, that is, for each known account, the face image in the new face attack record is compared with the face image of the user under the known account; if they are the same, the known account is regarded as a potential risk account. If the face attack record in the attack blacklist also contains information of other dimensions, such as communication number, address information, etc., the new face attack record can also be associated with the known account based on the communication number and address information. Take the communication number as an example: for each known account, the communication number in the new face attack record is compared with the communication number under the known account for face comparison; if they are the same, the known account is regarded as a potential risk account.

[0049] In an optional implementation, in order to more reasonably mine potential risk accounts, for new face attack records added to the attack blacklist, check whether the attacking face images in the new face attack records are stolen face images; if so, associate the new face attack records with known accounts, and treat the known accounts successfully associated with the new face attack records as potential risk accounts; if not, no association is performed. By reviewing whether the attacking face images in the new face attack records are used, it is avoided that users use their own face photos (such as: face pictures printed on color paper, face digital images on electronic device screens) to scan faces and other low-risk situations to mine potential risk accounts.

[0050] In an optional implementation, in order to more rigorously mine potential risk accounts, after adding the current face attack data to the attack blacklist to form a new face attack record in the attack blacklist, it is determined whether the historical face attack records in the attack blacklist that belong to the same attack as the new face attack record reach a preset number of records; if so, the new face attack record is associated with a known account, and the known account successfully associated with the new face attack record is used as a potential risk account; otherwise, the attack count threshold of the attack cluster where the current face attack data is located is increased.

[0051] Therefore, only when a certain number of identical attack records are reached in the attack blacklist, the new face attack record is mined for potential risk accounts. Take the preset number of records as 1 as an example: if there is a historical face attack record in the attack blacklist that belongs to the same attack as the new face attack record, it triggers the new face attack record to be associated with the known account. For the specific association processing, please refer to the previous description, and for the sake of brevity, it will not be repeated here.

[0052] Furthermore, for the discovered potential risk accounts, risk-related processing can be performed in any of the following ways to improve the safety level of the system:

[0053] Specifically, risk-related processing for potential risk accounts can be any of the following: freezing potential risk accounts, prohibiting potential risk accounts from starting face scanning, or setting more complex face scanning action verification when potential risk accounts are performing face scanning, such as adding face scanning action verification such as shaking head, blinking, nodding, etc. Of course, after mining potential risk accounts, other dimensions of risk screening can be conducted on potential risk accounts. For example, in the eKYC scenario, it can be checked whether there is a risk of fake ID in the potential risk account, and whether there are multiple attempts to scan the face within a preset time. If there are risks in other dimensions, freeze the potential risk account, prohibit the potential risk account from starting face scanning, or set more complex face scanning action verification when the potential risk account is performing face scanning.

[0054] Second, if the current facial data is detected as live facial data, it means that the face liveness detection model has missed the current facial data. However, assuming that the user has attacked multiple times before, the attacker may have mastered some attack techniques, and an "advanced" attack will occur, which greatly increases the probability of breaking the current facial attack data. In a certain facial data, due to some special operations (such as dark light environment, a certain degree of blur, a certain degree of tilt, etc.), the liveness algorithm is deceived. Therefore, the face liveness detection model may miss the attack.

[0055] In a specific embodiment of this specification, in order to effectively retrieve face attack data that is "missed" by the liveness algorithm, the following implementation process is provided:

[0056] If the face liveness detection model detects that the current pen face data is face liveness data, the current pen face data will be matched with the historical face attack data in the attack data pool for attack similarity; if it is matched that there is historical face attack data in the attack data pool whose similarity with the current pen face data is less than the preset attack similarity threshold, the current pen face data will be set as new face attack data and the new face attack data will be intercepted; if there is no historical face attack data in the attack data pool whose similarity with the current pen face data is less than the preset attack similarity threshold, the current pen face data will not be intercepted.

[0057] Furthermore, in order to reduce the amount of calculation, if the face liveness detection model detects that the current pen face data is face liveness data, it is determined whether the current pen face data is in a high-risk attack range; if so, the current pen face data is matched with the historical face attack data in the attack data pool for attack similarity; otherwise, the current pen face data is directly passed.

[0058] Specifically, the high-risk attack interval is a similarity value interval set according to the preset attack similarity threshold as the lower limit. Therefore, if the current pen face data is far different from the historical face attack data in the attack data pool, it will not fall into the high-risk attack interval, and will not be matched and retrieved in the attack data pool, reducing the matching retrieval amount.

[0059] More specifically, the current facial data is matched with the historical facial attack data in the attack data pool in terms of attack similarity, specifically: the attack similarity features of each historical facial attack data in the attack data pool and the attack similarity features of the current facial data are extracted; for each historical facial attack data in the attack data pool, similarity matching is performed based on the attack similarity features of the current facial data and the attack similarity features extracted from the historical facial attack data.

[0060] The attack similarity features extracted from the current facial data and each historical facial attack data in the attack data pool specifically include: attack medium features and attack face features.

[0061] Specifically, the attack medium features of the current face data and each historical face attack data in the attack data pool can be extracted through the FC (fully connected layers) layer in the CNN for face-anti-spoofing model, but are not limited to, to compare the attack medium similarity.

[0062] Specifically, the attacking face features of the current face data and each historical face attack data in the attack data pool can be extracted through the FC layer in the Tensorflow-implementation-of-LCNN model, but are not limited to, to compare the attacking face similarities.

[0063] By further mining the face attack data that was missed by the face liveness detection model, the attacks that were "missed" by the face liveness detection model can be effectively retrieved, thereby improving the security level of the system. It is often not enough to rely solely on the decision results of the visual algorithm to ensure the security of the system. The embodiments of this specification combine risk control and make comprehensive decisions with the visual algorithm through other dimensions to achieve face liveness detection.

[0064] In a second aspect, the present specification provides a method for processing face data, referring to Figure 3 As shown, the face data processing method includes the following steps:

[0065] S300, performing face liveness detection on the collected current face data through a face liveness detection model;

[0066] S302: if it is detected that the current pen face data is pen face liveness data, then the current pen face data is matched with the historical face attack data for attack similarity;

[0067] S304: If the historical face attack data with a similarity less than a preset attack similarity threshold between the current face data and the historical face attack data is matched, the current face data is set as the new face attack data;

[0068] S306: Intercept new face attack data.

[0069] For specific implementation details in the embodiment of the face data processing method described in the second aspect, reference can be made to the embodiment of the face data processing method described in the first aspect above, and for the sake of brevity of the specification, they will not be repeated here.

[0070] In a third aspect, the present invention provides a face data processing device, referring to Figure 4 As shown, including:

[0071] An attack adding unit 401 is used to add the current face attack data intercepted by the face liveness detection model to the attack data pool, which includes historical face attack data;

[0072] An attack clustering unit 402 is used to cluster the attack data pool according to the attack similarity to obtain the attack cluster where the current pen face attack data is located;

[0073] The adding blacklist unit 403 is used to add the current face attack data to the attack blacklist if the attack cluster where the current face attack data is located reaches the attack count threshold, so as to form a new face attack record in the attack blacklist.

[0074] In an optional implementation manner, the attack clustering unit 402 includes:

[0075] A first attack feature extraction subunit is used to extract attack medium features and attack face features of face attack data in the attack data pool;

[0076] The first clustering subunit is used to cluster the attack similarity of the attack data pool according to the attack medium characteristics of the face attack data in the attack data pool to obtain attack clusters of the same medium attack, and

[0077] The second clustering subunit is used to cluster the attack data pool according to the attack face features of the face attack data in the attack data pool, and obtain attack clusters of the same face attack;

[0078] The attack cluster determination subunit is used to determine the attack cluster where the current pen face attack data is located from the attack clusters obtained by clustering.

[0079] In an optional implementation manner, the face data processing device further includes:

[0080] An algorithm detection unit, used for performing face liveness detection on the collected current pen face data based on the face liveness detection model before adding the current pen face attack data intercepted by the face liveness detection model into the attack data pool;

[0081] The first attack interception unit is used to intercept the current pen face data if the face liveness detection model detects that the current pen face data is pen face attack data.

[0082] In an optional implementation manner, the face data processing device further includes:

[0083] An attack matching unit, for matching the current pen face data with the historical face attack data in the attack data pool for attack similarity if the face liveness detection model detects that the current pen face data is pen face liveness data;

[0084] A matching result processing unit, configured to set the current pen face data as new face attack data if the historical face attack data with a similarity less than a preset attack similarity threshold between the current pen face data and the matched attack data pool exists;

[0085] The second attack interception unit is used to intercept new face attack data.

[0086] In an optional implementation manner, the attack matching unit includes:

[0087] The second attack feature extraction subunit is used to extract the attack similarity feature of each historical face attack data in the attack data pool, and extract the attack similarity feature of the current face data;

[0088] The attack matching subunit is used to perform similarity matching on each piece of historical face attack data in the attack data pool according to the attack similarity feature of the current piece of face data and the attack similarity feature of the historical face attack data.

[0089] In an optional implementation manner, the face data processing device further includes:

[0090] An attack judgment unit is used to judge whether any face data collected belongs to face attack data according to the attack blacklist before performing face liveness detection on the face data through the face liveness detection model;

[0091] The third attack interception unit is used to intercept the facial data if the judgment result is yes; otherwise, call the algorithm detection unit to perform facial liveness detection on the facial data through the facial liveness detection model.

[0092] In an optional implementation manner, the face data processing device further includes:

[0093] A first association processing unit is used to associate the new face attack record with the known account, and to treat the known account successfully associated with the new face attack record as a potential risk account, or to review whether the attacking face image in the new face attack record is a stolen face image, and if so, to associate the new face attack record with the known account, and to treat the known account successfully associated with the new face attack record as a potential risk account;

[0094] The first risk processing unit is used to perform risk-related processing on potential risk accounts.

[0095] In an optional implementation manner, the face data processing device further includes:

[0096] An attack record retrieval unit, used to retrieve the historical face attack records in the attack blacklist that belong to the same attack as the new face attack record to see whether the number of records reaches a preset number;

[0097] A first association processing unit, configured to associate a new face attack record with a known account if a preset number of records is reached;

[0098] The second risk processing unit is used to treat the known account successfully associated with the new face attack record as a potential risk account; otherwise, increase the attack count threshold of the attack cluster where the current face attack data is located.

[0099] Regarding the above-mentioned device, the specific functions of each unit therein have been described in detail in the embodiment of the face data processing method provided in the first aspect of the embodiment of the present invention, and will not be elaborated here. The specific implementation process can refer to the embodiment of the face data processing method provided in the first aspect above.

[0100] In a fourth aspect, the embodiments of this specification provide a face data processing device, referring to Figure 5 As shown, including:

[0101] The algorithm detection unit 501 is used to perform face liveness detection on the collected current pen face data through a face liveness detection model;

[0102] The attack matching unit 502 is used to match the current pen face data with the historical face attack data for attack similarity if it is detected that the current pen face data is pen face live data;

[0103] The matching result processing unit 503 is used to set the current pen face data as new face attack data if the historical face attack data with a similarity less than a preset attack similarity threshold between the historical face attack data and the current pen face data is matched;

[0104] The attack interception unit 504 is used to intercept new face attack data.

[0105] Regarding the above-mentioned device, the specific functions of each unit therein have been described in detail in the embodiment of the face data processing method provided in the second aspect of the embodiment of the present invention, and will not be elaborated here. The specific implementation process can refer to the embodiment of the face data processing method provided in the second aspect above.

[0106] In a fifth aspect, based on the same inventive concept as the aforementioned face data processing method embodiment, the embodiment of this specification also provides an electronic device, such as Figure 6 As shown, it includes a memory 604, a processor 602, and a computer program stored in the memory 604 and executable on the processor 602. When the processor 602 executes the program, the steps of the aforementioned face data processing method embodiment are implemented.

[0107] Among them, Figure 6 In the embodiment of the present invention, a bus architecture (represented by bus 600) is shown, which may include any number of interconnected buses and bridges, and bus 600 links various circuits including one or more processors represented by processor 602 and memory represented by memory 604. Bus 600 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 606 provides an interface between bus 600 and receiver 601 and transmitter 603. Receiver 601 and transmitter 603 may be the same element, namely a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 602 is responsible for managing bus 600 and general processing, while memory 604 may be used to store data used by processor 602 when performing operations.

[0108] In the sixth aspect, based on the same inventive concept as the aforementioned face data processing method embodiment, the embodiment of this specification also provides a computer-readable storage medium, on which a computer program is stored, which implements the steps of the aforementioned face data processing method embodiment when executed by a processor.

[0109] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that has the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0112] It should be understood that the present embodiment is not limited to the precise structure described above and shown in the drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the present embodiment is limited only by the appended claims.

[0113] The above description is only a preferred embodiment of the present embodiment and is not intended to limit the present embodiment. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present embodiment should be included in the protection scope of the present embodiment.

Claims

1. A method for processing face data, comprising: For any facial data collected, determine whether it belongs to facial attack data based on the attack blacklist; If yes, intercept the facial data; Otherwise, perform face liveness detection on the collected current pen face data based on the face liveness detection model; If the face liveness detection model detects that the current pen face data is pen face attack data, intercepting the current pen face data; If the face liveness detection model detects that the current pen face data is pen face liveness data, the current pen face data is matched with the historical face attack data in the attack data pool for attack similarity; if the attack data pool matches the historical face attack data with a similarity less than a preset attack similarity threshold between the current pen face data, the current pen face data is set as new face attack data; and the new face attack data is intercepted; The current face attack data intercepted by the face liveness detection model is added to the attack data pool, which contains historical face attack data; the attack data pool is clustered according to attack similarity to obtain the attack cluster where the current face attack data is located; if the attack cluster where the current face attack data is located reaches the attack count threshold, the current face attack data is added to the attack blacklist to form a new face attack record in the attack blacklist.

2. The method according to claim 1, wherein clustering the attack data pool according to attack similarity to obtain the attack cluster where the current pen face attack data is located comprises: Extracting attack medium features and attack face features of face attack data in the attack data pool; performing attack similarity clustering on the attack data pool according to the attack medium characteristics of the face attack data in the attack data pool to obtain attack clusters of attacks with the same medium, and Performing attack similarity clustering on the attack data pool according to attack face features of face attack data in the attack data pool to obtain attack clusters of the same face attack; The attack cluster where the current pen face attack data is located is determined from the attack clusters obtained by clustering.

3. The method according to claim 1, wherein the attack similarity matching of the current facial data with the historical facial attack data in the attack data pool comprises: Extracting attack similarity features of each piece of historical face attack data in the attack data pool, and extracting attack similarity features of the current piece of face data; For each piece of historical face attack data in the attack data pool, similarity matching is performed based on the attack similarity feature of the current piece of face data and the attack similarity feature of the historical face attack data.

4. The method according to claim 1, after adding the current face attack data to the attack blacklist to form a new face attack record in the attack blacklist, further comprises: Associating the new face attack record with a known account, and treating the known account successfully associated with the new face attack record as a potential risk account, or Review whether the attacking face image in the new face attack record is a stolen face image. If so, associate the new face attack record with a known account, and treat the known account successfully associated with the new face attack record as a potential risk account; Perform risk-related processing on the potential risk accounts.

5. The method according to claim 1, after adding the current face attack data to the attack blacklist to form a new face attack record in the attack blacklist, further comprises: Determine whether the number of historical face attack records in the attack blacklist that belong to the same attack as the new face attack record reaches a preset number of records; If yes, the new face attack record is associated with a known account, and the known account successfully associated with the new face attack record is regarded as a potential risk account; Otherwise, the attack count threshold of the attack cluster where the current facial attack data is located is increased.

6. A facial data processing device, comprising: An attack judgment unit, used for judging whether the current facial data collected belongs to facial attack data according to the attack blacklist; A third attack interception unit, configured to intercept the facial data if the judgment result is yes; Otherwise, perform face liveness detection on the collected current pen face data based on the face liveness detection model; A first attack interception unit, configured to intercept the current pen face data if the face liveness detection model detects that the current pen face data is pen face attack data; an attack matching unit, configured to perform attack similarity matching between the current pen face data and the historical face attack data in the attack data pool if the face liveness detection model detects that the current pen face data is pen face liveness data; a matching result processing unit, configured to set the current facial data as new facial attack data if the historical facial attack data with a similarity less than a preset attack threshold value between the historical facial attack data and the current facial data is matched in the attack data pool; A second attack interception unit, used to intercept the new face attack data; An attack adding unit, used for adding the current pen face attack data intercepted by the face liveness detection model to the attack data pool, wherein the attack data pool contains historical face attack data; An attack clustering unit, used for clustering the attack data pool according to attack similarity to obtain an attack cluster where the current pen face attack data is located; The blacklist adding unit is used to add the current face attack data to the attack blacklist if the attack cluster where the current face attack data is located reaches the attack count threshold, so as to form a new face attack record in the attack blacklist.

7. The apparatus according to claim 6, wherein the attack clustering unit comprises: A first attack feature extraction subunit is used to extract attack medium features and attack face features of face attack data in the attack data pool; a first clustering subunit, configured to cluster the attack data pool according to attack medium characteristics of the face attack data in the attack data pool to obtain attack clusters of attacks with the same medium, and A second clustering subunit is used to perform attack similarity clustering on the attack data pool according to attack face features of the face attack data in the attack data pool to obtain attack clusters of the same face attack; The attack cluster determination subunit is used to determine the attack cluster where the current pen face attack data is located from the attack clusters obtained by clustering.

8. The device according to claim 6, wherein the attack matching unit comprises: A second attack feature extraction subunit is used to extract the attack similarity feature of each historical face attack data in the attack data pool, and to extract the attack similarity feature of the current face data; The attack matching subunit is used to perform similarity matching on each piece of historical face attack data in the attack data pool according to the attack similarity feature of the current piece of face data and the attack similarity feature of the historical face attack data.

9. The apparatus of claim 6, further comprising: A first association processing unit is used to associate the new face attack record with a known account, and use the known account successfully associated with the new face attack record as a potential risk account, or to review whether the attacking face image in the new face attack record is a stolen face image, and if so, associate the new face attack record with a known account, and use the known account successfully associated with the new face attack record as a potential risk account; The first risk processing unit is used to perform risk-related processing on the potential risk account.

10. The apparatus of claim 6, further comprising: An attack record retrieval unit, used to retrieve whether the number of historical face attack records in the attack blacklist that belong to the same attack as the new face attack record reaches a preset number of records; A first association processing unit, configured to associate the new face attack record with a known account if a preset number of records is reached; A second risk processing unit, configured to treat a known account successfully associated with the new face attack record as a potential risk account; Otherwise, the attack count threshold of the attack cluster where the current facial attack data is located is increased.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method as claimed in any one of claims 1 to 5 are implemented.

12. An electronic device, characterized in that: The invention comprises a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors to execute the operation instructions contained in the one or more programs for performing the method as claimed in any one of claims 1 to 5.

Images