A method and device for assisting in liveness detection in a face recognition system

Abstract

A method and device for assisting in live detection of a face recognition system, including: establishing a list of "attack suspects" and initializing it; extracting a face region using a face detection algorithm; calculating the live probability of a single-frame face using a neural network model; calculating face feature values; matching the "attack suspects" according to the face feature values; comparing the live probability of the suspect with a set threshold to adjust the confidence value, so as to determine whether the current suspect is true or false; the confidence of the attack suspect decays over time: traversing all "attack suspects" every interval of time T, with a confidence decay step size; when the confidence is 0, deleting the "attack suspect". The present invention establishes a list of "attack suspects" based on face feature value matching, which can filter out most long-term continuous attacks; dynamically adjusts during the matching process to avoid errors caused by occasional misjudgments, and the judgment result is more stable.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for assisting in live detection for a face recognition system. Background Art

[0002] During the real-time face recognition process, it is necessary to be vigilant against using a fake face for face recognition. Therefore, live detection needs to be performed before face recognition.

[0003] Many existing live detection algorithms are based on single-frame images. The specific approach is as follows: First, face detection is performed on the picture, and then the face area is cropped. Then, the face area is input into the live detection model to obtain the result. Since the result of single-frame image recognition often cannot have a very high accuracy, when an attacker continuously performs a live attack on-site for a long time, there is a certain probability that the detection will be breached at a certain moment. The longer the time, the higher the probability of breaching.

[0004] Therefore, in view of the situation of long-term attacks by attackers, there is an urgent need to propose a new method to solve the risk of long-term attacks. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and device for assisting in live detection for a face recognition system, which can greatly reduce the risk of long-term attacks.

[0006] In a first aspect, the present invention provides a method for assisting in live detection for a face recognition system, specifically including the following steps:

[0007] Step S1: Establish a "suspected attacker" list. Each suspected attacker includes two fields: confidence level P attack and face features, and initialize it. The initial state is empty.

[0008] Step S2: Extract the face area using a face detection algorithm.

[0009] Step S3: Calculate the live probability of a single-frame face using a neural network model, denoted as P live ;

[0010] Step S4: Calculate the face feature value.

[0011] Step S5: Match the "suspected attacker" according to the face feature value.

[0012] Step S6: Update the confidence level P of the "current suspected attacker" attack , and adjust the confidence level P live by comparing the suspect's live probability P live_threshold with a set threshold P attack to determine whether the current suspect is real or fake.

[0013] Step S7: Attack suspect confidence level P attack Decay over time: Traverse all "attack suspects" at regular intervals of time T, and the confidence level P attack - Decay step Δd2 until it reaches 0.0; when the confidence level P attack is 0, delete the "attack suspect"; go to Step S2.

[0014] Furthermore, the said Step S5 specifically includes:

[0015] Step S51: Traverse the "attack suspect" list, compare the similarity of facial features, which is higher than the threshold P live_threshold , and lock it as the "current attack suspect";

[0016] Step S52: If it does not exist, add a new "attack suspect" with the confidence level P attack being 0 and set it as the "current attack suspect".

[0017] Furthermore, the said Step S6 specifically includes:

[0018] Step S61: If the suspect liveness probability P live is less than the threshold P live_threshold , then the suspect confidence level P attack + update step Δd1 until the confidence level P attack reaches 1.0; if the current suspect confidence level P attack [[ID=**************]] attack_threshold , return the result as false;

[0019] Step S62: If the suspect liveness probability P live is greater than or equal to the threshold P live_threshold , then the suspect confidence level P attack - update step Δd1 until the confidence level P attack reaches 0.0; if the current suspect confidence level P attack is less than the threshold P attack_threshold , return the result as true, otherwise as false.

[0020] In a second aspect, the present invention provides a liveness detection assistance device for a face recognition system, including:

[0021] "Attack suspect" list initialization module, used to establish an "attack suspect" list, each attack suspect includes two fields of confidence level P attack and facial features, and initialize them, initially empty;

[0022] Facial region extraction module, used to extract the facial region by using a face detection algorithm; ​

[0023] A single-frame face liveness probability calculation module, which is used to calculate the liveness probability of a single-frame face by using a neural network model, denoted as P live ;

[0024] A face feature value calculation module, which is used to calculate face feature values;

[0025] An "attack suspect" matching module, which is used to match the "attack suspect" according to the face feature values

[0026] The confidence level P of the "current attack suspect" attack An update module, which is used to adjust the confidence level P by comparing the suspect liveness probability P live with the set threshold P live_threshold to determine whether the current suspect is true or false; attack by adjusting the value of P

[0027] The confidence level P of the attack suspect attack A confidence level P decay module over time, which is used to traverse all "attack suspects" at intervals of time T, and the confidence level P attack -0.1 until it reaches 0.0; when the confidence level P attack is 0, delete the "attack suspect".

[0028] Furthermore, the "attack suspect" matching module specifically includes:

[0029] Traverse the "attack suspect" list, compare the similarity of face features, and if it is higher than the threshold P live_threshold , lock it as the "current attack suspect";

[0030] If it does not exist, add a new "attack suspect" with the confidence level P attack being 0 and set it as the "current attack suspect".

[0031] Furthermore, the confidence level P update module of the "current attack suspect" attack specifically includes:

[0032] If the suspect liveness probability P live is less than the threshold P live_threshold , then the suspect confidence level P attack +0.05 until the confidence level P attack reaches 1.0; if the current suspect confidence level P attack is greater than the threshold P attack_threshold , return the result as false;

[0033] If the suspect liveness probability P live is greater than or equal to the threshold P live_threshold , then the suspect confidence level P attack-0.05 until the confidence level P attack up to 0.0; if the current suspect confidence level P attack is less than the threshold P attack_threshold , return the result as true, otherwise false.

[0034] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in the first aspect is implemented.

[0035] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method described in the first aspect is implemented.

[0036] The advantages of the present invention are as follows:

[0037] 1. Based on face feature value matching, the present invention establishes a "suspected attacker" list, which can filter out most long-term continuous attacks.

[0038] 2. The present invention sets a confidence level for the "suspected attacker", which is not simply black or white. It is dynamically adjusted during the matching process to avoid errors caused by occasional misjudgments, and the judgment result is more stable.

[0039] 3. The present invention clears the "suspected attacker" list based on time decay, avoiding additional prevention costs in the case of no attack. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present invention will be further described below with reference to the accompanying drawings in conjunction with embodiments.

[0041] Figure 1 is a schematic flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Embodiment 1

[0043] As Figure 1 shown, a method for assisting in live detection of a face recognition system specifically includes the following steps:

[0044] Step S1: Establish a "suspected attacker" list. Each suspected attacker includes two fields, namely, a confidence level P attack and a face feature, and is initialized to be empty;

[0045] Step S2: Use the MTCNN face detection algorithm to extract the face region;

[0046] Step S3: Use a neural network model to calculate the live probability of a single-frame face, denoted as P live ;

[0047] Step S4: Calculate the face feature value;

[0048] Step S5: Match the "attack suspect" according to the face feature value. The specific matching process includes:

[0049] Step S51: Traverse the "attack suspect" list, compare the similarity of face features, and if it is higher than the threshold P live_threshold , lock it as the "current attack suspect";

[0050] Step S52: If not, add a new "attack suspect" with a confidence level P attack of 0 and set it as the "current attack suspect";

[0051] Step S6: Update the confidence level P of the "current attack suspect" attack , specifically including:

[0052] Step S61: If the live probability P of the suspect live is less than the threshold P live_threshold , then the suspect confidence level P attack +0.05 (update step size Δd1 = 0.05, but not limited to this value) until the confidence level P attack reaches 1.0; if the current suspect confidence level P attack is greater than the threshold P attack_threshold , return the result as false;

[0053] Step S62: If the live probability P of the suspect live is greater than or equal to the threshold P live_threshold , then the suspect confidence level P attack -0.05 until the confidence level P attack reaches 0.0; if the current suspect confidence level P attack is less than the threshold P attack_threshold , return the result as true, otherwise as false;

[0054] Step S7: The confidence level P of the attack suspect attack decays with time: Traverse all "attack suspects" at regular intervals (for example: one minute), and the confidence level P attack -0.1 (decay step size Δd2 = 0.1, but not limited to this value) until it reaches 0.0; when the confidence level P attack is 0, delete this "attack suspect"; go to Step S2.

[0055] Based on the same inventive concept, the present application also provides a device corresponding to the method in Embodiment 1. For details, see Embodiment 2.

[0056] Embodiment 2

[0057] In this embodiment, a living body detection assistance device for a face recognition system is provided, including:

[0058] An "attack suspect" list initialization module, which is used to establish an "attack suspect" list. Each attack suspect contains two fields: confidence level P attack and face features, and is initialized to be empty;

[0059] A face region extraction module, which is used to extract the face region by using a face detection algorithm;

[0060] A living body probability calculation module for a single-frame face, which is used to calculate the living body probability of a single-frame face by using a neural network model, denoted as P live ;

[0061] A face feature value calculation module, which is used to calculate the face feature value;

[0062] An "attack suspect" matching module, which is used to match the "attack suspect" according to the face feature value. The specific matching process includes:

[0063] Traverse the "attack suspect" list, compare the similarity of face features, and if it is higher than the threshold P live_threshold , lock it as the "current attack suspect";

[0064] If it does not exist, add a new "attack suspect" with a confidence level P attack of 0 and set it as the "current attack suspect";

[0065] The confidence level P of the "current attack suspect" attack Update module, specifically including:

[0066] If the living body probability P of the suspect live is less than the threshold P live_threshold , then the confidence level P of the suspect attack +0.05 (the update step size Δd1 = 0.05, but not limited to this value), until the confidence level P attack reaches 1.0; if the current suspect's confidence level P attack is greater than the threshold P attack_threshold , return the result as false;

[0067] If the living body probability P of the suspect live is greater than or equal to the threshold P live_threshold , then the confidence level P of the suspect attack -0.05, until the confidence level P attack reaches 0.0; if the current suspect's confidence level P attack is less than the threshold P attack_threshold , return the result as true, otherwise as false;

[0068] Attack suspect confidence level P attack A time decay module for traversing all "attack suspects" at intervals of time T to obtain the confidence level P attack -0.1 (the decay step Δd2 = 0.1, but not limited to this value), until it reaches 0.0; when the confidence level P attack When the confidence level is 0, delete this "attack suspect".

[0069] Since the device introduced in the second embodiment of the present invention is the device used to implement the method of the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, those skilled in the art can understand the specific structure and variations of this device, so it will not be elaborated here. Any device used to implement the method of the first embodiment of the present invention falls within the scope of protection of the present invention.

[0070] Based on the same inventive concept, this application provides an electronic device embodiment corresponding to the first embodiment, as detailed in the third embodiment.

[0071] Embodiment Three

[0072] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, any implementation manner in the first embodiment can be realized.

[0073] Since the electronic device introduced in this embodiment is the device used to implement the method in the first embodiment of this application, based on the method introduced in the first embodiment of this application, those skilled in the art can understand the specific implementation manner and various variations of the electronic device in this embodiment. Therefore, how this electronic device realizes the method in the embodiments of this application will not be introduced in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application falls within the scope of protection of this application.

[0074] Based on the same inventive concept, this application provides a storage medium corresponding to the first embodiment, as detailed in the fourth embodiment.

[0075] Embodiment Four

[0076] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, any implementation manner in the first embodiment can be realized.

[0077] The main idea of the present invention is based on the continuity of deepfake attacks on the time axis. In actual scenarios, deepfake attacks usually continuously attack for a period of time. If there are continuous attacks within a period of time, the probability of continued attacks next is very high. At this time, the attack can be temporarily blocked to greatly increase the cost of deepfake attacks.

[0078] The key point of the present invention is to design a list of "attack suspects", and based on the similarity of face recognition feature values, recent attackers are screened. If a face feature appears in false face attacks multiple times, the face feature value will be recorded, simply referred to as an "attack suspect". The list stores the face feature values and confidence levels that may be false faces. The result of liveness detection needs to be compared with the list of "attack suspects", and only after it is determined that it is not an "attack suspect" can the next face recognition be carried out.

[0079] The probability of an "attack suspect" changes with the environment: when a single-frame face liveness judgment is false, the confidence level of this "attack suspect" increases; when a single-frame face recognition judgment is true, it still needs to be compared with the list of "attack suspects". If there is a list of "attack suspects" and the confidence level is higher than the threshold, the liveness fails. At the same time, the confidence level of this "attack suspect" will also decrease. When it drops to a certain level, the "attack suspect" is deleted. The confidence level of an "attack suspect" decays over time. When it is not triggered for a certain period of time, the "attack suspect" is deleted.

[0080] The above are only the preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for assisting in liveness detection of a face recognition system, characterized in that: It includes the following steps: Step S1: Establish a list of "attack suspects", where each attack suspect contains two fields: confidence level P attack and facial features, and initialize it. Initially, it is empty. Step S2: Extract the face region using a face detection algorithm; Step S3: Calculate the liveness probability of a single-frame face using a neural network model, denoted as P live ; Step S4: Calculate the face feature values; Step S5: Match the "attack suspect" according to the face feature values; Step S6: Update the confidence level P of the "current attack suspect" attack , based on the suspect's liveness probability P live and compare it with the set threshold P live_threshold to adjust the confidence level P attack value, thereby determining whether the current suspect is true or false; Step S7: Confidence level P of the attack suspect attack Decay over time: Traverse all "attack suspects" every time interval T, and the confidence level P attack - Decay step Δd2 until it reaches 0.0; when the confidence level P attack is 0, delete this "attack suspect"; go to step S2.

2. The live detection assistance method for a face recognition system according to claim 1, wherein: The specific content of step S5 includes: Step S51: Traverse the list of "attack suspects", compare the similarity of facial features, and if it is higher than the threshold P live_threshold , lock it as the "current attack suspect"; Step S52: If not present, add "attack suspect" with confidence level P attack being 0, and set it as the "current attack suspect".

3. The live detection assistance method for a face recognition system according to claim 1, wherein: The specific content of step S6 includes: Step S61: If the probability P of the suspect being alive live is less than the threshold P live_threshold , then the confidence level P of the suspect attack is increased by the update step Δd1 until the confidence level P attack reaches 1.0; if the current confidence level P of the suspect attack is greater than the threshold P attack_threshold , return the result as false; Step S62: If the probability P of the suspect being alive live is greater than or equal to the threshold P live_threshold , then the confidence level P of the suspect attack - Update the step size Δd1 until the confidence level P attack reaches 0.0; If the current confidence level P of the suspect attack is less than the threshold P attack_threshold , return the result as true, otherwise false.

4. A living body detection auxiliary device for a face recognition system, characterized in that: It includes: "Attack suspect" list initialization module, used to establish an "attack suspect" list, where each attack suspect contains two fields, confidence level P attack and facial features, and initialize them, initially empty; A face region extraction module, which is used to extract the face region by using a face detection algorithm; A single-frame face liveness probability calculation module, which is used to calculate the liveness probability of a single-frame face by using a neural network model, denoted as P live ; A face feature value calculation module, which is used to calculate the face feature values; An "attack suspect" matching module, which is used to match the "attack suspect" according to the face feature values. Confidence level P of "current suspect under attack" attack An update module, configured to adjust the confidence level P live by comparing the live probability P of the suspect live_threshold with a set threshold P attack so as to determine whether the current suspect is true or false; Attack suspect confidence P attack Decay over time module, used to traverse all "attack suspects" every time interval T, with confidence P attack - Decay step size Δd2, until it reaches 0.0; when the confidence P attack The confidence is 0, and this "attack suspect" is deleted.

5. The auxiliary device for live detection of a face recognition system according to claim 4, characterized in that: The specific content of the "attack suspect" matching module includes: Traverse the list of "attack suspects", compare the similarity of facial features, which is higher than the threshold P live_threshold , and lock it as the "current attack suspect"; If not present, add "attack suspect", with confidence level P attack being 0, set as "current attack suspect".

6. The auxiliary device for live detection of a face recognition system according to claim 4, characterized in that: Confidence level P of "current attack suspect" attack Update module, specifically including: If the live probability P of the suspect live is less than the threshold P live_threshold , then the confidence level P of the suspect attack + update step Δd1 until the confidence level P attack reaches 1.0; if the current confidence level P of the suspect attack is greater than the threshold P attack_threshold , return the result as false; If the live probability P of the suspect live is greater than or equal to the threshold P live_threshold , then the confidence level P of the suspect attack - Update step Δd1 until the confidence level P attack reaches 0.0; If the current confidence level P of the suspect attack is less than the threshold P attack_threshold , return the result as true, otherwise false.

7. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 3.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 3.

Images