Test method and device for preventing attack of prosthesis in face recognition system

Abstract

The invention discloses a testing method and device for preventing a prosthesis from presenting attacks in a face recognition system. The method comprises the following steps: acquiring a face prosthesis image; inputting any characteristic value of the face prosthesis image into a prosthesis presentation attack recognition model of the face recognition system to be detected, and predicting a living body detection confidence coefficient which is output by the face recognition system to be detected under the condition of the input characteristic value and is used for recognizing the face prosthesis image; the prosthesis presentation attack recognition model is obtained by training in advance according to the following mode: recognizing a pre-collected face prosthesis image by using a face recognition system to be detected, and outputting a living body detection confidence coefficient of the pre-collected face prosthesis image; and taking the in-vivo detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image as a training set, training and integrating various artificial intelligence models, and obtaining a prosthesis presentation attack recognition model. According to the invention, an anti-prosthesis presentation attack test is efficiently realized, a test result is provided based on different image features, and iterative improvement of a face recognition system is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of face recognition technology, and in particular to a face recognition system anti-pseudo-presentation attack testing method and device. Background Art

[0002] Attackers use facial prostheses (such as printed photos, masks, and facial images displayed on electronic screens) to deceive face recognition systems. This behavior is called a prosthesis presentation attack. This attack method seriously threatens the accuracy and reliability of face recognition systems and may cause system misjudgment, which in turn leads to a series of security issues. Therefore, it is necessary to discover security vulnerabilities in face recognition systems through prosthesis presentation attack tests, iteratively improve face recognition systems based on test results, improve the face recognition system's anti-attack capabilities, and ensure that only legitimate users can pass verification, thereby protecting personal privacy and important data from unauthorized access.

[0003] The existing test method for the anti-prosthesis presentation attack function of the face recognition system mainly includes the following steps: making a prosthetic mask; adjusting the angle of the dummy head; adjusting the lighting conditions projected onto the dummy head; using the face recognition system to identify the prosthetic mask with different angle parameters and lighting condition parameters, and the recognition result is the test result of the face recognition system's anti-prosthesis presentation attack function. However, the existing test method for the anti-prosthesis presentation attack function of the face recognition system has the following limitations: the tester needs to wear a prosthetic mask for identification, which is not efficient, and the test results only show whether the prosthesis presentation attack detection is passed or not, and the statistical analysis process is complicated. It is difficult to provide guidance for the subsequent iterative improvement of the face recognition system based on the analysis results of the test results. Summary of the invention

[0004] The embodiment of the present invention provides a method for testing a face recognition system against a prosthesis presentation attack, which is used to efficiently implement a prosthesis presentation attack test, provide test results for different prosthesis image features, and facilitate subsequent iterative improvements of the face recognition system. The method includes:

[0005] Collecting facial prosthesis images;

[0006] Input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows:

[0007] Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image;

[0008] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0009] The embodiment of the present invention also provides a face recognition system anti-prosthesis presentation attack test device, which is used to efficiently implement the prosthesis presentation attack test, provide test results for different prosthesis image features, and facilitate subsequent iterative improvements of the face recognition system. The device includes:

[0010] An image acquisition module, used for acquiring images of a facial prosthesis;

[0011] The anti-pseudo-presentation attack test module is used to input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows:

[0012] Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image;

[0013] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0014] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned face recognition system anti-pseudo-presentation attack testing method when executing the computer program.

[0015] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned face recognition system anti-pseudo-presentation attack testing method is implemented.

[0016] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned face recognition system anti-prosthesis presentation attack testing method is implemented.

[0017] The traditional test device for the anti-pseudo-presentation attack function of the face recognition system needs to use a mechanical arm or other device to repeatedly clamp the face mask or photo to perform the anti-pseudo-presentation attack detection of the face recognition system. The anti-pseudo-presentation attack test is not efficient, and the obtained detection result only shows whether the anti-pseudo-presentation attack detection is passed or not. Compared with the traditional method, in the embodiment of the present invention, a prosthesis presentation attack recognition model for the face recognition system to be tested is proposed, and any feature value of the face prosthesis image is used as the input of the model to obtain the predicted value of the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition. The steps of using a mechanical arm or other device to assist in the traditional method are not required, and the prosthesis presentation attack test is efficiently implemented, and the anti-pseudo-presentation attack detection results for different prosthesis image features are obtained, that is, the recognition ability of the face recognition system to be tested for different prosthesis image features, which is helpful in the subsequent iterative improvement of the face recognition system to be tested. According to the recognition ability of the face recognition system to be tested for different prosthesis image features, an improvement plan is formulated, thereby achieving the effect of effectively guiding the iterative improvement of the face recognition system to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0019] Figure 1 It is a processing flow chart of a method for testing a face recognition system against a fake presentation attack in an embodiment of the present invention;

[0020] Figure 2 A flowchart of a prosthetic presentation attack recognition model for training in an embodiment of the present invention;

[0021] Figure 3 A data processing flow chart of a training prosthetic presentation attack recognition model in an embodiment of the present invention;

[0022] Figure 4 Schematic diagram of a face recognition system anti-pseudo-presentation attack testing device according to an embodiment of the present invention;

[0023] Figure 5 Schematic diagram of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0025] In order to efficiently implement the anti-pseudo-presentation attack test and obtain the test results that can provide efficient guidance for the subsequent iterative improvement of the face recognition system, an embodiment of the present invention provides a face recognition system anti-pseudo-presentation attack test method. Figure 1 FIG. 1 is a processing flow chart of a method for testing a face recognition system against a fake body presentation attack according to an embodiment of the present invention. Figure 1 As shown, the method may include:

[0026] Step 101, collecting a facial prosthesis image;

[0027] Step 102: Input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows:

[0028] Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image;

[0029] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0030] In the existing testing method for the anti-pseudo-presentation attack function of the face recognition system, the detection results of the anti-pseudo-presentation attack function of the face recognition system do not have statistical characteristics. The detection results only show whether the anti-pseudo-presentation attack detection is passed or not. The statistical analysis process is complicated. It is difficult to provide guidance for the subsequent iterative improvement of the face recognition system based on the analysis results of the detection results. Compared with the prior art, the face recognition system anti-prosthesis presentation attack test method of the embodiment of the present invention trains an artificial intelligence model by using the feature values ​​of the face prosthesis image and the liveness detection confidence of the face prosthesis image output by the face recognition system to be tested, and adopts integration technology to obtain the prosthesis presentation attack recognition model of the face recognition system to be tested. It can predict the liveness detection confidence output by the face recognition system to be tested under specific face attribute conditions, which is helpful to obtain the prosthesis presentation attack detection results of different prosthesis image features, that is, the recognition ability of the face recognition system to be tested for different prosthesis image features. In the subsequent iterative improvement of the face recognition system to be tested, an improvement plan can be formulated according to the recognition ability of the face recognition system to be tested for different prosthesis image features, which is helpful to the iterative improvement of the face recognition system to be tested.

[0031] In the embodiment of the present invention, the face prosthesis image is collected. Step 101 can be implemented in the following manner: the tester wears a prosthetic mask of sandstone, silicone, resin, plaster or other types (the prosthetic mask has eyes dug out), uses a camera to collect the face prosthesis image of the person wearing the mask, annotates the collected face prosthesis image, and stores it in the database.

[0032] In an embodiment of the present invention, any eigenvalue of a facial prosthesis image is input into a prosthesis presentation attack recognition model of a face recognition system to be tested, and the liveness detection confidence of the facial prosthesis image recognized by the face recognition system to be tested under the input eigenvalue condition is predicted.

[0033] Step 102 can be implemented by the following operation: input any feature value of any facial image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and the prosthesis presentation attack recognition model outputs the liveness detection confidence that the face recognition system to be tested will output under the face attribute detection condition.

[0034] The prosthetic presentation attack recognition model in step 102 can be obtained by training in the following manner:

[0035] Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image;

[0036] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the trained multiple artificial intelligence models are integrated to obtain a prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0037] Figure 2 Flowchart of training the prosthetic presentation attack recognition model in an embodiment of the present invention. Figure 2 As shown, the training process of the prosthesis presentation attack recognition model of the face recognition system to be tested in the embodiment of the present invention can be mainly divided into five operation steps: face prosthesis processing operation 201, face prosthesis image generation operation 202, face recognition system anti-prosthesis presentation attack detection operation 203, anti-prosthesis presentation attack detection result analysis operation 204, and anti-prosthesis presentation attack result prediction operation 205.

[0038] Figure 2 The face prosthesis processing operation 201 is the same as the operation performed in step 101. The face prosthesis image is collected in advance before model training. The implementation method can refer to the implementation method of step 101.

[0039] In the embodiment of the present invention, the face recognition system to be tested is used to recognize the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output.

[0040] Since the production cost of prosthetic masks is high, in the embodiments of the present invention, taking into account the need to reduce costs and improve the feasibility of the anti-prosthetic presentation attack testing method, the facial prosthesis image generation operation 202 performs image environment transformation and / or semantic level transformation on the pre-collected facial prosthesis image, thereby greatly reducing the cost of facial prosthesis production.

[0041] In order to predict the confidence level of liveness detection output by the face recognition system under test for the face prosthesis image under the input feature value condition, a detection result with statistical characteristics is obtained, such as Figure 2 In the face recognition system anti-prosthesis presentation attack detection operation 203, the liveness detection confidence of the face prosthesis image output by the face recognition system to be tested is obtained, and a training set of the prosthesis presentation attack recognition model is constructed.

[0042] In one embodiment, the pre-collected face prosthesis image is subjected to image environment transformation and / or semantic level transformation, the pre-collected face prosthesis image and the transformed image are identified using a face recognition system to be tested, and the liveness detection confidence of the pre-collected face prosthesis image and the transformed image is output; the liveness detection confidence of the pre-collected face prosthesis image and the transformed image, and the feature values ​​of the pre-collected face prosthesis image and the transformed image are used as training sets.

[0043] Perform image environment transformation and / or semantic level transformation on the pre-collected facial prosthesis image, transform the brightness, contrast, age, expression, angle, accessories, hairstyle, etc. of the facial image, then output the transformed facial prosthesis image, annotate the transformed facial prosthesis image, and store it in the database.

[0044] In order to further improve the efficiency of obtaining the liveness detection confidence output by the face recognition system, in the embodiment of the present invention, the API interface of the face recognition system to be tested is called, and the face recognition system to be tested is used to identify the pre-collected face prosthesis image, so as to realize automatic testing at the software level and greatly improve the detection efficiency.

[0045] In one embodiment, the API interface of the face recognition system to be tested is called, the face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output.

[0046] In one embodiment, a pre-collected face prosthesis image is identified using a face recognition system to be tested, and after the liveness detection confidence of the pre-collected face prosthesis image is output, the liveness detection confidence of the pre-collected face prosthesis image is associated with a feature value of the face prosthesis image and stored in a database.

[0047] For example, the face recognition system to be tested can be used to identify the transformed face prosthesis image, and the liveness detection confidence of the transformed face prosthesis image and the prosthetic mask material, brightness, contrast, age, expression, angle, accessories, hairstyle and other data of the transformed face prosthesis image can be associated and stored in a database.

[0048] In an embodiment of the present invention, the liveness detection confidence of a pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by the training are integrated to obtain a prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input, and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0049] In order to improve the accuracy of the prosthesis presentation attack recognition model in predicting the liveness detection confidence of the face recognition system under test in identifying the face prosthesis image output, Figure 2 In the anti-prosthesis presentation attack detection result analysis operation 204, consider preprocessing the training set, where the preprocessing operation can be: performing a correlation analysis on the liveness detection confidence of the face prosthesis image and each feature value of the face prosthesis image.

[0050] In one embodiment, correlation analysis is performed on the liveness detection confidence of the pre-collected face prosthesis image and each eigenvalue of the pre-collected face prosthesis image; eigenvalues ​​whose correlation with the liveness detection confidence is lower than a threshold are eliminated from the database; and the liveness detection confidence of the pre-collected face prosthesis image and the eigenvalues ​​of the pre-collected face prosthesis image are used as a training set, including: adding the eigenvalues ​​retained in the database to the training set.

[0051] In specific implementation, the data in the database is divided into a training data set and a test data set in a ratio of 2:1.

[0052] In one embodiment, a soft voting method is used to integrate multiple trained artificial intelligence models to obtain the model.

[0053] In one embodiment, a decision tree model, a random forest model, and a gradient boosting model are trained.

[0054] Figure 2 In the anti-prosthesis presentation attack result prediction operation 205, the training data and decision tree, random forest, and gradient boosting models can be used for model training, the above models can be integrated using the soft voting method, and the model prediction results can be verified using the test data set. The integrated model is the prosthesis presentation attack recognition model of the face recognition system to be tested.

[0055] For example, the data processing flow of training a prosthetic presentation attack recognition model is described in detail. Figure 3 This is a data processing flow chart of the training prosthetic presentation attack recognition model in an embodiment of the present invention. The symbols and expressions in this embodiment can be found in Figure 3 .

[0056] In a dark room, collect pictures x∈X of the tester wearing a facial prosthesis (where X is all the collected pictures and x is one of them), and record the facial attribute values ​​in the label of the picture, with label value lable{material, skin color, brightness, contrast, age, expression, angle, accessories, hairstyle}, and lable∈L (where L is the set of label values ​​of all collected pictures and lable is the set of attribute values ​​of one of them). Record X and L in the database data∈D.

[0057] The STYLEGAN network can be used to transform the face of image X, with different ranges, such as material, brightness, contrast, skin color, age, expression, angle, accessories, and hairstyle. The transformed image x′∈X′ (where X′ is all the transformed images and x′ is one of them) and the corresponding label value lable′∈L′ (where L′ is the set of label values ​​of all images collected and lable′ is the set of attribute values ​​of one of them) can be recorded in the database data′∈D′.

[0058] Use the image X′ as the input of the face recognition system to be tested. The face recognition system outputs the corresponding liveness detection confidence score and records the score in the database score∈S.

[0059] Merge data D′ and S into data set W, analyze the label value corresponding to image X′ in D′, that is, the possibility of relationship between the feature value L′ of face attribute and the confidence level S of liveness detection, and select three feature values ​​that have important relationship with the confidence level S of liveness detection. The method is as follows: For data set W, calculate the information gain of each feature, compare their sizes, and select the three features with the largest information gain. Calculate the entropy H(W) of data set W, Among them, p i represents the proportion of the i-th sample in the data set W, and k represents the number of categories in the data set W, that is, the number of features. k ∈A, calculate information gain Among them, {a1=lable1, a2=lable2,…, a3=lable3}.

[0060] Keep the three features a′={a1,a2,a3,} and the corresponding scores with the largest information gain in the data set W, remove other data from the data set W, and record it as the data set W′.

[0061] The training data set and the test data set are obtained by data separation. The data set W′ is divided into the training data set W1′ and the test data set W2′ in a ratio of 2:1. The training data set is input into the decision tree DT, random forest RF, and gradient boosting GB models. The confidence of the data set W1′ output by each model belonging to each category is pi(j) (j=1, 2, 3). The above models are integrated to output the integrated model where w i is the weight of the above three models, and the average value is w i =1 / 3.

[0062] The test dataset W2′ is input into the integrated model M, and the model M is used to output the predicted data score′∈S′. The consistency of score′ and score is compared to evaluate the prosthesis presentation attack recognition model of the face recognition system under test.

[0063] Randomly generate a set of pictures x random , its label value lable random Input the prosthesis to attack the recognition model M, then the model M predicts the face recognition system under test under the face attribute detection condition, that is, label random The corresponding liveness detection result under the value is score predict .

[0064] To sum up: the face recognition system anti-prosthesis presentation attack test method proposed in the embodiment of the present invention collects face prosthesis images; any feature value of the face prosthesis image is input into the prosthesis presentation attack recognition model of the face recognition system to be tested, and the liveness detection confidence of the face prosthesis image output by the face recognition system to be tested under the input feature value condition is predicted; the prosthesis presentation attack recognition model is pre-trained in the following manner: the face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output; the liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets, and multiple artificial intelligence models are trained and integrated to obtain a prosthesis presentation attack recognition model. The method does not require the steps of using a robot arm or other device to assist in the traditional method, and can efficiently implement the anti-prosthesis presentation attack test. The prosthesis presentation attack recognition model of the face recognition system to be tested can predict the liveness detection confidence output by the face recognition system to be tested under specific facial attribute conditions, and obtain the anti-prosthesis presentation attack detection results for different prosthesis image features, that is, the recognition ability of the face recognition system to be tested for different prosthesis image features. This is helpful in formulating improvement plans according to the recognition ability of the face recognition system to be tested for different prosthesis image features in the subsequent iterative improvement of the face recognition system to be tested, which is helpful for iterative improvement of the face recognition system to be tested.

[0065] The embodiment of the present invention also provides a face recognition system anti-pseudo-presentation attack test device, as described in the following embodiment. Since the principle of solving the problem by the device is similar to the face recognition system anti-pseudo-presentation attack test method, the implementation of the device can refer to the implementation of the face recognition system anti-pseudo-presentation attack test method, and the repeated parts will not be repeated.

[0066] Figure 4 FIG. 1 is a schematic diagram of a face recognition system anti-pseudo-presentation attack testing device according to an embodiment of the present invention. Figure 4 As shown, the device comprises:

[0067] An image acquisition module 401 is used to acquire a face prosthesis image;

[0068] The anti-pseudo-presentation attack test module 402 is used to input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows:

[0069] Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image;

[0070] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

[0071] The training method of the prosthesis presentation attack recognition model in the anti-prosthesis presentation attack test module 402 can be implemented in the specific method of the following embodiment.

[0072] In one embodiment, a face recognition system to be tested is used to identify a pre-collected face prosthesis image, and a liveness detection confidence of the pre-collected face prosthesis image is output; the liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including:

[0073] Performing image environment transformation and / or semantic level transformation on the pre-collected face prosthesis image, using the face recognition system to be tested to identify the pre-collected face prosthesis image and the transformed image, and outputting the liveness detection confidence of the pre-collected face prosthesis image and the transformed image;

[0074] The liveness detection confidence of the pre-collected face prosthesis image and the transformed image, and the feature values ​​of the pre-collected face prosthesis image and the transformed image are used as training sets.

[0075] In one embodiment, using the face recognition system to be tested to identify a pre-collected face prosthesis image and outputting a liveness detection confidence of the pre-collected face prosthesis image includes:

[0076] By calling the API interface of the face recognition system to be tested, the face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output.

[0077] In one embodiment, after using the face recognition system to be tested to recognize the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes:

[0078] The liveness detection confidence of the pre-collected human face prosthesis image and the feature value of the pre-collected human face prosthesis image are associated and stored in a database.

[0079] In one embodiment, after using the face recognition system to be tested to recognize the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes:

[0080] Performing correlation analysis on the liveness detection confidence of the pre-collected face prosthesis image and each feature value of the pre-collected face prosthesis image; removing feature values ​​whose correlation with the liveness detection confidence is lower than a threshold from the database;

[0081] The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including:

[0082] The feature values ​​retained in the database are added to the training set.

[0083] In one embodiment, multiple artificial intelligence models obtained through training are integrated to obtain a prosthetic presentation attack recognition model, including:

[0084] Using the soft voting method, multiple trained artificial intelligence models are integrated to obtain a prosthetic presentation attack recognition model.

[0085] In one embodiment, multiple artificial intelligence models are trained, including:

[0086] Train decision tree models, random forest models, and gradient boosting models.

[0087] Compared with the technical solution in the prior art that requires the use of a robotic arm or other device to repeatedly clamp a face mask or photo to perform anti-prosthesis presentation attack detection of the face recognition system, and the obtained detection result only shows whether the anti-prosthesis presentation attack detection is passed or not, the prosthesis presentation attack recognition model of the face recognition system to be tested proposed in the embodiment of the present invention is used, and any feature value of the face prosthesis image is used as the input of the model to obtain the predicted value of the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition. The steps of using a robotic arm or other device to assist in the traditional method are not required, and the anti-prosthesis presentation attack test is efficiently implemented, and the anti-prosthesis presentation attack detection results for different prosthesis image features are obtained, that is, the recognition ability of the face recognition system to be tested for different prosthesis image features. This helps to formulate an improvement plan according to the recognition ability of the face recognition system to be tested for different prosthesis image features in the subsequent iterative improvement of the face recognition system to be tested, thereby achieving the effect of effectively guiding the iterative improvement of the face recognition system to be tested.

[0088] Figure 5 Schematic diagram of a computer device in an embodiment of the present invention. Figure 5 As shown, an embodiment of the present invention further provides a computer device 500, including a processor 501, a memory 502, and a computer program 503 stored in the memory 502 and executable on the processor 501, wherein the processor 501 implements the above-mentioned face recognition system anti-pseudo-presentation attack test method when executing the computer program 503.

[0089] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned face recognition system anti-pseudo-presentation attack testing method is implemented.

[0090] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned face recognition system anti-prosthesis presentation attack testing method is implemented.

[0091] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 provides the functions specified in a block or multiple blocks.

[0093] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising 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.

[0094] 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.

[0095] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for testing a face recognition system against a fake presentation attack, characterized in that: include: Collecting facial prosthesis images; Input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows: Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

2. The method according to claim 1, characterized in that Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including: Performing image environment transformation and / or semantic level transformation on the pre-collected face prosthesis image, using the face recognition system to be tested to identify the pre-collected face prosthesis image and the transformed image, and outputting the liveness detection confidence of the pre-collected face prosthesis image and the transformed image; The liveness detection confidence of the pre-collected face prosthesis image and the transformed image, and the feature values ​​of the pre-collected face prosthesis image and the transformed image are used as training sets.

3. The method according to claim 1, characterized in that The face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output, including: By calling the API interface of the face recognition system to be tested, the face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output.

4. The method according to claim 1, characterized in that After using the face recognition system to be tested to identify the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes: The liveness detection confidence of the pre-collected human face prosthesis image and the feature value of the pre-collected human face prosthesis image are associated and stored in a database.

5. The method according to claim 4, characterized in that After using the face recognition system to be tested to identify the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes: Performing correlation analysis on the liveness detection confidence of the pre-collected face prosthesis image and each feature value of the pre-collected face prosthesis image; removing feature values ​​whose correlation with the liveness detection confidence is lower than a threshold from the database; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including: The feature values ​​retained in the database are added to the training set.

6. The method according to claim 1, characterized in that The multiple artificial intelligence models obtained through training are integrated to obtain the prosthetic presentation attack recognition model, including: The soft voting method is used to integrate the multiple artificial intelligence models obtained through training to obtain the prosthesis presentation attack recognition model.

7. The method according to claim 1, characterized in that Train multiple AI models, including: Train decision tree models, random forest models, and gradient boosting models.

8. A face recognition system anti-pseudo-presentation attack test device, characterized in that: include: An image acquisition module, used for acquiring images of a facial prosthesis; The anti-pseudo-presentation attack test module is used to input any feature value of the face prosthesis image into the prosthesis presentation attack recognition model of the face recognition system to be tested, and predict the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested under the input feature value condition; the prosthesis presentation attack recognition model is pre-trained as follows: Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as training sets to train multiple artificial intelligence models, and the multiple artificial intelligence models obtained by training are integrated to obtain the prosthesis presentation attack recognition model; the prosthesis presentation attack recognition model takes the feature value of the face prosthesis image as input and takes the liveness detection confidence of the face prosthesis image recognized by the face recognition system to be tested as output.

9. The device according to claim 8, characterized in that Using the face recognition system to be tested to identify the pre-collected face prosthesis image, and outputting the liveness detection confidence of the pre-collected face prosthesis image; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including: Performing image environment transformation and / or semantic level transformation on the pre-collected face prosthesis image, using the face recognition system to be tested to identify the pre-collected face prosthesis image and the transformed image, and outputting the liveness detection confidence of the pre-collected face prosthesis image and the transformed image; The liveness detection confidence of the pre-collected face prosthesis image and the transformed image, and the feature values ​​of the pre-collected face prosthesis image and the transformed image are used as training sets.

10. The device according to claim 8, characterized in that The face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output, including: By calling the API interface of the face recognition system to be tested, the face recognition system to be tested is used to identify the pre-collected face prosthesis image, and the liveness detection confidence of the pre-collected face prosthesis image is output.

11. The device according to claim 8, characterized in that After using the face recognition system to be tested to identify the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes: The liveness detection confidence of the pre-collected human face prosthesis image and the feature value of the pre-collected human face prosthesis image are associated and stored in a database.

12. The device according to claim 11, characterized in that After using the face recognition system to be tested to identify the pre-collected face prosthesis image and outputting the liveness detection confidence of the pre-collected face prosthesis image, the method further includes: Performing correlation analysis on the liveness detection confidence of the pre-collected face prosthesis image and each feature value of the pre-collected face prosthesis image; removing feature values ​​whose correlation with the liveness detection confidence is lower than a threshold from the database; The liveness detection confidence of the pre-collected face prosthesis image and the feature value of the pre-collected face prosthesis image are used as a training set, including: The feature values ​​retained in the database are added to the training set.

13. The device according to claim 8, characterized in that The multiple artificial intelligence models obtained through training are integrated to obtain the prosthetic presentation attack recognition model, including: The soft voting method is used to integrate the multiple artificial intelligence models obtained through training to obtain the prosthesis presentation attack recognition model.

14. The device according to claim 8, characterized in that Train multiple AI models, including: Train decision tree models, random forest models, and gradient boosting models.

15. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

17. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images