An Authentication Method for Cooperative Identity Identification and False Attribute Detection

Abstract

The present invention discloses an authentication method for collaborative identity identification and false attribute detection, which relates to the technical field of trusted identity verification. This method solves the problem that unknown attacks are difficult to detect in the prior art. The method includes: obtaining multiple first features respectively corresponding to the multi-dimensional identity attributes of the object to be tested, and searching for multiple pieces of data corresponding to each first feature in the database; determining the optimal depth; combining the above parameters to obtain a consensus identity set, and calculating the convergence degree respectively to determine the optimal consensus identity; judging whether multiple second features of the optimal consensus identity are in the top δ of the first set to obtain a judgment result, and performing feature identification on the multiple second features; outputting the feature identification to complete the identity authentication of the object to be tested; realizing the use of an unsupervised unknown attack detection method, improving the accuracy of the identity identification method and the security performance of the system, so as to achieve the purpose of accurate identity identification.

Description

Technical Field

[0001] The present invention relates to the technical field of trusted identity authentication, and particularly to an authentication method for collaborative identity identification and false attribute detection. Background Art

[0002] Biometric systems face security threats from false attributes, such as photo attacks, video attacks, silicone masks, and false fingerprints. These attacks aim to affect the normal operation of biometric systems to confuse or impersonate others, and new unknown attacks pose a major challenge. This paper attempts to solve unknown types of attacks in an unsupervised learning manner through the collaborative recognition of multiple biometric attributes to improve the security of identity recognition systems.

[0003] In complex dynamic scenarios, single-modal biometric recognition systems are vulnerable to limitations such as non-generality, environmental changes, and spoofing attacks. These factors threaten the recognition accuracy and system security of single-modal biometric recognition systems. Inspired by the fact that multi-source information fusion technology can reduce the error probability and enhance the system robustness, researchers have turned their attention to multi-modal biometric recognition technology, which completes identity recognition by fusing different biometric information.

[0004] Regarding the problem that current identity recognition systems face spoofing attacks or unknown attacks, the mainstream methods mainly learn the difference between genuine and fake samples through manual feature extraction or machine learning. They are better for detecting 2D attacks and some 3D attacks, but less effective for detecting unknown attacks. Most existing unknown attack methods are based on supervised learning with prior knowledge to learn the decision boundary between genuine and fake. However, in real complex scenarios, prior knowledge may not be available for learning, so unsupervised unknown attack detection methods will be the future development direction. Summary of the Invention

[0005] The present invention effectively solves the problems in the prior art that unknown attacks are difficult to detect and prior knowledge cannot be used for learning in real complex scenarios by providing an authentication method for collaborative identity identification and false attribute detection, and further realizes the use of unsupervised unknown attack detection methods, improves the accuracy of the identity identification method and the security performance of the system so as to achieve the purpose of accurate identity identification.

[0006] In a first aspect, an embodiment of the present invention provides an authentication method for collaborative identity identification and false attribute detection, the method comprising:

[0007] Obtain multiple first features respectively corresponding to multi-dimensional identity attributes of a to-be-tested object, search for multiple pieces of data corresponding to each of the first features in a database, and merge the multiple pieces of data into an identification result set topL;

[0008] Input the recognition result set topL into the trained deep trainer to obtain the optimal depth;

[0009] Use the step-by-step consensus method to calculate the recognition result set topL and the optimal depth to obtain a consensus identity set, and calculate the convergence degree of each consensus identity in the consensus identity set respectively, and determine that the consensus identity with the smallest convergence degree is the optimal consensus identity;

[0010] Judge whether multiple second features of the optimal consensus identity are in the first set topδ to obtain a judgment result, and assign values to the feature identifiers according to the judgment result to obtain the feature identifiers of multiple second features of the optimal consensus identity respectively; where

[0011] Output the feature identifiers corresponding to multiple second features of the optimal consensus identity to complete the identity authentication of the object to be measured.

[0012] Combined with the first aspect, in a possible implementation manner, the database includes data of multiple known objects, and each data of the known object includes: a consensus identity, multi-dimensional identity attributes, and multiple third features corresponding to the multi-dimensional identity attributes.

[0013] Combined with the first aspect, in a possible implementation manner, the step of searching for multiple data corresponding to each of the first features in the database and merging the multiple data into a recognition result set topL specifically includes:

[0014] Calculate the similarity between each first feature in the multiple first features and the same features in the multiple third features of each data in the database respectively, and sort the known objects according to the similarity to obtain the sorting of the known objects corresponding to each first feature;

[0015] Determine that the multiple data with the top L sorting of the known objects corresponding to each first feature are used to form a recognition result set topL.

[0016] Combined with the first aspect, in a possible implementation manner, the step of using the step-by-step consensus method to calculate the recognition result set topL and the optimal depth to obtain a consensus identity set specifically includes:

[0017] Initialize the consensus identity set, the consensus value set, and the convergence degree value set; where the size of the consensus value set is the same as the number of data in the database;

[0018] In the recognition result set topL, search for the data with the feature values equal to the multiple first features respectively to obtain multiple search data;

[0019] Count the number of occurrences of the same data in the multiple search data, and record the statistical result in the consensus value set;

[0020] Combined with the optimal depth, obtain the data in the consensus value set where the statistical result is greater than or equal to the second threshold, and store the corresponding known objects in the consensus identity set to obtain the consensus identity set.

[0021] Combined with the first aspect, in a possible implementation manner, the specific steps of calculating the convergence degree of each consensus identity in the consensus identity set include:

[0022] Respectively determine whether multiple second features of the multi-dimensional identity attributes of the consensus identity are in the recognition result set topL. If not, assign the second feature convergence degree corresponding to the second feature as 1, and the value range of the convergence degree is (0, 1];

[0023] If so, respectively calculate the convergence degree of multiple second features of the multi-dimensional identity attributes of the consensus identity to obtain the values of multiple second feature convergence degrees, and add and average the values of multiple second convergence degrees to obtain the convergence degree of the consensus identity.

[0024] Combined with the first aspect, in a possible implementation manner, the convergence degree is specifically expressed as:

[0025]

[0026] Among them, s represents the sorting position of the known object m in the recognition result set topL; N represents the dimension of the multi-dimensional identity attribute; L represents the size of the recognition result set topL; K(m) represents the convergence degree of the known object m.

[0027] Combined with the first aspect, in a possible implementation manner, the size of the first set topδ is expressed as:

[0028] δ = min(8 + S, L)

[0029] Among them, S represents the optimal depth; L represents the size of the recognition result set topL.

[0030] Combined with the first aspect, in a possible implementation manner, the specific steps of assigning a feature identifier according to the judgment result include:

[0031] If the judgment result is yes, set the feature identifier of the feature to true;

[0032] If the judgment result is no, set the feature identifier of the feature to false.

[0033] In combination with the first aspect, in a possible implementation, the multi-dimensional identity attributes of the object to be measured include real attributes and false attributes.

[0034] In a second aspect, the present invention provides an authentication device for collaborative identity identification and false attribute detection, and the device includes:

[0035] An identification result set acquisition module, configured to obtain a plurality of first features respectively corresponding to the multi-dimensional identity attributes of the object to be measured, search for multiple pieces of data corresponding to each of the first features in a database, and merge the multiple pieces of data into an identification result set topL;

[0036] An optimal depth calculation module, configured to input the identification result set topL into a trained depth trainer to obtain an optimal depth;

[0037] An optimal consensus identity acquisition module, configured to calculate the identification result set topL and the optimal depth by using a step-by-step consensus method to obtain a consensus identity set, and respectively calculate the convergence degrees of the consensus identities in the consensus identity set, and determine the consensus identity with the smallest convergence degree as the optimal consensus identity;

[0038] A feature identifier acquisition module, configured to determine whether multiple second features of the optimal consensus identity are in a first set topδ, obtain a judgment result, and assign a feature identifier according to the judgment result to respectively obtain the feature identifiers of the multiple second features of the optimal consensus identity; wherein,

[0039] A feature identifier output module, configured to output the feature identifiers corresponding to the multiple second features of the optimal consensus identity to complete the identity authentication of the object to be measured.

[0040] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0041] The present invention adopts an authentication method for collaborative identity identification and false attribute detection, and the method includes: obtaining a plurality of first features respectively corresponding to multi-dimensional identity attributes of an object to be measured, searching for multiple pieces of data corresponding to each first feature in a database, and merging the multiple pieces of data into an identification result set topL, converting the false identity attribute detection from a low-dimensional space anti-counterfeiting detection problem into a high-dimensional space sample difference research problem, getting rid of the method restriction that traditional supervised detection techniques need to learn a large amount of known attack data, and can well solve unknown attacks; inputting the identification result set topL into a trained deep trainer to obtain an optimal depth, better solving the problem of attribute forgery through unsupervised learning by studying the optimal convergence depth difference of different samples in the high-dimensional space; using a step-by-step consensus method to calculate the identification result set topL and the optimal depth to obtain a consensus identity set, and respectively calculating the convergence degrees of each consensus identity in the consensus identity set, determining the consensus identity with the smallest convergence degree as the optimal consensus identity, judging whether multiple second features of the optimal consensus identity are in the first set topδ to obtain a judgment result, and assigning a feature identifier according to the judgment result to respectively obtain the feature identifiers of multiple second features of the optimal consensus identity; where Fusing multi-dimensional attribute information for identity recognition can improve the accuracy of identity identification and avoid the problem of the decline in recognition accuracy caused by algorithm or environmental errors; outputting the feature identifiers corresponding to multiple second features of the optimal consensus identity to complete the identity authentication of the object to be measured; effectively solving the problems that unknown attacks are difficult to detect in the prior art and prior knowledge cannot be used for learning in realistic complex scenarios, and further realizing the use of an unsupervised unknown attack detection method to improve the accuracy of the identity identification method and the security performance of the system so as to achieve the purpose of accurate identity identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments of the present invention or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 It is a flowchart of the steps of the authentication method for collaborative identity identification and false attribute detection provided by the embodiment of the present invention;

[0044] Figure 2 It is a flowchart of a specific usage embodiment provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] In a first aspect, an authentication method for collaborative identity identification and false attribute detection is provided in an embodiment of the present invention. As Figure 1 shown, the method includes the following steps S101 to S105.

[0047] S101, obtain multiple first features corresponding to the multi-dimensional identity attributes of the object to be tested, find multiple pieces of data corresponding to each first feature in the database, and merge the multiple pieces of data into an identification result set topL.

[0048] The database includes data of multiple known objects, and each piece of data of a known object includes: a consensus identity, multi-dimensional identity attributes, and multiple third features corresponding to the multi-dimensional identity attributes.

[0049] In a specific embodiment provided by the present invention, the present invention is based on an established database template, given N identity attributes of an object with an unknown identity, which may include more than 50% false attributes. According to the identification result sequence of the N attributes, false attributes are analyzed and detected, and the true identity is calculated to ensure the credible identification of the identity. Moreover, the algorithm is simple and the computational complexity is low. The mixed attribute set is defined as the identity attributes of a to-be-tested object that simultaneously include normally collected attributes and false attributes. The distance between the mixed attribute set and the true attribute set: refers to the distance between the mixed attribute set of the to-be-tested object and the attribute template corresponding to its actual identity.

[0050] The principle of the technical solution of the present invention is that the optimal depths of samples caused by normal / false attributes are different, that is, the normal / false attribute distributions of the to-be-tested object are significantly different in their optimal depths. The higher the forgery ratio, the deeper the optimal depth of its identity calculation; for samples with a low forgery ratio, their optimal depths are relatively higher. Based on this difference, we found that under the condition that the false attribute ratio is greater than 50%, there are partial convergence characteristics between the distance of the mixed attribute set - the true attribute set, while the false attributes are relatively divergent. In the case of random forgery, the true attributes are converged together in the multi-dimensional attribute results, and since the false attributes are randomly generated, even if the false attributes account for the vast majority, their distributions in the multi-dimensional attribute set are relatively divergent, and only in extremely rare cases will there be convergence.

[0051] Therefore, based on the differences in the aggregation states of true / false attributes, analyze the recognition result sequence of multi-dimensional attributes to find the true identity and detect false attributes.

[0052] In a specific embodiment provided by the present invention, the data in the constructed database includes:

[0053] Step 1: Collect the N-dimensional identity attribute information of M objects and establish a known identity attribute information library, where the identities are represented as O1, O2, O3, …, O m , and the attributes are represented by lowercase letters a, b, c, d…, n. The specific collection and establishment methods are as follows:

[0054] Successively collect the N attributes of M identity objects. First, collect the a attributes of M identity objects and uniquely identify the identities with standardized naming. Secondly, collect the b attributes of M identity objects and name them in the same standardized way. And so on, collect the remaining attributes of M objects until the N attributes are collected completely, then M identity object libraries can be formed, where C(i,j) represents the j-th dimension attribute of the i-th object;

[0055] Step 2: Establish the feature templates of the identity attributes of M objects to form N feature templates. The establishment methods are as follows:

[0056] Successively extract the features corresponding to the N identity attributes of M identity objects. The selection of features is determined according to the types of attributes. One feature template is established for each type of attribute. The corresponding features and identity identifiers are saved in the feature templates. First, extract the feature of attribute a of S identity objects, secondly, extract the feature of attribute b, and finally, extract the feature of attribute c. And so on, extract the features of the remaining attributes to form N feature templates, which are respectively denoted as F(j), where j ∈ {1, 2,..., N}.

[0057] Given the N-dimensional identity attributes of the object to be measured (unknown identity) X, its N-dimensional identity attributes are X(j), where j ∈ {1, 2,..., N}. Among these N-dimensional attributes, there are both normal attributes and false attributes. The number of false attributes is N Fake , satisfying the condition N Fake ≥50% × N. The prerequisite of this method requires at least two or more normally collected attributes so that the true attributes will converge.

[0058] In step S101, search for multiple pieces of data corresponding to each first feature in the database and merge the multiple pieces of data into the recognition result set topL, which specifically includes the following steps:

[0059] (1) Calculate the similarity between each of the multiple first features and the same features among the multiple third features of each piece of data in the database respectively, and sort the known objects according to the similarity to obtain the sorting of the known objects corresponding to each first feature.

[0060] (2) Determine the multiple pieces of data whose sorting of the known objects corresponding to each first feature is among the top L, and form the recognition result set topL with the multiple pieces of data.

[0061] In a specific embodiment provided by the present invention, for the N-dimensional identity attribute X(j) of the object to be measured X, where j ∈ {1, 2,..., N}, select the preface recognition algorithm for the corresponding attribute, obtain N features of the corresponding attribute, and compare them with the feature template F(j) in the database to obtain N recognition result sequences. In the recognition results, select the multiple pieces of data whose identity selection similarity sorting is among the top L, and the recognition result is denoted as R a ,R b ,R c ,..., the recognition method is analyzed as follows:

[0062] For the identity attribute X(1), use F(1) as the known identity object feature template library, extract the features of the attribute X(1), and perform matching comparison with the template F(1). Select the multiple pieces of data whose similarity sorting is among the top L as the recognition result of the attribute a1, and denote it as R a , for example R a ={O2, O4, O6,..., O L}; and so on, the recognition results R b , R c of the attributes X(2), X(3)... can be obtained. Add the recognition results R a , R b , R c to the recognition result set topL.

[0063] S102, Input the recognition result set topL into the trained deep trainer to obtain the optimal depth.

[0064] S103, Use the step-by-step consensus method to calculate the recognition result set topL and the optimal depth to obtain the consensus identity set, and calculate the convergence degree of each consensus identity in the consensus identity set respectively. Determine the consensus identity with the smallest convergence degree as the optimal consensus identity. Since the proportion of false attributes is relatively large in the scenario of this project, it is not excluded that there are two or more consensus identities. Therefore, it is necessary to calculate the convergence degree to determine the unique consensus identity.

[0065] In step S103, using the step-by-step consensus method to calculate the recognition result set topL and the optimal depth to obtain the consensus identity set specifically includes the following steps:

[0066] (1) Initialize the consensus identity set, the consensus value set, and the convergence degree value set; among them, the size of the consensus value set is the same as the number of data in the database. Take the M identities as consensus identities, and set their consensus values to 0, that is, d[O i = 0, and the convergence degree value is the same, that is, dcon[O i = 0.

[0067] (2) In the recognition result set topL, respectively search for the data whose feature values are equal to those of multiple first features to obtain multiple search data.

[0068] (3) Count the number of occurrences of the same data among the multiple search data, and record the statistical results in the consensus value set.

[0069] (4) Combine the optimal depth, obtain the data in the consensus value set whose statistical results are greater than or equal to the second threshold, and store the corresponding known objects into the consensus identity set to obtain the consensus identity set.

[0070] The convergence degree is specifically expressed as:

[0071]

[0072] Among them, s represents the sorting position of the known object m in the recognition result set topL; N represents the dimension of the multi-dimensional identity attribute; L represents the size of the recognition result set topL; K(m) represents the convergence degree of the known object m.

[0073] If the object m does not appear in the first L positions of the attribute set, then take s = L, then

[0074]

[0075] In the case of high recognition algorithm accuracy and ideal acquisition environment, the value of L is relatively low, generally around 5, while when the recognition algorithm accuracy is low or the acquisition sample environment is poor, the value of L is relatively high, generally around 10. The subsequent values are considered not to be within the scope of convergence, so they are not considered.

[0076] S104. Determine whether the multiple second features of the optimal consensus identity are in the first set topδ to obtain a judgment result, and assign values to the feature identifiers according to the judgment result to obtain the feature identifiers of the multiple second features of the optimal consensus identity; among them, The size of the first set topδ is expressed as:

[0077] δ = min(8 + S, L)

[0078] Among them, S represents the optimal depth; L represents the size of the recognition result set topL.

[0079] In step S104, the convergence degrees of each consensus identity in the consensus identity set are calculated respectively, which specifically includes:

[0080] (1) Determine whether multiple second features of the multi-dimensional identity attributes of the consensus identity are in the recognition result set topL respectively. If not, assign the second feature convergence degree corresponding to the second feature as 1, and the value range of the convergence degree is (0, 1].

[0081] (2) If so, calculate the convergence degrees of multiple second features of the multi-dimensional identity attributes of the consensus identity respectively, obtain the values of multiple second convergence degrees, and add the values of multiple second convergence degrees and take the average to obtain the convergence degree of the consensus identity.

[0082] In step S104, the feature identifier is assigned according to the judgment result, which specifically includes:

[0083] (1) If the judgment result is yes, set the feature identifier of the feature to true, flag_a = True.

[0084] (2) If the judgment result is no, set the feature identifier of the feature to false, flag_a = False.

[0085] The present invention converts the false identity attribute detection from a low-dimensional space anti-counterfeiting detection problem into a high-dimensional space sample difference research problem, getting rid of the method restriction that traditional supervised detection technologies need to learn a large amount of known attack data, and can well solve unknown attacks. Traditional false attribute detection methods have good solutions for known attacks, and the solution for unknown attacks is through learning the training data of known attacks. However, in real life, attackers can always come up with endless means to confuse identity recognition methods, and there are certain defects in the supervised learning method. Therefore, the present invention uses an unsupervised learning method to study the data distribution differences between attacked and normally collected samples in a high-dimensional space for false attribute detection and real identity calculation.

[0086] The present invention studies the false attribute detection problem in a high-dimensional space, fuses multi-dimensional attribute information for identity recognition, can improve the identity recognition accuracy, avoid the problem of decreased recognition accuracy caused by algorithm or environmental errors, and the present invention can be applied to non-cooperative and unaware identity authentication scenarios, optimize the user experience, and improve the system security. Traditional multi-modal biometric detection only focuses on the preferred identity and performs identity recognition through the majority decision idea. In the case of serious recognition algorithm errors and environmental interference, the accuracy of identity recognition decreases. The present invention studies the sorting sequence of the recognized identity results, studies the identities ranked second, third, etc. after the preferred identity, and studies the convergence phenomenon between real attributes in a high-dimensional space for false attribute detection and real identity calculation.

[0087] S105. Output the feature identifiers corresponding to multiple second features of the optimal consensus identity to complete the identity authentication of the object to be tested.

[0088] The present invention provides a specific usage embodiment. As Figure 2 shown, for the N identity attributes of the object to be tested, the false attributes contained therein are detected through analysis. To meet the requirement that there are at least 2 true attributes and the number of false attributes is greater than 50% of the total number of attributes in the experiment, obviously, the value of N is a natural number greater than or equal to 4. The technical solution of the present invention can be automatically run by using computer software technology. In the embodiment, the value of N is 5, and its implementation process includes the following steps:

[0089] Step 1. Collect the N-dimensional identity attribute information of M objects to establish a known identity attribute information library, where the identities are represented as O1, O2, O3, …, O m , and the attributes are represented by lowercase letters a, b, c, d, e. The specific collection and establishment method is as follows: In the embodiment, the value of N is 5, and the attributes are represented by a, b, c, d, e. The 5 attributes of M identity objects are collected in sequence. First, collect the a attribute of M identity objects and uniquely identify the identity with a standardized name. Secondly, collect the b attribute of M identity objects and also standardize the name. And so on, collect the remaining attributes of M objects until the 5 attributes are collected, and then an M identity object library can be formed, where C(i,j) represents the jth-dimensional attribute of the ith object.

[0090] Step 2. Establish the feature templates of the identity attributes of M objects to form N feature templates. The establishment method is as follows: In the embodiment, the value of N is 5. Sequentially extract the features corresponding to the 5 identity attributes of M identity objects. The selection of features is determined according to the type of attributes. One feature template is established for each type of attribute. The corresponding features and identity identifiers are saved in the feature template to form 5 feature templates, which are respectively denoted as F(1), F(2), F(3), F(4), F(5) and stored in the recognition result set topL.

[0091] Step 3. Establish an optimal depth predictor and train it so that for each input sample, the optimal depth of the sample can be predicted.

[0092] Step 4. Given the N-dimensional identity attribute X(j) of the object to be tested X, where j = 1 to N, among these N dimensions of attributes, there are both normal attributes and false attributes. The number of false attributes N Fake , satisfies the condition N Fake ≥50%×N. The prerequisite of this method requires at least two or more normally collected attributes so that the true attributes will converge.

[0093] Step 5. For the N-dimensional identity attributes X(j) of the object X to be measured, when N = 5, the identity attributes X(j) are expressed as: X(1), X(2), X(3), X(4), X(5). Select the recognition algorithms corresponding to the attributes to obtain N recognition results corresponding to the attributes. Select the top L identities with the similarity ranking of the first features corresponding to each, and identify the recognition result set topL.

[0094] Step 6. Input the recognition result set topL into the optimal depth trainer to predict the optimal depth S of the sample.

[0095] Step 7. For the recognition result set topL obtained in Step 5, perform step-by-step consensus calculation to obtain the most likely consensus identity Identity. Since the proportion of false attributes is relatively large in the scenario of this project, it is not excluded that there are two or more consensus identities. Therefore, it is necessary to calculate the convergence degree to determine the unique consensus identity;

[0096] Step 7.1. Initialize the consensus identity, consensus value, and convergence degree value. Take the M identities as the consensus identities, and set their consensus values to 0, that is, d[O i = 0, and the convergence degree value is the same, that is, d_con[O i = 0.

[0097] Step 7.2. Obtain the identities with the consensus value d[O i ≥2 through the step-by-step consensus algorithm. The implementation method is as follows. Traverse the recognition results R a , R b , R c , R d , R e step by step according to the step-by-step consensus algorithm, update the consensus value according to the number of occurrences of the identities in the recognition results, and retain the identities with the consensus value d[O i ≥2.

[0098] Step 7.3. Judge whether there are multiple identities with the consensus value d[O i ≥2. If not, that is, there is only one consensus identity, output the obtained consensus identity Identity and enter Step 8. If there are multiple identities, enter Step 7.4.

[0099] Step 7.4. Calculate the local convergence degree values of the corresponding sequences of multiple identities O m , O t respectively. The convergence degree calculation formula is as follows.

[0100] The convergence degree of object m on multi-dimensional attributes is calculated by adding the convergence degree values K(m,n) of each of the N dimensions and taking the average, which is the convergence degree K(m) between the object X to be measured and the m-th object in the dataset. The value range of the convergence degree is 0 < K(m) ≤ 1, and the smaller the value, the greater the convergence degree.

[0101]

[0102] Among them, s represents that the identity matching result of object m is ranked at the s-th position (selected from the first L positions). If object m does not appear in the first L positions of the attribute set, then s = L. Then:

[0103]

[0104] In the experiment, the value of L is judged by experience. In the case of high recognition algorithm accuracy and ideal acquisition environment, the value of L is relatively low, generally around 5. When the recognition algorithm accuracy is low or the acquisition sample environment is poor, the value of L is relatively high, generally around 10. The subsequent values are considered not to be within the range of convergence, so they are not considered.

[0105] Step 7.5, judge the size of the convergence degree value, and take the corresponding identity with the smallest convergence degree value as the consensus identity Identity, and enter step 8.

[0106] Step 8, judge whether the consensus identity identity is in the recognition result set topL. If the consensus identity identity is in the R of the recognition result set topL of the to-be-detected attribute a1 a , set the feature flag flag_a = True, otherwise set flag_a = False. Similarly, the feature flags of the to-be-detected attributes b1 and c1 can be obtained.

[0107] In the embodiment, the value of N is 5, and the recognition results of the to-be-detected attributes X(1), X(2), X(3), X(4), X(5) are respectively R a , R b , R c , R d , R e , combined with the identity obtained in step 5, judge in turn whether the consensus identity identity is in the R of the recognition result sets topL of the to-be-detected attributes X(1), X(2), X(3), X(4), X(5) a , R b , R c , R d , R eAmong them, obtain the feature identifiers flag_a, flag_b, flag_c, flag_d, flag_e of the attributes X(1), X(2), X(3), X(4), X(5) to be detected.

[0108] Step 9, determine the feature identifiers obtained in Step 6. If the feature identifier flag_a is False, then attribute a1 is considered a false attribute; otherwise, attribute a1 is considered a true attribute. Similarly, determine whether the remaining attributes to be detected, such as b1, c1..., are true or false attributes.

[0109] In the embodiment, the value of N is 5. According to the feature identifiers flag_a, flag_b, flag_c, flag_d, flag_e obtained in Step 6 in sequence, determine the truth or falsehood of the attributes X(1), X(2), X(3), X(4), X(5) to be detected.

[0110] Step 10, output the detected false identity attributes, and the consensus identity Identity is used as the trusted identity. The false detection and identity calculation are completed.

[0111] By using the false identity attribute detection method provided by the present invention, first establish a known object identity attribute library. Through the above steps, it is possible to detect the multi-dimensional attributes of the obtained unknown identity object, discover the false attributes contained therein, and calculate its true identity. The algorithm execution efficiency is high, which is conducive to realizing the trusted identification of identities to a certain extent.

[0112] In a second aspect, the present invention provides an authentication device for collaborative identity identification and false attribute detection. The device includes: an identification result set acquisition module, an optimal depth calculation module, an optimal consensus identity acquisition module, a feature identifier acquisition module, and a feature identifier output module.

[0113] The identification result set acquisition module is used to obtain multiple first features corresponding to the multi-dimensional identity attributes of the object to be measured, search for multiple pieces of data corresponding to each first feature in the database, and merge the multiple pieces of data into an identification result set topL.

[0114] The optimal depth calculation module is used to input the identification result set topL into a trained depth trainer to obtain the optimal depth.

[0115] The optimal consensus identity acquisition module is used to calculate the identification result set topL and the optimal depth by using the step-by-step consensus method to obtain a consensus identity set, and calculate the convergence degree of each consensus identity in the consensus identity set respectively, and determine the consensus identity with the smallest convergence degree as the optimal consensus identity.

[0116] A feature identification acquisition module, configured to determine whether multiple second features of an optimal consensus identity are in the first set topδ, obtain a determination result, and assign a feature identifier according to the determination result, respectively obtaining the feature identifiers of multiple second features of the optimal consensus identity; wherein,

[0117] A feature identifier output module, configured to output the feature identifiers corresponding to multiple second features of the optimal consensus identity, and complete the identity authentication of the object to be tested.

[0118] The structure of the present invention can be three modules: an identity recognition module, a sample optimal depth prediction module, and a false attribute detection module. The identity recognition module includes obtaining the identity attribute information of an unknown object, as well as sorting the identity recognition results, and sorting the identity recognition results of each attribute.

[0119] The sample optimal depth prediction module consists of two parts: prediction network training and sample prediction. The training and prediction of this part are still implemented by integrating existing open-source codes, and the optimal depth prediction of the identity recognition results of each attribute of the input identity is performed. Since the sample data in different forgery situations are significantly different, we can predict the optimal depth through different sample differences. Then, false attribute detection is performed based on the consensus identity obtained from the predicted depth, and the credible identity is calculated.

[0120] The false attribute detection module includes aggregating consensus calculation based on the predicted optimal depth and forgery detection based on the position of the consensus identity. For the false attribute detection module, we obtain the consensus identity of the unknown object based on the aggregation consensus algorithm, and then perform false attribute detection according to the position of the consensus identity in the recognition result sequence.

[0121] The present invention belongs to unsupervised learning. Different from the traditional method that mostly relies on the idea of "prior knowledge and data" to learn the decision boundary, the present invention focuses on studying the distribution law and differences of errors such as algorithms, environments, and attacks in the high-dimensional data space, and proposes a conjecture that "the optimal depths of true aggregation of different samples in the high-dimensional space are different". Traditional false attribute detection methods belong to supervised learning, and the present invention better solves the problem of attribute forgery through unsupervised learning by studying the differences in the optimal aggregation depths of different samples in the high-dimensional space.

[0122] Traditional methods have achieved good results in solving known attacks, but unknown attacks are difficult to effectively prevent through existing experience and data. The present invention uses unsupervised learning to explore the characterization mechanism of the human attack dimension at the decision-making level based on the multi-dimensional identity identification data of the object, so as to solve the problem of unknown attack detection.

[0123] The present invention solves the problem of credible identity identification under the condition that the false attributes are greater than 50% faster and more accurately. The previous step-by-step consensus algorithm has well solved the problem of credible identity identification under the condition that the proportion of false attributes is less than 50% and greater than 50%. The present invention further solves the problems of poor previous effect and low efficiency. According to the principle that the optimal convergence depth of different samples is different in the high-dimensional space, false attribute detection and credible identity calculation are carried out, and more complex scenarios can be adapted compared with the previous methods.

[0124] Some modules in the device of the present invention can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that execute specific tasks or implement specific abstract data types. The present invention can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0125] From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary hardware. Based on such an understanding, the technical solution of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product, or can also be embodied in the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0126] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, reference can be made to each other. The focus of each embodiment is to illustrate the differences from other embodiments. All or part of the present invention can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, mobile communication terminals, multi-processor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on.

[0127] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting the present invention; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. An authentication method for collaborative identity recognition and false attribute detection, characterized in that, Including: Obtain multiple first features corresponding to the multi-dimensional identity attributes of the object to be tested, search for multiple pieces of data corresponding to each of the first features in the database, and merge the multiple pieces of data into an identification result set topL; Input the identification result set topL into a trained deep trainer to obtain an optimal depth; Use a step-by-step consensus method to calculate the identification result set topL and the optimal depth to obtain a consensus identity set, and calculate the convergence degree of each consensus identity in the consensus identity set respectively, and determine the consensus identity with the smallest convergence degree as the optimal consensus identity; Determine whether multiple second features of the optimal consensus identity are in the top δ of the first set to obtain a judgment result, and assign a feature identifier according to the judgment result to respectively obtain the feature identifiers of the multiple second features of the optimal consensus identity; wherein, Output the feature identifiers corresponding to the multiple second features of the optimal consensus identity to complete the identity authentication of the object to be tested.

2. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, wherein The database includes multiple pieces of data of known objects, and each piece of data of the known object includes: a consensus identity, multi-dimensional identity attributes, and multiple third features corresponding to the multi-dimensional identity attributes.

3. The authentication method for collaborative identity recognition and false attribute detection according to claim 2, wherein The searching for multiple pieces of data corresponding to each of the first features in the database and merging the multiple pieces of data into an identification result set topL specifically includes: Calculate the similarity between each first feature in the multiple first features and the same features in the multiple third features of each piece of data in the database respectively, and sort the known objects according to the similarity to obtain the sorting of the known objects corresponding to each of the first features; Determine multiple pieces of data with the sorting of the known objects corresponding to each of the first features being the top L, and form the multiple pieces of data into an identification result set topL.

4. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, characterized in that, The using a step-by-step consensus method to calculate the identification result set topL and the optimal depth to obtain a consensus identity set specifically includes: Initialize the consensus identity set, the consensus value set, and the convergence degree value set; where the size of the consensus value set is the same as the number of data in the database; In the identification result set topL, search for the data with the feature values equal to the multiple first features respectively to obtain multiple pieces of searched data; Count the number of times the same data appears in the multiple pieces of searched data, and record the statistical result in the consensus value set; Combined with the optimal depth, obtain the data in the consensus value set whose statistical result is greater than or equal to the second threshold, and store the corresponding known objects in the consensus identity set to obtain the consensus identity set.

5. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, characterized in that The calculating the convergence degree of each consensus identity in the consensus identity set respectively specifically includes: Judge whether the multiple second features of the multi-dimensional identity attributes of the consensus identity are in the identification result set topL respectively. If not, assign the second feature convergence degree corresponding to the second feature as 1, and the value range of the convergence degree is (0, 1]; If so, calculate the convergence degree of the multiple second features of the multi-dimensional identity attributes of the consensus identity respectively to obtain multiple second feature convergence degree values, and add the multiple second convergence degree values and take the average to obtain the convergence degree of the consensus identity.

6. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, characterized in that, The convergence degree is specifically expressed as: Among them, s represents the sorting position of the known object m in the recognition result set topL; N represents the dimension of the multi-dimensional identity attribute; L represents the size of the recognition result set topL; K(m) represents the convergence degree of the known object m.

7. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, wherein The size of the first set topδ is expressed as: δ = min(8 + S, L) Among them, S represents the optimal depth; L represents the size of the recognition result set topL.

8. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, wherein Assigning a feature identifier according to the judgment result specifically includes: If the judgment result is yes, set the feature identifier of the feature to true; If the judgment result is no, set the feature identifier of the feature to false.

9. The authentication method for collaborative identity recognition and false attribute detection according to claim 1, wherein The multi-dimensional identity attributes of the object to be tested include real attributes and false attributes.

10. An authentication device for collaborative identity recognition and false attribute detection, characterized in that, It includes: A recognition result set acquisition module, configured to acquire multiple first features corresponding to the multi-dimensional identity attributes of the object to be tested respectively, search for multiple pieces of data corresponding to each of the first features in the database, and merge the multiple pieces of data into a recognition result set topL; An optimal depth calculation module, configured to input the recognition result set topL into a trained depth trainer to obtain an optimal depth; An optimal consensus identity acquisition module, configured to use a step-by-step consensus method to calculate the recognition result set topL and the optimal depth to obtain a consensus identity set, and calculate the convergence degree of each consensus identity in the consensus identity set respectively, and determine the consensus identity with the smallest convergence degree as the optimal consensus identity; A feature identification acquisition module, configured to determine whether multiple second features of the optimal consensus identity are in the first set topδ, obtain a determination result, and assign a feature identifier according to the determination result, respectively obtaining the feature identifiers of the multiple second features of the optimal consensus identity; wherein, A feature identifier output module, configured to output the feature identifiers corresponding to the multiple second features of the optimal consensus identity, and complete the identity authentication of the object to be tested.

Images