CN111582312A - Secure biological hash code generation method for resisting relation attack based on periodic function - Google Patents

Secure biological hash code generation method for resisting relation attack based on periodic function Download PDF

Info

Publication number
CN111582312A
CN111582312A CN202010272440.8A CN202010272440A CN111582312A CN 111582312 A CN111582312 A CN 111582312A CN 202010272440 A CN202010272440 A CN 202010272440A CN 111582312 A CN111582312 A CN 111582312A
Authority
CN
China
Prior art keywords
hash
biological
network
data set
function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010272440.8A
Other languages
Chinese (zh)
Other versions
CN111582312B (en
Inventor
沃焱
汪强降
韩国强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202010272440.8A priority Critical patent/CN111582312B/en
Publication of CN111582312A publication Critical patent/CN111582312A/en
Application granted granted Critical
Publication of CN111582312B publication Critical patent/CN111582312B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Collating Specific Patterns (AREA)

Abstract

The invention discloses a secure biological hash code generation method for resisting relation attack based on a periodic function, which comprises the following steps: 1) extracting a feature vector from the biological image by using a residual error network to obtain a biological feature data set; 2) inputting the biological characteristic data set into a Hash generation network for forward propagation; 3) calculating a loss function, reversely propagating and calculating a gradient, and updating a weight; 4) repeating the steps 2) to 3) until convergence, and obtaining a Hash generation model; 5) inputting the biological characteristic data set into the Hash generation model obtained in the step 4), namely obtaining a corresponding biological Hash code, wherein the Hash code has high distinguishing capacity and can be used for biological identification. The method not only has strong defense capability to the attack based on the relationship, but also has better identification performance compared with the existing biological characteristic hash method. Experiments are carried out on the iris data set CASIA-v4-interval and the face data set LFW, and the effectiveness of the method can be guaranteed.

Description

Secure biological hash code generation method for resisting relation attack based on periodic function
Technical Field
The invention relates to the technical field of biological characteristic image processing, in particular to a secure biological hash code generation method for resisting relation attack based on a periodic function.
Background
Biometric identification refers to the use of human biometric features, such as irises and faces, to identify a person who may be authorized to access a system, device or data. Since biometrics is unique to individuals and is easier to use than passwords, biometrics is widely used as an authentication method in banks, hospitals, social network sites, and the like. In order to make a one-to-many comparison, biometric identification systems typically need to assess similarity between the input query and the database records. The biological hash technology is widely applied to a biological recognition system due to the advantages of small calculation amount, high storage efficiency and the like.
However, due to the high sensitivity and invariance of biometric data, the widespread use of biometric technology has raised concerns about privacy leakage caused by biometric technology. Although the specific algorithms are different, most of the bio-hashes essentially require that the similarity or distance in the original biometric space be maintained in the hash space to obtain higher distinguishing capability and matching accuracy, which is called relationship maintenance. Based on this relationship preservation, an attacker can approach the attacked target by reducing the difference between the two original biometrics by reducing the difference between the hashes. We collectively refer to such attacks as relationship-based attacks.
There are several methods currently used to defend against relationship-based attacks to secure biometric features. These methods can be broadly divided into encryption and non-encryption. The former is mostly based on homomorphic encryption and scrambling code circuits, and can process encrypted biological characteristic data on the basis of no need of decryption. Since there is no distance measurement in the encrypted domain, attacks can be effectively prevented. However, in a one-to-many recognition scenario, the similarity search must return a number of candidates that need to be ranked in terms of similarity to the query. This process not only involves a dense distance calculation, but also a dense distance comparison, which is particularly troublesome in the encrypted domain. The latter, i.e. non-cryptographic methods, are typically improved on the basis of the normal hash, taking security or privacy into account at the beginning of the design, and are typically based on quantized random projections and locality sensitive hashes. However, the method based on the quantitative random projection is independent of data in the random projection process, and the identification performance of the method has a great space for improvement. The locality sensitive hashing based method is subject to artificial adjustment depending on threshold values and probabilities, and requires a larger coding space and a longer coding length to realize the constraints of locality sensitive hashing. For large data sets, the storage and calculation requirements of the method are high.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method for generating a secure biological hash code based on a periodic function and capable of resisting relation attack. By maximizing the minimum entropy of the sign of the distance difference in the original space and the hash space, the leakage of distance relationships over the original biometric can be minimized. The optimization goal can be realized by using a periodic function as a distance mapping and then using a deep neural network for training. The method not only has strong capability of defending the attack based on the relationship, but also has excellent biological characteristic identification performance.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: the secure biological hash code generation method for resisting relation attack based on the periodic function comprises the following steps:
1) extracting a feature vector from the biological image by using a residual error network to obtain a biological feature data set;
2) inputting the biological characteristic data set into a Hash generation network for forward propagation;
3) calculating a loss function, reversely propagating and calculating a gradient, and updating a weight;
4) repeating the steps 2) to 3) until convergence, and obtaining a Hash generation model;
5) inputting the biological characteristic data set into the Hash generation model obtained in the step 4), namely obtaining a corresponding biological Hash code, wherein the Hash code has high distinguishing capacity and can be used for biological identification.
In step 1), extracting by using a residual error network ResNet50Obtaining a set of biometric data, specifically operating to: let O be { O ═ Oi|oi1,2,3, N represents the original image dataset, oiRepresenting one sample or data point in O, N representing the number of samples in the original data set, inputting these data to a residual network ResNet50 to obtain a corresponding feature data set X, wherein the residual network ResNet50 used needs to be pre-trained on ImageNet in advance, and
Figure BDA0002443588990000031
in the formula xiRepresents one sample or one data point in X,
Figure BDA0002443588990000032
representing the real number domain, m representing the dimension of a single biometric, N representing the number of biometrics, outputs the biometric dataset, i.e. outputs X to the next step.
In step 2), inputting the biological characteristic data set into a Hash generation network for forward propagation, wherein the Hash generation network is composed of a full-connection neural network, a quantization function and a loss function, and specifically, a training set, a test set and a verification set are separated from X by using the biological characteristic data set X obtained in the previous step and using a K-fold cross validation method, and then two data X are arbitrarily taken from the training seti,xjAnd inputting the two data into two full-connection neural networks sharing parameters respectively, and after the data pass through the full-connection neural networks, delivering the data to a quantization function for processing to obtain corresponding binary strings yi,yjWhere the quantization function is q, numbers greater than 0 are mapped to 1, other numbers are mapped to 0, and the parameter trained in the fully-connected network is the matrix W, and the following equation holds:
yi＝q(Wxi) (1)
yj＝q(Wxj) (2)。
in step 3), a loss function is calculated, the gradient is propagated reversely, and the weight is updated, wherein the loss function is defined as follows:
Figure BDA0002443588990000033
where q is a quantization function, W represents a parameter matrix in a fully-connected network, xi,xjRepresenting input biometric features, f1And f2Respectively representing a linear function and a periodic function, wherein the periodic function refers to the fact that the value of a dependent variable of the function shows periodic change along with the increase of an independent variable, and specifically refers to the fact that the function f2There is a non-zero constant T such that any two arguments x and x + T present in the domain satisfy f2(x+T)＝f2(x) The two functions can be selected according to the actual requirements, and in addition, CijIs represented by xi,xjWhether the model is homogeneous or heterogeneous, if the model is homogeneous, the value is 0, and if the model is heterogeneous, the value is 1, after the value of the loss function is solved, the partial derivative of the weight of the corresponding model is solved through back propagation, and the model parameter or the weight in the network is updated through a random gradient descent method, as shown in the following formula:
Figure BDA0002443588990000041
where t denotes the number of iterations ηtIndicates the learning rate, Wt+1And WtRespectively representing the parameter weight matrix of the neural network at the t +1 th iteration and the t th iteration, JtRepresenting the value of the loss function calculated at the t-th iteration.
In step 4), step 2) and step 3) are interleaved using a K-fold cross-validation method until the value of the loss function satisfies the following condition:
|Jt+1-Jt|＜ (5)
in the formula, Jt+1And JtRepresents the calculated loss function values for the t +1 th and the t-th iterations, respectively, but is a hyperparameter, representing an error or threshold, below which the network is deemed to have converged; and after the network is converged, storing the whole network to obtain the Hash generation model.
In step 5), the whole Hash generation model is regarded as a Hash generation function h, and then the biological characteristic data set is processed
Figure BDA0002443588990000042
Inputting the hash values into h to obtain a corresponding hash set:
Y＝{yi|yi∈{0,1}n,i＝1,2,3,...,N} (6)
wherein x isiRepresents one sample or one data point in X,
Figure BDA0002443588990000043
representing a real number domain, m representing a dimension of a single biometric, N representing a number of biometrics; y isi＝h(xi) Denotes xiThe corresponding hash code is a binary string, h is the entire hash generation model, and n represents the dimension of the generated hash code.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the two-stage distance mapping based on the periodic function provided by the invention simultaneously considers the safety and the usability, and realizes high-precision biological feature identification on the basis of protecting the biological features from being leaked.
2. The Hash generation network provided by the invention uses the weak label for learning, and has a wider application range. At the same time, the network parameters are shared for both inputs, simplifying the network.
3. The method is easy to realize and has better generalization capability on unseen samples.
In summary, the invention constructs a complete hash learning framework to describe the hash method for defending the relational attack, and realizes the corresponding secure hash aiming at the specific biological characteristic identification problem, and the secure hash uses a periodic function as distance mapping to respectively map in classes and among classes, thereby realizing the defense for the relational attack on the basis of certain availability. Therefore, in the big data era, the private data can bring more convenient and personalized services to people under the condition of not revealing individual privacy.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a hash generation network architecture diagram of the present invention.
Detailed Description
The present invention will be further described with reference to the following specific examples.
As shown in fig. 1 and fig. 2, the method for generating a secure biological hash code based on a periodic function and resisting a relational attack according to the present embodiment includes the following steps:
1) extracting a feature vector from the biological image by using a residual error network to obtain a biological feature data set, wherein the method specifically comprises the following operations:
let O be { O ═ Oi|oi1,2,3, N represents the original image dataset, where oiRepresenting one sample or data point in O, N represents the number of samples in the original data set, inputting the data to a residual network ResNet50 to obtain a corresponding biological feature data set X, wherein the used residual network ResNet50 is pre-trained on ImageNet in advance, ResNet50 has 2 basic blocks, one is Identity Block, the input and output dimensions are the same, so that a plurality of basic blocks can be connected in series, the other basic Block is Conv Block, the input and output dimensions are different, so that the basic blocks cannot be connected in series, and the function of the method is to change the dimension of the feature vector. While
Figure BDA0002443588990000051
In the formula xiRepresents one sample or one data point in X,
Figure BDA0002443588990000052
representing the real number domain, m representing the dimension of a single biometric, N representing the number of biometrics, outputs the biometric dataset, i.e. outputs X to the next step.
2) Using the biological characteristic data set to carry out forward propagation on the Hash generation network, and specifically operating as follows:
the Hash generation network is composed of a fully-connected neural network, a quantization function and a loss function, as shown in figure 2, a biological characteristic data set X obtained in the last step is utilized, a K-fold cross validation method is used, and two data X are randomly selected from a training seti,xjAnd inputting the two parameters into two full-connection neural networks sharing the parameters respectively, and after passing through the neural networks, delivering the data to a quantization function for processing to obtain corresponding binary strings yi,yjWhere the quantization function is q, numbers greater than 0 are mapped to 1, other numbers are mapped to 0, and the parameter trained in the fully-connected network is the matrix W, and the following equation holds:
yi＝q(Wxi) (1)
yj＝q(Wxj) (2)
3) and calculating a loss function, reversely propagating and calculating a gradient, and updating the weight. Wherein the loss function is defined as follows:
Figure BDA0002443588990000061
where q is a quantization function, W represents a parameter matrix in a fully-connected network, xi,xjRepresenting input biometric features, f1And f2Respectively representing a linear function and a periodic function, wherein the periodic function refers to the fact that the value of a dependent variable of the function shows periodic change along with the increase of an independent variable, and specifically refers to the fact that the function f2There is a non-zero constant T such that any two arguments x and x + T present in the domain satisfy f2(x+T)＝f2(x) The two functions can be selected according to actual requirements. CijIs represented by xi,xjWhether the same type or different type exists, if the same type exists, the value is 0, and if the same type exists, the value of the different type exists, the value is 1.
After the value of the loss function is obtained, the partial derivative of the weight corresponding to the model can be obtained through back propagation, and the model parameter or the weight in the network is updated through a random gradient descent method, as shown in the following formula:
Figure BDA0002443588990000062
where t denotes the number of iterations ηtIndicates the learning rate, WtRepresenting the parameter weight matrix of the network at the t-th iteration, JtRepresenting the loss function value calculated for the t-th iteration.
4) Interleaving steps 2) and 3) using a K-fold cross-validation method until the value of the loss function satisfies the following condition:
|Jt+1-Jt|＜ (5)
in the formula, Jt+1And JtThe loss function values calculated in the t +1 th iteration and the t th iteration are respectively represented, but the super parameter represents an error or a threshold value, and when the value is lower than the value, the network is considered to be converged; and after the network is converged, storing the whole network to obtain the Hash generation model. In the iterative training process, the parameters of the hash generation network slowly tend to be stable until a network with high distinguishing capability is finally generated.
5) Inputting the biological characteristic data set into a Hash generation model, namely obtaining a biological Hash code for defending the relational attack, wherein the biological Hash code has higher distinguishing capability and can be used for biological identification, and the method specifically comprises the following steps:
treating the whole Hash generation model as a Hash generation function h, and then collecting the biological characteristic data set
Figure BDA0002443588990000071
Inputting the hash values into h to obtain a corresponding hash set:
Y＝{yi|yi∈{0,1}n,i＝1,2,3,...,N} (6)
wherein, yi＝h(xi) Denotes xiThe corresponding hash code is a binary string, h is the entire hash generation model, and n represents the dimension of the generated hash code.
The above-mentioned embodiments are merely preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, so that the changes in the shape and principle of the present invention should be covered within the protection scope of the present invention.

Claims (6)

1. The secure biological hash code generation method for resisting the relation attack based on the periodic function is characterized by comprising the following steps of:
1) extracting a feature vector from the biological image by using a residual error network to obtain a biological feature data set;
2) inputting the biological characteristic data set into a Hash generation network for forward propagation;
3) calculating a loss function, reversely propagating and calculating a gradient, and updating a weight;
4) repeating the steps 2) to 3) until convergence, and obtaining a Hash generation model;
5) inputting the biological characteristic data set into the Hash generation model obtained in the step 4), namely obtaining a corresponding biological Hash code, wherein the Hash code has high distinguishing capacity and can be used for biological identification.
2. The method for generating secure biological hash code against relational attack based on periodic function according to claim 1, wherein: in step 1), a biometric data set is extracted using a residual network ResNet50, specifically operating as: let O be { O ═ Oi|oi1,2,3, N represents the original image dataset, oiRepresenting one sample or data point in O, N representing the number of samples in the original data set, inputting these data to a residual network ResNet50 to obtain a corresponding feature data set X, wherein the residual network ResNet50 used needs to be pre-trained on ImageNet in advance, and
Figure FDA0002443588980000011
in the formula xiRepresents one sample or one data point in X,
Figure FDA0002443588980000012
representing the real number domain, m represents the dimension of a single biometric feature,n denotes the number of biometrics, and outputs a biometrics data set, i.e., outputs X to the next step.
3. The method for generating secure biological hash code against relational attack based on periodic function according to claim 1, wherein: in step 2), inputting the biological characteristic data set into a Hash generation network for forward propagation, wherein the Hash generation network is composed of a full-connection neural network, a quantization function and a loss function, and specifically, a training set, a test set and a verification set are separated from X by using the biological characteristic data set X obtained in the previous step and using a K-fold cross validation method, and then two data X are arbitrarily taken from the training seti,xjAnd inputting the two data into two full-connection neural networks sharing parameters respectively, and after the data pass through the full-connection neural networks, delivering the data to a quantization function for processing to obtain corresponding binary strings yi,yjWhere the quantization function is q, numbers greater than 0 are mapped to 1, other numbers are mapped to 0, and the parameter trained in the fully-connected network is the matrix W, and the following equation holds:
yi＝q(Wxi) (1)
yj＝q(Wxj) (2)。
4. the method for generating secure biological hash code against relational attack based on periodic function according to claim 1, wherein: in step 3), a loss function is calculated, the gradient is propagated reversely, and the weight is updated, wherein the loss function is defined as follows:
Figure FDA0002443588980000021
where q is a quantization function, W represents a parameter matrix in a fully-connected network, xi,xjRepresenting input biometric features, f1And f2Respectively representing a linear function and a periodic function, wherein the periodic function refers to that the value of a dependent variable of the function follows an independent variableShows a periodic variation, in particular the function f2There is a non-zero constant T such that any two arguments x and x + T present in the domain satisfy f2(x+T)＝f2(x) The two functions can be selected according to the actual requirements, and in addition, CijIs represented by xi,xjWhether the model is homogeneous or heterogeneous, if the model is homogeneous, the value is 0, and if the model is heterogeneous, the value is 1, after the value of the loss function is solved, the partial derivative of the weight of the corresponding model is solved through back propagation, and the model parameter or the weight in the network is updated through a random gradient descent method, as shown in the following formula:
Figure FDA0002443588980000022
where t denotes the number of iterations ηtIndicates the learning rate, Wt+1And WtRespectively representing the parameter weight matrix of the neural network at the t +1 th iteration and the t th iteration, JtRepresenting the value of the loss function calculated at the t-th iteration.
5. The method for generating secure biological hash code against relational attack based on periodic function according to claim 1, wherein: in step 4), step 2) and step 3) are interleaved using a K-fold cross-validation method until the value of the loss function satisfies the following condition:
|Jt+1-Jt|＜ (5)
in the formula, Jt+1And JtRepresents the calculated loss function values for the t +1 th and the t-th iterations, respectively, but is a hyperparameter, representing an error or threshold, below which the network is deemed to have converged; and after the network is converged, storing the whole network to obtain the Hash generation model.
6. The method for generating secure biological hash code against relational attack based on periodic function according to claim 1, wherein: in step 5), the whole hash generation model is regarded asA hash is generated to a function h, and then the biometric data set is generated
Figure FDA0002443588980000031
Inputting the hash values into h to obtain a corresponding hash set:
Y＝{yi|yi∈{0,1}n,i＝1,2,3,...,N} (6)
wherein x isiRepresents one sample or one data point in X,
Figure FDA0002443588980000032
representing a real number domain, m representing a dimension of a single biometric, N representing a number of biometrics; y isi＝h(xi) Denotes xiThe corresponding hash code is a binary string, h is the entire hash generation model, and n represents the dimension of the generated hash code.
CN202010272440.8A 2020-04-09 2020-04-09 Secure biological hash code generation method for resisting relation attack based on periodic function Active CN111582312B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010272440.8A CN111582312B (en) 2020-04-09 2020-04-09 Secure biological hash code generation method for resisting relation attack based on periodic function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010272440.8A CN111582312B (en) 2020-04-09 2020-04-09 Secure biological hash code generation method for resisting relation attack based on periodic function

Publications (2)

Publication Number Publication Date
CN111582312A true CN111582312A (en) 2020-08-25
CN111582312B CN111582312B (en) 2023-04-07

Family

ID=72120495

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010272440.8A Active CN111582312B (en) 2020-04-09 2020-04-09 Secure biological hash code generation method for resisting relation attack based on periodic function

Country Status (1)

Country Link
CN (1) CN111582312B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115633129A (en) * 2022-10-13 2023-01-20 同济大学 Stack type sparse self-encoder and GAN chaotic sequence image encryption method and device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109918528A (en) * 2019-01-14 2019-06-21 北京工商大学 A kind of compact Hash code learning method based on semanteme protection
CN110188227A (en) * 2019-05-05 2019-08-30 华南理工大学 A kind of hashing image search method based on deep learning and low-rank matrix optimization

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109918528A (en) * 2019-01-14 2019-06-21 北京工商大学 A kind of compact Hash code learning method based on semanteme protection
CN110188227A (en) * 2019-05-05 2019-08-30 华南理工大学 A kind of hashing image search method based on deep learning and low-rank matrix optimization

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115633129A (en) * 2022-10-13 2023-01-20 同济大学 Stack type sparse self-encoder and GAN chaotic sequence image encryption method and device
CN115633129B (en) * 2022-10-13 2024-03-05 同济大学 Stacked sparse self-encoder and GAN chaotic sequence image encryption method and device

Also Published As

Publication number Publication date
CN111582312B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
Lee et al. Defending against neural network model stealing attacks using deceptive perturbations
Mireshghallah et al. Shredder: Learning noise distributions to protect inference privacy
Agrawal et al. QUOTIENT: Two-party secure neural network training and prediction
Vatsalan et al. Privacy-preserving record linkage for big data: Current approaches and research challenges
Wang et al. CloudBI: Practical privacy-preserving outsourcing of biometric identification in the cloud
Yang et al. A comprehensive survey on secure outsourced computation and its applications
US8165352B1 (en) Reconstruction of biometric image templates using match scores
Engelsma et al. HERS: Homomorphically encrypted representation search
Zhao et al. PrivateDL: privacy‐preserving collaborative deep learning against leakage from gradient sharing
Feng et al. Masquerade attack on transform-based binary-template protection based on perceptron learning
Yang et al. Effectiveness of distillation attack and countermeasure on neural network watermarking
Ding et al. Privacy-preserving feature extraction via adversarial training
Jain et al. Adversarial Detection Avoidance Attacks: Evaluating the robustness of perceptual hashing-based client-side scanning
Zhang et al. Privacy threats and protection in machine learning
Liu et al. Face image publication based on differential privacy
CN111582312B (en) Secure biological hash code generation method for resisting relation attack based on periodic function
Gu et al. FedPass: privacy-preserving vertical federated deep learning with adaptive obfuscation
CN117494183A (en) Knowledge distillation-based privacy data generation method and system for generating countermeasure network model
Duan et al. SSGD: A safe and efficient method of gradient descent
Cui et al. Deep index-of-maximum hashing for face template protection
Li et al. One factor cancellable fingerprint scheme based on novel minimum hash signature and secure extended feature vector
CN114398660A (en) High-efficiency fuzzy searchable encryption method based on Word2vec and ASPE
Yu et al. Secure biometric hashing against relation-based attacks via maximizing min-entropy
Yao et al. SNN-PPRL: A secure record matching scheme based on siamese neural network
Zhu et al. A performance-optimization method for reusable fuzzy extractor based on block error distribution of iris trait

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant