SM4 algorithm-based privacy protection lightweight face authentication method

Abstract

The invention discloses a privacy protection lightweight face authentication method based on an SM4 algorithm, and belongs to the technical field of face authentication. In a user registration stage, a user uploads a face image to a client through a sensor device, the client outputs a face feature vector representing an original image by loading a pre-trained FaceNet model, and the face feature vector is used as a face feature vector representing the original image; and then a hash value is calculated through a p-stable LSH function. And the client calculates a ciphertext for registration through a symmetric encryption algorithm MS4, and finally sends the ciphertext of the face features and the corresponding hash value to the server. And in the identity authentication process, the client judges whether elements smaller than a threshold value exist in the similarity set or not after decryption, and if yes, authentication is passed. Otherwise, the authentication fails. By means of the mode, the data are encrypted through the security encryption algorithm SM4, it is ensured that the data are protected and cannot be tampered or leaked in the transmission process, and therefore the integrity and privacy of the data are maintained.

Description

Technical Field

[0001] The present invention relates to the technical field of face authentication, and in particular to a privacy-preserving lightweight face authentication method based on an SM4 algorithm. Background Art

[0002] Face recognition technology has very important theoretical significance and application value. The research on face recognition has a significant driving force for the development of image processing, pattern recognition, computer vision, computer graphics and other fields. It is also widely used in various fields such as biometric authentication, video surveillance, and security. It has brought great convenience to people's lives and demonstrated considerable commercial value and huge development potential.

[0003] However, with the continuous development of big data technology, data is growing exponentially. With the continuous emergence of privacy leaks and other problems, people are paying more and more attention to individual privacy, and issues related to privacy protection are becoming more and more urgent to be solved.

[0004] At present, face recognition systems still face the following important problems: data is at risk of being tampered with or leaked during transmission, and privacy is difficult to guarantee.

[0005] Based on this, the present invention designs a privacy-preserving lightweight face recognition method based on the SM4 algorithm to solve the above problems. Summary of the invention

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention provides a privacy-preserving lightweight face authentication method based on the SM4 algorithm.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A privacy-preserving lightweight face authentication method based on the SM4 algorithm comprises the following steps:

[0009] In the initialization phase, the server selects the FaceNet model and the p-stable LSH function and deploys them on each client. Each client generates a long-term private key MK combined with a timestamp and expiration information and stores it in the secret storage area.

[0010] During the user registration phase, the user uploads the face image to the client through the sensor device. After the client obtains the face image, it loads the pre-trained FaceNet model, outputs the face feature vector representing the original image, and then calculates the hash value through the p-stable LSH function; the client calculates the ciphertext for registration through the symmetric encryption algorithm MS4, and finally sends the ciphertext of the face feature and the corresponding hash value to the server; the server then finds the corresponding classification set based on the hash value and stores the ciphertext of the face feature in the set;

[0011] During the identity authentication process, the client obtains the hash value of the user's face image and the ciphertext of the face features, and sends them to the server; the server quickly matches the face to the classification set based on the hash value, and calculates the ciphertext of the ciphertext and the ciphertexts in the set one by one in the set to obtain a ciphertext set of similarity, and returns the set to the client. After decryption, the client obtains the similarity between the face features and the elements in the classification set, and determines whether there are elements in the similarity set that are less than the threshold. If so, the authentication is successful and the protocol terminates; otherwise, the client sends the verification failure result to the server.

[0012] Furthermore, in step S1, the client generates a combination of a long-term private key MK and a timestamp and expiration date information:

[0013] K=H(MK||T||PT)

[0014] Among them, T is the current timestamp and PT is the validity period.

[0015] Furthermore, we select a random number r and then a hash function H:{0,1} * →{0,1} l To generate authentication information; let E k (m) represents the ciphertext of m encrypted using the key k of the symmetric encryption algorithm MS4, D k (m) represents the plaintext decrypted using the key k of the corresponding symmetric encryption algorithm MS4; the block size of the symmetric encryption algorithm MS4 and the p-stable LSH function h(·) are set to 20 bytes in length, and the secret key is 16 bytes.

[0016] Furthermore, the specific steps of step S2 are:

[0017] Step 2.1, the user sends the facial image and registration request to the client through a dedicated channel, i.e. the camera device of the client;

[0018] Step 2.2: After receiving the facial image, the client first extracts the facial feature vector through the trained FaceNet model; the client uses the p-stable SLH function algorithm to obtain the hash value hi ;

[0019] Then, the client calculates the hash value H(h i ||r), to prevent server-side forgery attacks;

[0020] Next, the client uses the symmetric encryption algorithm MS4 to calculate the ciphertext data for registration, namely E k (H(h i ||r));

[0021] Finally, the client sends the ciphertext of the facial features and the corresponding hash value <h i ,E k (H(h i ||r))>Send to the server;

[0022] Step 2.3: The server receives <h i ,E k (H(h i ||r))>, the server finds the corresponding classification set in the database DB according to the hash value and stores the ciphertext of the facial features in the set.

[0023] Furthermore, the specific steps of step S3 are:

[0024] Step 3.1: The user initiates an authentication request to the client, and the client receives and responds to the request;

[0025] Step 3.2: The client collects multiple facial images of user U through a camera device

[0026] Step 3.3: First, load the pre-trained FaceNet model, adjust the image format to RGB image, and then extract the multi-dimensional facial feature vector A = {V 1 ,V 2 ,V 3 ,…V n}, perform FaceNet facial feature extraction on A to obtain the biological feature vector;

[0027] Step 3.4: Use the obtained biometric vector through the p-stable LSH function to obtain the hash value h i ;

[0028] Step 3.5: The client calculates the hash value H(h i ||r) and ciphertext data E k (H(h i ||r)); Finally, the client sends <h i ,E k (H(hi ||r))> and authentication request;

[0029] Step 3.6: The server receives <h i ,E k (H(h i ||r))>, the server traverses the hash value of each face image data in the database DB and finds the hash value closest to h i The hash value of matches the classification set where the face belongs;

[0030] Then the ciphertext is calculated, and the similarity of the ciphertext M is calculated by Euclidean distance;

[0031] The ciphertext set with similarity <h j ,M||E k (H(h j ||r))>Send to the client;

[0032] Step 3.7: The client receives the traversal results from the server <h j ,M||E k (H(h j ||r))>, the received result is decrypted to determine whether the returned similarity is within the specified threshold. If it is not within the specified threshold, the user authentication fails; otherwise, the user authentication passes.

[0033] Furthermore, in step 3.3, the image format is adjusted to a 224×224×3 png RGB image, and then the multi-dimensional facial feature vector A={V 1 ,V 2 ,V 3 ,…V n}, perform FaceNet facial feature extraction on A to obtain a single 128-dimensional biometric feature vector.

[0034] Furthermore, L2norm processing is performed on a single 128-dimensional biometric feature vector.

[0035] Furthermore, in step 3.4, each hash function in the LSH function family based on p-stable distribution is expressed as:

[0036]

[0037] Furthermore, in step 3.6, the server traverses each data item in the database DB <h x ,E k (H(h x ||r))>(1≤x≤|DB|); calculate h j ={h x |min <hx ; E k (H(h x ||r)∈DB|h x -h i |, get the closest hash value and corresponding ciphertext, that is, <h j ,E k (H(h j ||r))>; where |h x -h j | indicates h x With h j The absolute value between |DB| represents the number of items in the database DB.

[0038] Furthermore, in step 3.6, the Euclidean distance formula is as follows:

[0039]

[0040] Among them, A and B are both vectors.

[0041] Compared with the prior art, the present invention has the following beneficial effects: Facial feature privacy: The present invention uses the secure encryption algorithm SM4 to encrypt data to ensure that the data is protected during transmission and is not tampered with or leaked, thereby maintaining the integrity and privacy of the data. During the authentication stage, the server performs facial similarity calculation in a ciphertext state to ensure that the facial data and similarity have not been tampered with or leaked, thereby ensuring the integrity of the data during the processing process.

[0042] User anonymity: User anonymity means that after the user encrypts and uploads his or her facial feature information, the database server cannot steal the user's facial feature plaintext information. During the authentication stage, the homomorphic encryption algorithm used for face matching is performed in a ciphertext state, and the user's facial feature information will not be leaked.

[0043] Preventing replay attacks: In encrypted communications, timestamps can be used to prevent replay attacks, where an attacker intercepts and resends previous communications. By checking timestamps, the system can reject expired requests.

[0044] Enhanced security: By setting an expiration date, even if a key is obtained by an unauthorized person, there is only a limited window of time in which the key can be used. This increases the security of the system because it limits the damage that can be caused by unauthorized access.

[0045] Key rotation: To prevent keys from being used for a long time and increasing the risk of leakage, it is common practice to change keys regularly. Timestamps and expiration dates can help implement key rotation strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is a schematic diagram of the initialization phase in a privacy-preserving lightweight face authentication method based on the SM4 algorithm of the present invention;

[0048] Figure 2 It is a schematic diagram of the registration phase in a privacy-preserving lightweight face authentication method based on the SM4 algorithm of the present invention;

[0049] Figure 3 This is a schematic diagram of the authentication phase in a privacy-preserving lightweight face authentication method based on the SM4 algorithm of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Terminology explanation:

[0052] SM4 cipher algorithm: a block cipher algorithm with a block length of 128 bits (i.e. 16 bytes, 4 words) and a key length of 128 bits (i.e. 16 bytes, 4 words). The encryption and decryption process uses a 32-round iteration mechanism (similar to DES (block algorithm using key encryption) and AES (Advanced Encryption Standard)), and each round requires a round key (similar to DES and AES).

[0053] FaceNet: A face recognition system based on deep learning, proposed by Google researchers in 2015. The core idea of ​​this model is to map face images into a high-dimensional vector space, so that the face images of the same person are close in space, while the distances between face images of different people are relatively far. The FaceNet model usually uses a convolutional neural network (CNN) to extract image features and uses a triplet loss function to train the network to optimize the effect of feature extraction.

[0054] p-stable LSH (Locality-Sensitive Hashing) function: a special class of hash functions that can efficiently map data points to hash code space while maintaining data locality. In machine learning and data mining, LSH is an important technique for approximate nearest neighbor search in large data sets. p-stable LSH is an LSH function based on p-stable distribution, where p-stable distribution is a probability distribution whose linear combination also obeys the same distribution. The most common p-stable distribution is Gaussian distribution (when p=2), which is what we often call normal distribution. P-stable LSH function is often used to process data in Euclidean space. The specific implementation of p-stable LSH function usually involves random projection, where data points are projected through randomly generated vectors and the projection results are used to determine the hash code. The advantage of this method is that it is simple and easy to parallelize, which is very suitable for processing large-scale data in modern computing environments.

[0055] Face recognition system: includes user, client and server. The user uploads facial features through the client's sensor device; the client is responsible for the registration and authentication of the face recognition system; the server is responsible for storing and managing the ciphertext data of the facial feature values ​​uploaded by the client, and is responsible for similarity calculation in the ciphertext state during the authentication stage.

[0056] Embodiment 1: In some embodiments, please refer to the drawings of the specification Figure 1-Figure 3 , a privacy-preserving lightweight face authentication method based on SM4 algorithm, comprising the following steps:

[0057] In the initialization phase, the server selects the FaceNet model and the p-stable LSH function and deploys them on each client. Each client generates a long-term private key MK combined with information such as timestamp and expiration date, and stores it in the secret storage area.

[0058] During the user registration phase, the user uploads the face image to the client through the sensor device. After the client obtains the face image, it loads the pre-trained FaceNet model, outputs the face feature vector representing the original image, and then calculates the hash value through the p-stable LSH function. The client calculates the ciphertext for registration through the symmetric encryption algorithm MS4, and finally sends the ciphertext of the face feature and the corresponding hash value to the server. The server then finds the corresponding classification set based on the hash value and stores the ciphertext of the face feature in the set.

[0059] During the identity authentication process, as in the registration phase, the client obtains the hash value of the user's face image and the ciphertext of the face features, and sends them to the server. Based on the hash value, the server quickly matches the classification set where the face is located, calculates the ciphertext of the ciphertext and the ciphertexts in the set one by one, obtains the ciphertext set of similarity, and returns the set to the client. After decryption, the client obtains the similarity between the face features and the elements in the classification set, and determines whether there are elements in the similarity set that are less than the threshold. If so, the authentication is passed and the protocol terminates. Otherwise, the client sends the verification failure result to the server.

[0060] Embodiment 2: In some embodiments, Figure 1-Figure 3 As a preferred embodiment of the present invention, a privacy-preserving lightweight face recognition method based on the SM4 algorithm comprises the following steps:

[0061] S1. Initialization phase

[0062] During the initialization phase, the client generates a long-term private key MK and combines it with information such as timestamp and expiration date:

[0063] K=H(MK||T||PT)

[0064] Among them, T is the current timestamp and PT is the expiration date;

[0065] And choose a random number r, and then choose a hash function H: {0,1} * →{0,1} l To generate authentication information; used to prevent the server from forging authentication data.

[0066] Assume E k (m) represents the ciphertext of m encrypted using the key k of the symmetric encryption algorithm MS4, D k (m) represents the plaintext decrypted using the key k of the corresponding symmetric encryption algorithm MS4;

[0067] The block size of the symmetric encryption algorithm MS4 and the p-stable LSH function h(·) are set to 20 bytes long, and the secret key is 16 bytes;

[0068] The long-term private key MK is combined with information such as timestamp and expiration date and stored in the secret storage area;

[0069] S2. User registration stage

[0070] Before a new user Ui accesses the server S, he first provides a face image to the client C and a set of face image related data that meets privacy protection requirements to the server S for registration. The specific steps are:

[0071] 2.1. User Ui sends his facial image and registration request to the client through a dedicated channel, namely the client's camera device (facial capture device).

[0072] 2.2. After the client receives the facial image mi from Ui, it first extracts the facial feature vector through the trained FaceNet model; in order to support efficient facial image retrieval, the client uses the p-stable SLH function algorithm to obtain the hash value h i ;

[0073] Then, the client calculates the hash value H(h i ||r), to prevent server-side forgery attacks;

[0074] Next, the client uses the symmetric encryption algorithm MS4 to calculate the ciphertext data for registration, namely E k (H(h i ||r));

[0075] Finally, the client sends the ciphertext of the facial features and the corresponding hash value <h i ,E k (H(h i ||r))>Send to the server.

[0076] 2.3、The server receives <h i ,E k (H(h i ||r))>, the server finds the corresponding classification set in the database DB according to the hash value and stores the ciphertext of the facial features in the set.

[0077] S3, Identity Authentication Stage

[0078] 3.1. The user initiates an authentication request to the client, and the client receives and responds to the request;

[0079] 3.2. The client collects multiple facial images of user U through a camera device

[0080] 3.3. First, load the pre-trained FaceNet model, adjust the image format to 224×224×3 png format RGB image, and then extract the multi-dimensional facial feature vector A={V 1 ,V 2 ,V 3 ,…V n}. Perform FaceNet facial feature extraction on A to obtain a single 128-dimensional biometric feature vector. Perform L2norm (Euclidean norm) processing on this vector. This L2 norm is to make different networks belong to the same order of magnitude.

[0081] 3.4. The obtained 128-dimensional biometric feature vector is passed through the p-stable LSH function to obtain the hash value h i , each hash function in the LSH function family based on p-stable distribution can be expressed as: α, b are two random variables in the hash function, where α is a 128-dimensional feature vector, b is a random number in the range [0, w], and w is a parameter set manually. v is the input feature vector.

[0082] 3.5. The client calculates the hash value H(h i ||r) and ciphertext data E k (H(h i ||r)); Finally, the client sends <h i ,E k (H(h i ||r))> and authentication request.

[0083] 3.6、The server receives from the client <h i ,E k (H(h i ||r))>, the server traverses the hash value of each face image data in the database DB and finds the hash value closest to h i The hash value of the face is quickly matched to the classification set.

[0084] Specifically, the server traverses each data item in the database DB <h x ,E k (H(h x ||r))>(1≤x≤|DB|); calculate h j ={h x |min <h x ; E k (H(h x ||r)∈DB|h x -h i |, get the closest hash value and corresponding ciphertext, that is, <h j ,E k (H(h j ||r))>;

[0085] Among them, |h x -h j | indicates h x With h j The absolute value between |DB| represents the number of items in the database DB. r means a random number, the purpose is to prevent leakage caused by man-in-the-middle attacks. j and h xAre hash values, where H j represents the hash value found by the server in the database that is most similar to the face image feature vector submitted by the current user. x Represents the value obtained by hashing the x-th user face image feature vector stored in the database through the LSH function.

[0086] Then the ciphertext is calculated, and the similarity of the ciphertext M is calculated by Euclidean distance. The Euclidean distance formula is as follows:

[0087]

[0088] Among them, A and B are both vectors. D represents the dimension, that is, the dimension of the vector. i ,b i Represents the element values ​​at the same position of two different vectors.

[0089] The ciphertext set with similarity <h j ,M||E k (H(h j ||r))>Sent to the client.

[0090] 3.7. The client receives the traversal results from the server <h j ,M||E k (H(h j ||r))>, the received result is decrypted to determine whether the returned similarity is within the specified threshold. If it is not within the specified threshold, the user Ui authentication fails; otherwise, the user Ui authentication passes.

[0091] The present invention has the following advantages:

[0092] Facial feature privacy: The present invention uses the secure encryption algorithm SM4 to encrypt the data to ensure that the data is protected during transmission and is not tampered with or leaked, thereby maintaining the integrity and privacy of the data. During the authentication phase, the server performs facial similarity calculations in a ciphertext state to ensure that the facial data and similarity have not been tampered with or leaked, thereby ensuring the integrity of the data during processing.

[0093] User anonymity: User anonymity means that after the user encrypts and uploads his or her facial feature information, the database server cannot steal the user's facial feature plaintext information. During the authentication stage, the homomorphic encryption algorithm used for face matching is performed in a ciphertext state, and the user's facial feature information will not be leaked.

[0094] Preventing replay attacks: In encrypted communications, timestamps can be used to prevent replay attacks, where an attacker intercepts and resends previous communications. By checking timestamps, the system can reject expired requests.

[0095] Enhanced security: By setting an expiration date, even if a key is obtained by an unauthorized person, there is only a limited window of time in which the key can be used. This increases the security of the system because it limits the damage that can be caused by unauthorized access.

[0096] Key rotation: To prevent keys from being used for a long time and increasing the risk of leakage, it is common practice to change keys regularly. Timestamps and expiration dates can help implement key rotation strategies.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A privacy-preserving lightweight face authentication method based on SM4 algorithm, characterized in that: The following steps are involved: In the initialization phase, the server selects the FaceNet model and the p-stable LSH function and deploys them on each client. Each client generates a long-term private key MK combined with a timestamp and expiration information and stores it in the secret storage area. During the user registration phase, the user uploads the face image to the client through the sensor device. After the client obtains the face image, it loads the pre-trained FaceNet model, outputs the face feature vector representing the original image, and then calculates the hash value through the p-stable LSH function; the client calculates the ciphertext for registration through the symmetric encryption algorithm MS4, and finally sends the ciphertext of the face feature and the corresponding hash value to the server; the server then finds the corresponding classification set based on the hash value and stores the ciphertext of the face feature in the set; During the identity authentication process, the client obtains the hash value of the user's face image and the ciphertext of the face features and sends them to the server; The server quickly matches the classification set where the face belongs based on the hash value, and performs ciphertext calculations on the ciphertext and the ciphertexts in the set one by one to obtain a ciphertext set of similarity, and returns the set to the client. After decryption, the client obtains the similarity between the face feature and the elements in the classification set, and determines whether there is an element in the similarity set that is less than the threshold. If so, the authentication is passed and the protocol terminates. Otherwise, the client sends the verification failure result to the server.

2. According to claim 1, the privacy-preserving lightweight face authentication method based on the SM4 algorithm is characterized in that: In step S1, the client generates a combination of a long-term private key MK and a timestamp and expiration date information: K=H(MK||T||PT) Among them, T is the current timestamp and PT is the validity period.

3. The privacy-preserving lightweight face recognition method based on the SM4 algorithm according to claim 2 is characterized in that: Choose a random number r, and then choose a hash function H: {0,1} * →{0,1} l To generate authentication information; let E k (m) represents the ciphertext of m encrypted using the key k of the symmetric encryption algorithm MS4, D k (m) represents the plaintext decrypted using the key k of the corresponding symmetric encryption algorithm MS4; the block size of the symmetric encryption algorithm MS4 and the p-stable LSH function h(·) are set to 20 bytes in length, and the secret key is 16 bytes.

4. The privacy-preserving lightweight face recognition method based on the SM4 algorithm according to claim 3 is characterized in that: The specific steps of step S2 are: Step 2.1, the user sends the facial image and registration request to the client through a dedicated channel, i.e. the camera device of the client; Step 2.2: After receiving the facial image, the client first extracts the facial feature vector through the trained FaceNet model; the client uses the p-stable SLH function algorithm to obtain the hash value h i ; Then, the client calculates the hash value H(h i ||r), to prevent server-side forgery attacks; Next, the client uses the symmetric encryption algorithm MS4 to calculate the ciphertext data for registration, namely E k (H(h i ||r)); Finally, the client sends the ciphertext of the facial features and the corresponding hash value <h i ,E k (H(h i ||r))>Send to the server; Step 2.3: The server receives <h i ,E k (H(h i ||r))>, the server finds the corresponding classification set in the database DB according to the hash value and stores the ciphertext of the facial features in the set.

5. The privacy-preserving lightweight face authentication method based on the SM4 algorithm according to claim 4 is characterized in that: The specific steps of step S3 are: Step 3.1: The user initiates an authentication request to the client, and the client receives and responds to the request; Step 3.2: The client collects multiple facial images of user U through a camera device Step 3.3: First load the pre-trained FaceNet model, adjust the image format to RGB image, and then extract the multi-dimensional facial feature vector A = {V1, V2, V3, ... V n }, perform FaceNet facial feature extraction on A to obtain the biological feature vector; Step 3.4: Use the obtained biometric vector through the p-stable LSH function to obtain the hash value h i ; Step 3.5: The client calculates the hash value H(h i ||r) and ciphertext data E k (H(h i ||r)); Finally, the client sends <h i ,E k (H(h i ||r))> and authentication request; Step 3.6: The server receives <h i ,E k (H(h i ||r))>, the server traverses the hash value of each face image data in the database DB and finds the hash value closest to h i The hash value of matches the classification set where the face belongs; Then the ciphertext is calculated, and the similarity of the ciphertext M is calculated by Euclidean distance; The ciphertext set with similarity <h j ,M||E k (H(h j ||r))>Send to the client; Step 3.7: The client receives the traversal results from the server <h j ,M||E k (H(h j ||r))>, the received result is decrypted to determine whether the returned similarity is within the specified threshold. If it is not within the specified threshold, the user authentication fails; otherwise, the user authentication passes.

6. The privacy-preserving lightweight face authentication method based on the SM4 algorithm according to claim 5 is characterized in that: In step 3.3, the image format is adjusted to 224×224×3 png format RGB image, and then the multi-dimensional facial feature vector A={V1,V2,V3,…V n }, perform FaceNet facial feature extraction on A to obtain a single 128-dimensional biometric feature vector.

7. The privacy-preserving lightweight face authentication method based on the SM4 algorithm according to claim 6 is characterized in that: Perform L2norm processing on a single 128-dimensional biometric feature vector.

8. The privacy-preserving lightweight face recognition method based on the SM4 algorithm according to claim 7 is characterized in that: In step 3.4, each hash function in the LSH function family based on p-stable distribution is expressed as:

9. The privacy-preserving lightweight face authentication method based on the SM4 algorithm according to claim 8, characterized in that: In step 3.6, the server traverses each data item in the database DB <h x ,E k (H(h x ||r))>(1≤x≤|DB|); calculate h j ={h x |min <h x ; E k (H(h x ||r)∈DB|h x -h i |, get the closest hash value and corresponding ciphertext, that is, <h j ,E k (H(h j ||r))>; where |h x -h j | indicates h x With h j The absolute value between |DB| represents the number of items in the database DB.

10. The privacy-preserving lightweight face authentication method based on the SM4 algorithm according to claim 9, characterized in that: In step 3.6, the Euclidean distance formula is as follows: Among them, A and B are both vectors.

Images