Face feature encryption matching method and system for privacy protection

Abstract

The application discloses a face feature encryption matching method and system for privacy protection, and the method comprises the following steps: preprocessing and feature extraction are performed on collected face images; a scaling factor is dynamically determined based on face feature distribution characteristics, and CKKS algorithm is improved; a feature set factor is constructed based on the concentration degree of feature values in a face feature vector, an encoding reference factor is constructed based on the representative degree of face feature values to the scaling factor in the CKKS algorithm by analyzing the feature set factor, and a dynamic scaling factor is constructed based on the feature set factor and the encoding reference factor; and face features are encrypted and matched based on the improved CKKS algorithm. Through dynamic analysis on the distribution characteristics of the face feature vector, the scaling factor of homomorphic encryption is adaptively determined, the optimization balance of precision and efficiency is realized, and the real-time performance of face feature encryption matching is significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of face recognition, in particular to a face feature encryption matching method and system for privacy protection. BACKGROUND

[0002] Face feature encryption matching is a technology for comparing face biometric features in the ciphertext domain. While bringing convenience with face recognition technology, the core feature data must be encrypted for protection to ensure that the data is available while being invisible, thereby effectively avoiding the risk of privacy leakage. In the prior art, homomorphic encryption schemes are usually used for ciphertext calculation, such as the CKKS homomorphic encryption algorithm based on lattice cryptography, which can directly support addition and multiplication operations on encrypted floating-point numbers, and is suitable for face feature vectors usually composed of high-dimensional floating-point numbers, and can directly calculate the Euclidean distance or cosine similarity between face feature vectors in the encrypted state. Chinese patent CN114093001A discloses a face recognition method for protecting privacy security, comprising: a face registration stage: collecting face images and extracting face features, and encrypting the face features by CKKS full homomorphic encryption algorithm to obtain a source face template; storing the source face template in a database; a face matching stage: collecting a face image to be identified and extracting face features to be identified, and encrypting the face features to be identified by CKKS full homomorphic encryption algorithm to obtain a target face template; matching the target face template with the source face template in the database. The invention uses CKKS full homomorphic encryption algorithm to encrypt face features, solving the problem of privacy security of face feature data. However, face feature vectors have unique characteristics of high dimension, uneven data distribution and dynamic change of numerical range. When the traditional CKKS algorithm is directly used for face feature encryption matching, the calculation accuracy and efficiency are difficult to balance, that is, in order to ensure the accuracy of the calculation result, the encryption parameters of the CKKS algorithm need to be globally preset according to the maximum possible range of the data; such static parameters cannot adapt to the actual statistical distribution of face feature data, often leading to the selection of parameters that are too large to accommodate extreme values, thereby causing ciphertext volume expansion and dramatic increase in homomorphic operation overhead, making it difficult to meet the real-time requirements of face recognition scenarios. SUMMARY

[0003] The present application aims to provide a face feature encryption matching method and system for privacy protection that can reduce homomorphic operation complexity while balancing matching accuracy.

[0004] Technical solution: The face feature encryption matching method for privacy protection comprises the following steps:

[0005] Preprocessing and feature extraction are performed on the collected face images;

[0006] The CKKS algorithm is improved based on the dynamic determination of the scaling factor based on the distribution characteristics of the face features, including constructing a feature set factor based on the concentration degree of the feature values in the face feature vector, constructing an encoding reference factor based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm based on the feature set factor, and constructing a dynamic scaling factor based on the feature set factor and the encoding reference factor.

[0007] The face features are encrypted and matched based on the improved CKKS algorithm.

[0008] Preferably, the preprocessing and feature extraction include converting the collected face image into a grayscale image, denoising the grayscale image, and then extracting the face part of the face image to input into the face recognition model to output a face feature vector.

[0009] Preferably, the extraction of the face part of the face image includes locating the position and boundary of the face in the face image using the MTCNN network and removing the background part.

[0010] Preferably, the construction of the feature set factor based on the concentration degree of the feature values in the face feature vector includes taking the absolute value of each element in the face feature vector and arranging it in descending order, selecting the top k head features, and constructing the feature set factor based on the ratio of the mean of the k head features to the mean of all features.

[0011] Preferably, the construction of the encoding reference factor based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm based on the feature set factor includes determining an adaptive threshold for the feature set factor, mapping the difference between the feature set factor and the adaptive threshold in the interval (0, 1), and adaptively assigning the weight of the maximum feature value and the mean of the head features to construct the encoding reference factor.

[0012]

[0013] wherein E represents the encoding reference factor of the face feature vector, represents the feature value with the largest absolute value in the face feature vector, and sig() represents the sigmoid function, F represents the feature set factor of the face feature vector, and F' represents the adaptive threshold of the feature set factor, represents the mean of the k head features.

[0014] Preferably, the adaptive threshold of the feature set factor is determined by the Otsu threshold segmentation method.

[0015] Preferably, the calculation method of the dynamic scaling factor is:

[0016] Wherein Q represents a dynamic scaling factor, E represents an encoding reference factor of a face feature vector, represents a rounding up function, F represents a feature set factor of a face feature vector.

[0017] The face feature encryption matching system for privacy protection provided by the application comprises an image acquisition device, a user terminal and a server terminal, the output end of the image acquisition device is connected with the user terminal, and the output end of the user terminal is connected with the server terminal.

[0018] The image acquisition device is used for acquiring a face image and outputting the acquired face image to the user terminal.

[0019] The user terminal is used for pre-processing the face image and extracting a feature vector, dynamically determining a scaling factor in the CKKS algorithm based on the face feature distribution characteristics, improving the CKKS algorithm, and then encrypting the face feature by using the improved CKKS algorithm to generate a user terminal ciphertext and transmitting the ciphertext and the corresponding scaling factor to the server terminal.

[0020] The dynamic determination of the scaling factor in the CKKS algorithm and the improvement of the CKKS algorithm comprise constructing a feature set factor based on the concentration degree of the feature values in the face feature vector, constructing an encoding reference factor based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm by analyzing the face feature values based on the feature set factor, and constructing a dynamic scaling factor based on the feature set factor and the encoding reference factor.

[0021] The server terminal is used for receiving the user terminal ciphertext and the corresponding scaling factor and matching the ciphertext domain in the database to output a face feature matching result.

[0022] The electronic device provided by the application comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the computer program realizes the face feature encryption matching method for privacy protection when being loaded to the processor.

[0023] The computer readable storage medium provided by the application stores a computer program, and the computer program realizes the face feature encryption matching method for privacy protection when being executed by the processor.

[0024] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: By dynamically analyzing the distribution characteristics of facial feature vectors, the scaling factor of homomorphic encryption is adaptively determined, achieving an optimized balance between accuracy and efficiency, and significantly improving the real-time performance of facial feature encryption matching; at the same time, the method has stronger robustness to various facial features, and through end-to-end full-process ciphertext processing, a comprehensive privacy protection system is constructed while ensuring high matching accuracy, effectively solving the problem of difficulty in balancing computational efficiency and security performance in the prior art. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0027] like Figure 1 As shown, the facial feature encryption matching method for privacy protection according to the present invention includes the following steps:

[0028] (1) Preprocess and extract feature vectors from the collected face images.

[0029] The system acquires raw facial images to be identified using image acquisition devices, such as high-definition cameras, infrared cameras, or 3D structured light cameras, which are installed on mobile terminals, access control machines, or self-service terminals. The acquisition process can be actively triggered by the user, such as by looking directly at the camera to take a picture; or passively triggered by the system through motion detection sensors, automatically capturing clear, evenly lit, and orthogonally oriented raw facial images within a designated area.

[0030] To ensure image quality, the acquired raw face images are preprocessed to obtain feature vectors after face image transformation. Specifically, this includes:

[0031] (1.1) Convert the acquired original face image into a grayscale image and perform noise reduction processing on the grayscale image; the noise reduction processing can be performed using a Gaussian filtering algorithm;

[0032] (1.2) Extract the face portion from the face image, specifically by using the MTCNN network to accurately locate the position and boundaries of the face in the image, and then removing the background portion;

[0033] (1.3) The extracted face portion is used as the input to the face recognition model, and the output of the face recognition model is the face feature vector of the face portion. The value of each dimension in the face feature vector corresponds to a feature value of the face. The face recognition model can be the ArcFace model, whose output face feature vector has 512 dimensions.

[0034] (2) Based on the distribution characteristics of facial features, the scaling factor in CKKS algorithm is dynamically determined to improve the CKKS algorithm. Specifically, it includes:

[0035] (2.1) A feature concentration factor is constructed based on the concentration degree of the feature values in the facial feature vector. The numerical distribution of the facial feature vector as a highly condensed biological information expression has the characteristics of sparse activation. That is, due to the hierarchical abstraction mechanism of the face recognition model, high-level semantic features are only encoded by specific neuron combinations, such as high nose bridge and wide eye distance, which will cause the feature extraction model to tend to produce activation values much larger than the mean on a few key dimensions, while the values of the remaining dimensions are relatively small or close to zero. Based on this, the feature concentration factor is constructed by dynamically analyzing the distribution characteristics of the facial feature vector to reflect the concentration of key features in the facial feature vector. It includes the following steps:

[0036] (2.1.1) Take the absolute value of each element in the facial feature vector to get the absolute value vector ; all elements in the absolute value vector are arranged in descending order to get the descending vector ; wherein, represents the i-th facial feature vector; represents the absolute value of the i-th facial feature vector; represents the i-th feature value in the descending vector S; n represents the number of all features;

[0037] (2.1.2) Select k head features to construct the feature concentration factor

[0038]

[0039] wherein, F represents the feature concentration factor of the facial feature vector, which is used to quantify the concentration degree of the numerical distribution in the facial feature vector; k represents the number of selected head features, represents the i-th feature value in the descending vector, represents the mean of the k head features, represents the mean of all features in the descending vector.

[0040] If the face feature vector has significant sparse activation characteristics, the values of the top k elements are much higher than those of other elements, resulting in a head feature mean significantly greater than the overall feature mean, and a larger factor in the feature set obtained by calculation; if the numerical distribution of each dimension in the face feature vector is relatively uniform, the difference between the head feature mean and the overall feature mean is smaller, the factor in the feature set obtained by calculation is smaller, and is close to 1. Therefore, the larger the value of the feature set factor, the more concentrated the feature distribution in the face feature vector. The head feature mean and the overall feature mean are used to construct the feature set factor, which accurately quantifies the concentration degree of the numerical distribution of each dimension in the face feature vector, that is, the concentration of the key features in the face feature vector, and provides a core basis for subsequent adaptive parameter adjustment.

[0041] (2.2) Construct an encoding reference factor based on the representative degree of the feature value in the face feature vector to the scaling factor in the CKKS algorithm.

[0042] In the traditional CKKS algorithm, the scaling factor is usually set according to the global maximum value in the face feature vector. This processing method is easy to cause the ciphertext volume to expand and the homomorphic operation overhead to increase dramatically in the subsequent steps, which reduces the efficiency of face feature encryption matching. Therefore, the present application considers the internal distribution characteristics of sparse activation in the face feature vector, analyzes the representative degree of the feature value in the face feature vector to the scaling factor in the CKKS algorithm, and constructs an encoding reference factor. Specifically, it includes:

[0043] (2.2.1) Obtain multiple face images from the database, and extract the face feature vectors of the multiple face images according to step (1);

[0044] (2.2.2) Calculate the feature set factor of each face feature vector according to step (2.1);

[0045] (2.2.3) Calculate the feature set factor of each face feature vector according to step (2.1); F As the input of the Otsu threshold segmentation method, the output is the adaptive threshold F' of the feature set factor;

[0046] (2.2.4) Based on the mapping of the difference between the feature set factor and the adaptive threshold in the interval (0, 1), the weights of the maximum feature value and the head feature mean are adaptively assigned, thereby constructing the encoding reference factor

[0047]

[0048] Wherein E The encoding reference factor of the face feature vector is used to dynamically evaluate the representative degree of the feature value, represents the first eigenvalue in the descending order vector, that is, the eigenvalue corresponding to the largest absolute value in the face feature vector, sig() represents the sigmoid function, and is used to map the data in the parentheses to the range of (0, 1), F represents the concentration factor in the feature set of the face feature vector, and F' represents the adaptive threshold of the concentration factor in the feature set, represents the average of the first k head features in the descending order vector.

[0049] When the concentration of the eigenvalue in the face feature vector is large, the concentration factor in the feature set is significantly greater than the adaptive threshold, indicating that the first k head features in the face feature vector can represent the scaling factor in the CKKS algorithm, and the corresponding is significantly greater than 0. However, after the sigmoid function processing, the term is close to 1, is close to 0, and the calculated encoding reference factor is closer to the eigenvalue with the largest absolute value in the face feature vector. Therefore, the eigenvalue with the largest absolute value is more accurate, and the eigenvalue with the largest absolute value should be used to adjust the scaling factor;

[0050] When the concentration of the eigenvalue in the face feature vector is small, the concentration factor in the feature set is significantly less than the adaptive threshold, indicating that the first k head features in the face feature vector cannot represent the scaling factor in the CKKS algorithm, and the corresponding is significantly less than 0. However, after the sigmoid function processing, the term is close to 0, is close to 1, and the calculated encoding reference factor is closer to the average of the first k head features . Therefore, the average of the head features is more accurate, and the more stable average of the head features should be used to adjust the scaling factor.

[0051] The adaptive threshold of the concentration factor in the feature set is determined based on the Otsu threshold segmentation method, and then the encoding reference factor is constructed, the dynamic switching between the maximum eigenvalue and the average of the head features is realized, and the most suitable reference value is used for parameter adjustment in the feature distribution, solving the limitation of the traditional algorithm which single relies on the global maximum value to set the parameter.

[0052] (2.3) Construct a dynamic scaling factor based on the encoding reference factor, and improve the CKKS algorithm.

[0053] In the CKKS algorithm, the precision and efficiency of encrypting the face feature vector has a strong correlation with the core parameter scaling factor. When the scaling factor is large, more decimal information of each feature value in the face feature vector can be retained during encoding, thereby improving the precision of the encrypted calculation, so that the face feature vector resolved in the ciphertext domain is highly consistent with the plaintext face feature vector, ensuring the accuracy of matching, but a larger scaling factor will result in a decrease in encoding operation efficiency. When the scaling factor is small, the operation speed is faster, thereby having high operation efficiency, but a smaller scaling factor will result in a larger error when encoding each feature value in the face feature vector, thereby affecting the calculation precision, and this error will be amplified after multiple operations, resulting in a large deviation between the final resolved face feature vector and the plaintext face feature vector, thereby causing matching failure or false recognition. Therefore, the present application considers that the concentration degree of the feature values in the face feature vector reflects the degree to which the feature values can represent the scaling factor, and a dynamic scaling factor is constructed,

[0054]

[0055] wherein Q represents a dynamic scaling factor, i.e. the scaling factor for homomorphic encryption of the face feature vector in the traditional CKKS algorithm is improved, denotes a logarithmic function with 2 as the base, E denotes an encoding reference factor of the face feature vector, denotes a rounding up function, F denotes a feature set factor of the face feature vector, denotes a logarithmic function with a natural constant as the base; by a number closest to the power of 2 of the encoding reference factor is obtained, and by the scaling factor is adjusted on the basis of the basic value.

[0056] In the formula, a number closest to the power of 2 of the encoding reference factor can be obtained, and the basic value of the dynamic scaling factor is the power of 2, in the CKKS algorithm, after performing homomorphic multiplication operation, the plaintext scale corresponding to the ciphertext will be squared, therefore, when the basic value of the dynamic scaling factor is the power of 2, the operation efficiency of the CKKS algorithm can be improved. For example, when E is 30, then , i.e. the number closest to the power of 2 of 30 is 32;

[0057] In the formula, the scaling factor can be finely adjusted on the basis of the basic value, since the value of the feature set factor is always greater than or equal to 1, therefore Constant is greater than or equal to 1, when the feature value in the face feature vector is highly concentrated in the head feature, the value of the feature concentration factor in the feature set is larger, at this time, a larger scaling factor should be ensured to ensure the accuracy of the matching; when the feature value in the face feature vector is relatively dispersed, the value of the feature concentration factor in the feature set is smaller, at this time, a smaller scaling factor should be ensured to ensure the efficiency of the matching;

[0058] Therefore, based on And The dynamic scaling factor obtained after multiplication can be dynamically determined according to different face feature vectors, and the matching accuracy and efficiency are considered.

[0059] The scaling factor of the face feature vector is dynamically determined by encoding the reference factor, the CKKS algorithm is improved, so that for the vast majority of face vectors with concentrated feature distribution and moderate dynamic range, a smaller scaling factor can be automatically used, the ciphertext volume is significantly reduced, the homomorphic operation complexity is reduced, and the matching efficiency is improved.

[0060] The improved CKKS algorithm is used for encrypted matching of face features.

[0061] When encrypting the face feature vector at the user end, a dynamic scaling factor is used, and in the CKKS algorithm encoding stage, any feature value v in the face feature vector is encoded as a coefficient in the plaintext polynomial, which is , Indicates the integer function, v represents the feature value, Q represents the dynamic scaling factor, the same processing is performed on each feature value in the face feature vector, and the face feature vector is mapped into a plaintext polynomial; the plaintext polynomial is encrypted into ciphertext using the encryption algorithm and the public key of CKKS. After encryption, the user end sends the ciphertext and the corresponding dynamic scaling factor to the server end for matching.

[0062] On the server side, for any one of the encrypted face feature vectors and its scaling factor in the database, denoted as database ciphertext and database scaling factor, the maximum value of the database scaling factor and the dynamic scaling factor from the user side is taken as the public scaling factor, the ciphertext with smaller scaling factor is processed by homomorphic multiplication using the public scaling factor, and the cosine similarity between the database ciphertext and the ciphertext from the user side is calculated in the ciphertext domain, denoted as ciphertext similarity; all the encrypted face feature vectors and scaling factors in the database are traversed according to the above steps, and the face corresponding to the database ciphertext with the maximum ciphertext similarity is taken as the matching result of the face feature of the user side; and the maximum ciphertext similarity is compared with the preset threshold value, and when the maximum ciphertext similarity is greater than or equal to the preset threshold value, it is judged that the to-be-identified person is the matched person in the database, and when the maximum ciphertext similarity is less than the preset threshold value, it is judged that the to-be-identified person is not the matched person in the database, so that an efficient and privacy-protected face feature encryption matching is completed. In the embodiment, the preset threshold value is 0.5, which can be selected according to actual conditions.

[0063] The present application constructs the characteristic factor and the encoding reference factor in the feature set by analyzing the sparse distribution characteristics of the face feature vector, dynamically determines the scaling factor of the face feature vector through the encoding reference factor, improves the CKKS algorithm, so that for most face vectors in the feature distribution set and in the dynamic range, the system can automatically use a smaller scaling factor, significantly reducing the ciphertext volume and the homomorphic operation complexity, and improving the matching efficiency; and the present method can allocate computing resources as needed according to the actual situation of each face feature, fundamentally overcoming the problem of low efficiency caused by the compatibility of extreme values in the traditional static parameter method, while ensuring the matching accuracy, further improving the operation efficiency of the encryption matching. Due to the dynamic scaling factor mechanism, unnecessary parameter amplification is avoided, the generated ciphertext volume is smaller, and the homomorphic operation cost is significantly reduced, so that the present application is especially suitable for face recognition authentication scenes such as access control and financial payment which require high real-time response, greatly enhancing the feasibility of the encryption technology in practical application. Through the core technology of parameter dynamic self-adaptation, the present application solves the contradiction between precision and efficiency in the field of face recognition, provides a privacy-protected face feature encryption matching solution with high security, high accuracy and high efficiency, and has high industrial utilization value.

[0064] Based on the same inventive concept, the present application provides a privacy-protected face feature encryption matching system, which comprises an image acquisition device, a user side and a server side, the output end of the image acquisition device is connected with the user side, and the output end of the user side is connected with the server side.

[0065] The image acquisition device includes a high-definition camera, an infrared camera, or a 3D structured light camera, etc., and is used to acquire a face image and output the acquired face image to a user terminal, so as to provide high-quality input data for subsequent feature extraction.

[0066] The user terminal is used to pre-process the original face image and extract a feature vector; a scaling factor in the CKKS algorithm is dynamically determined based on the face feature distribution characteristics, and the CKKS algorithm is improved; then the improved CKKS algorithm is used to encrypt the face features, to generate a ciphertext, and the ciphertext and the corresponding scaling factor are transmitted to the server terminal.

[0067] The user terminal includes a mobile terminal, an access control machine, or a self-service terminal, etc.; the pre-processing of the original face image and the extraction of the feature vector include converting the face image into a grayscale image, Gaussian filter denoising, using an MTCNN network to locate and extract a face region, and generating a face feature vector through an ArcFace model; the scaling factor in the CKKS algorithm is dynamically determined based on the face feature distribution characteristics, including calculating a feature set factor and an encoding reference factor of the face feature vector; the scaling factor of the CKKS homomorphic encryption algorithm is dynamically determined based on the encoding reference factor and the feature set factor, so as to realize adaptive encryption parameter adjustment.

[0068] The server terminal is used to receive the user terminal ciphertext and the scaling factor from the user terminal; the similarity between the user terminal ciphertext and the server terminal ciphertext is calculated based on the ciphertext domain and the scaling factor in the database, and compared with a preset threshold, to output a face feature matching result.

[0069] Through the cooperation of the image acquisition device, the user terminal, and the server terminal, the present application realizes end-to-end privacy protection from image acquisition to encrypted matching; the dynamic scaling factor mechanism effectively balances the calculation accuracy and efficiency, reduces the ciphertext volume and homomorphic operation overhead, improves the real-time performance of the face recognition scene, and ensures that the biometric data is always in an encrypted state during storage, transmission, and processing, thereby meeting the needs of high-security application scenarios such as financial payment and access control.

[0070] Based on the same inventive concept, the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program realizes the face feature encryption matching method for privacy protection when loaded into the processor.

[0071] Based on the same inventive concept, the present application provides a computer readable storage medium, which stores a computer program, and the computer program realizes the face feature encryption matching method for privacy protection when executed by a processor.

Claims

1. A face feature encryption matching method for privacy protection, characterized in that, The method comprises the following steps: Preprocessing and feature extraction are performed on the collected face image; The CKKS algorithm is improved by dynamically determining the scaling factor based on the distribution characteristics of the face features; including constructing a feature set factor based on the concentration degree of the feature values in the face feature vector, constructing an encoding reference factor based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm by analyzing the feature set factor, and constructing a dynamic scaling factor based on the feature set factor and the encoding reference factor; The feature set factor is constructed based on the concentration degree of the feature values in the face feature vector, which includes taking the absolute value of each element in the face feature vector and arranging it in descending order, selecting the first k head features, and constructing the feature set factor based on the ratio of the mean value of the k head features to the mean value of all features; The encoding reference factor is constructed based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm by analyzing the feature set factor, which includes determining the adaptive threshold of the feature set factor, mapping the difference between the feature set factor and the adaptive threshold in the interval (0, 1), and adaptively assigning the weight of the maximum feature value and the head feature mean value to construct the encoding reference factor wherein E denotes an encoding reference factor of the face feature vector, denotes the largest absolute value of the feature values in the face feature vector, and sig() denotes a sigmoid function, F denotes a feature set factor of the face feature vector, denotes an adaptive threshold of the feature set factor, denotes the k head feature mean values; The calculation method of the dynamic scaling factor is: where Q represents a dynamic scaling factor, E represents an encoding reference factor of the face feature vector, represents a ceiling function, represents a feature set factor of the face feature vector; The face features are encrypted and matched based on the improved CKKS algorithm.

2. The method for privacy-preserving face feature encryption matching of claim 1, wherein, The preprocessing and feature extraction include converting the collected face image into a grayscale image, denoising the grayscale image, and then extracting the face part of the face image to input into the face recognition model to output the face feature vector.

3. The method for privacy-preserving face feature encryption matching of claim 2, wherein, The face part of the face image is extracted by using the MTCNN network to locate the position and boundary of the face in the face image and remove the background part.

4. The method for privacy-preserving face feature encryption matching of claim 1, wherein, The adaptive threshold of the feature set factor is determined by using the Otsu threshold segmentation method.

5. A face feature encryption matching system for privacy protection, for implementing the face feature encryption matching method for privacy protection according to any one of claims 1-4, characterized in that, It comprises an image acquisition device, a user terminal and a server terminal, the output end of the image acquisition device is connected with the user terminal, and the output end of the user terminal is connected with the server terminal; The image acquisition device is used to collect face images and output the collected face images to the user terminal; The user terminal is used for preprocessing and feature vector extraction of face images; The CKKS algorithm is improved by dynamically determining the scaling factor in the CKKS algorithm based on the distribution characteristics of the face features; and then the face features are encrypted by using the improved CKKS algorithm to generate user terminal ciphertext, and the ciphertext and the corresponding scaling factor are transmitted to the server terminal; The dynamic determination of the scaling factor in the CKKS algorithm, which improves the CKKS algorithm, includes constructing a feature set factor based on the concentration degree of the feature values in the face feature vector, constructing an encoding reference factor based on the representative degree of the face feature values to the scaling factor in the CKKS algorithm by analyzing the feature set factor, and constructing a dynamic scaling factor based on the feature set factor and the encoding reference factor; The server terminal is used for receiving the user terminal ciphertext and the corresponding scaling factor and matching with the ciphertext domain in the database to output the face feature matching result.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is loaded into the processor to realize the face feature encryption matching method for privacy protection according to any one of claims 1-4.

7. A computer-readable storage medium storing a computer program, wherein the computer program comprises the following steps of: receiving a request for a resource from a client; determining whether the client is authorized to access the resource; and if the client is authorized to access the resource, providing the resource to the client. The computer program is executed by a processor to implement the face feature encryption matching method for privacy protection according to any one of claims 1-4.

Images