Data encryption sharing method and efficient data anonymous tracing retrieval method based on inner product function encryption

Abstract

The invention discloses a data encryption sharing method and an efficient data hiding tracing retrieval method based on inner product function encryption, and the method comprises the steps: a data owner extracts the features of personal data, and generates a corresponding personal data feature vector; the data owner sends the personal data feature vector to a key generation center; the key generation center executes a key generation algorithm and generates a corresponding private key for the personal data feature vector of the data owner; the retrieval executor collects personal data of the data retrieval party; the retrieval execution party acquires the public parameters from the key generation center, and encrypts the personal data of the data retrieval party based on the public parameters to form a vector ciphertext; and a retrieval executor executes a decryption algorithm on the vector ciphertext based on the private key and the public parameter, and retrieves from the private key of the data owner.

Description

Technical Field

[0001] The present invention belongs to the field of data security and privacy computing technology, and specifically provides a data encryption sharing method and an efficient data anonymous retrieval method based on inner product function encryption. Background Art

[0002] As a privacy-preserving data retrieval technology in multi-user scenarios, cloud-based anonymous data retrieval aims to ensure privacy during the retrieval process. Specifically, this technology requires that the retrieval executor (such as a cloud service provider) and other participants cannot access any sensitive information related to the retrieval content or the retrieval process during the retrieval process. Data encryption, as a key technology for ensuring data confidentiality, has been widely used in data storage and sharing. In this context, researchers have proposed numerous cryptography-based anonymous cloud data retrieval schemes.

[0003] Existing schemes can be divided into three categories based on the cryptographic techniques employed: those based on data feature summaries, those based on secure multi-party computation, and those based on homomorphic encryption. Schemes based on data feature summaries achieve efficient retrieval by extracting key data features. However, when processing similar data, this can lead to a decrease in retrieval accuracy due to duplicate feature summaries. Schemes based on secure multi-party computation ensure the security and accuracy of the retrieval process, but require multiple rounds of interactive communication between all participants. In multi-user scenarios, this significantly increases the communication overhead and computational complexity of the system, severely impacting retrieval efficiency. In comparison, schemes based on homomorphic encryption have become a hot topic of research due to their advantages, such as support for lossless ciphertext computation, ability to execute complex retrieval logic, and the lack of frequent interaction in multi-user environments. However, these schemes still face the inherent ciphertext bloat and computational efficiency bottlenecks of homomorphic encryption algorithms, placing significant communication, storage, and computational pressures on all system participants. The existing fully homomorphic encryption algorithm (CHEON JH, KIM A, KIM M, et al. Homomorphic encryption for arithmetic of approximate numbers [C] / / International Conference on the Theory and Applications of Cryptology and Information Security. Hong Kong, China, 2017: 409-437) has a large computational overhead due to the inherent complexity of the underlying mathematical operations. Secondly, since the security of the fully homomorphic encryption algorithm depends on difficult mathematical problems in high-dimensional lattice spaces (such as the LWE problem), and the noise mechanism introduced to support ciphertext operations will cause the size of the generated homomorphic ciphertext to increase significantly.

[0004] Therefore, how to break through the efficiency bottleneck of existing technical solutions while ensuring the privacy and security of retrieval data and processes, and build a new data retrieval solution with both high security and high retrieval performance has become a key scientific problem that needs to be urgently solved in the field of anonymous cloud data retrieval. Summary of the Invention

[0005] Purpose of the invention: To address the problems of high computational overhead and large ciphertext size faced by existing fully homomorphic encryption algorithms, the present invention proposes a data encryption sharing method and an efficient data anonymous retrieval method based on inner product function encryption. It is aimed at the field of privacy-preserving data retrieval in multi-user scenarios and can meet the functional requirements and privacy protection requirements in complex data retrieval processes.

[0006] Technical solution: A data encryption sharing method, comprising the following steps:

[0007] The data owner extracts the features of personal data and generates the corresponding personal data feature vector;

[0008] The data owner sends the personal data feature vector to the key generation center;

[0009] The key generation center executes the key generation algorithm to generate the corresponding private key for the data owner's personal data feature vector;

[0010] The key generation center returns the generated private key to the data owner;

[0011] The data owner uploads the private key to the cloud service provider.

[0012] Furthermore, the key generation algorithm includes the following steps:

[0013] The key generation center obtains the security parameter λ and the maximum length of the vector

[0014] Use the security parameter λ to generate the group parameters corresponding to the asymmetric group through the group generation algorithm: in is the multiplicative cyclic group, q is the order of the group;

[0015] The key generation center randomly selects the generator of the group

[0016] For any The key generation center selects a random number and calculate

[0017] The key generation center generates public parameters and master key

[0018] The key generation center broadcasts the public parameters pp and stores the master key msk in a secure area;

[0019] Assuming the data owner's personal data feature vector x, the key generation center generates the vector according to the following formula The corresponding private key:

[0020]

[0021] The present invention also discloses an efficient data anonymization and retrieval method based on inner product function encryption, comprising the following steps:

[0022] Step 1: The key generation center uses a data encryption sharing method to generate public parameters, master keys, and private keys for the personal data of different data owners; and sends the generated private keys to the corresponding data owners;

[0023] Step 2: The retrieval execution party collects the personal data of the data retrieval party;

[0024] Step 3: The key generation center obtains public parameters from the retrieval execution direction and encrypts the personal data of the data retrieval party based on the public parameters to form a vector ciphertext;

[0025] Step 4: The retrieval execution party performs a decryption algorithm on the vector ciphertext based on the private key and public parameters, and retrieves it from the data owner's private key;

[0026] Among them, the data encryption sharing method is the data encryption sharing method disclosed above.

[0027] Furthermore, in step 3, the retrieval execution direction obtains public parameters from the key generation center, and encrypts the personal data of the data retrieval party based on the public parameters. The specific operations include:

[0028] Retrieve the execution direction key generation center to obtain the public parameter pp;

[0029] Pick a random number And calculate C = g r ,D=h r ,

[0030] According to the following formula, the personal data of the data retriever is encrypted to form the corresponding vector ciphertext

[0031] Furthermore, in step 4, the retrieval execution party executes a decryption algorithm on the vector ciphertext based on the private key and public parameters, and retrieves the data owner's private key. The specific operations include:

[0032] Based on the private key and vector ciphertext, the retrieval execution party first calculates the intermediate parameter E' according to the following formula, and finally calculates log g(E') to obtain the inner product result of vector x and vector y<x,y> , to achieve retrieval from the data owner’s private key;

[0033]

[0034] Among them E i The ciphertext parameter obtained by the encryption vector, x i are the elements in vector x, C and D are random parameters in the ciphertext, s x ,t x are the random parameters in the key respectively.

[0035] Furthermore, if the inner product result<x,y> If the decryption algorithm is in the integer interval {0,…,L}, the Pollard kangaroo method is used to optimize the decryption algorithm to O(L 1 / 2 ) to complete the decryption algorithm, and l represents the upper limit of the calculation result range.

[0036] Furthermore, by pre-calculating a scale of O(l 1 / 3 ) table to optimize the decryption algorithm and reduce the time complexity of the decryption algorithm to O(l 1 / 3 ).

[0037] Furthermore, the execution of the decryption algorithm is optimized according to the following optimization process:

[0038] According to the Euclidean distance calculation formula, the Euclidean distance between two vectors x and y is calculated as follows:

[0039]

[0040] Since x and y belong to different entities and cannot be cross-calculated, the vectors are transformed as follows:

[0041]

[0042] Therefore, the characteristics based on the inner product operation are:

[0043]

[0044] square term and The high and low bits of are split so that the bit lengths corresponding to the split items are close to those of other elements in the vector. The specific optimization method is as follows:

[0045] First, let the square term where Ψ (h) and Ψ (l) Represents the square terms The high and low positions of Φ (h) and Φ (l) Represents the square terms high and low;

[0046] The original vector is then represented as follows:

[0047]

[0048] Calculation based on the Arbitrary Inner Product Function Encryption (IPFE) scheme<x″,y″> , by calculating<x″,y″> , the equation dist(x,y)=<x″,y″> Still holds true.

[0049] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0050] (1) To address the problems of high computational overhead and large ciphertext size in existing fully homomorphic encryption algorithms, this paper proposes a data retrieval method based on inner product function encryption, thereby effectively reducing the computational overhead and corresponding ciphertext size in the data retrieval process;

[0051] (2) The present invention further optimizes the performance of the data retrieval algorithm (i.e., the decryption algorithm) by using the idea of ​​pre-computation, thereby significantly improving the efficiency of data retrieval;

[0052] (3) The method of the present invention can achieve efficient and privacy-preserving data retrieval in multi-user scenarios, that is, it can achieve data sharing, storage and retrieval without leaking any private information, assuming that the third-party cloud server is honest and curious. On the one hand, the data owner uses the key corresponding to the data instead of the data itself, and uploads it to the cloud server for outsourced storage and retrieval, thereby saving data communication and storage overhead while protecting the privacy of the data. On the other hand, when performing data retrieval, the third-party cloud server can perform data retrieval without decrypting the data, and the data retrieval method based on data similarity measurement can ensure efficient and lossless retrieval on ciphertext data. Therefore, the method of the present invention can support efficient and lossless data retrieval while ensuring privacy in the data retrieval process in multi-user scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the data encryption sharing stage;

[0054] Figure 2 Schematic diagram of the data retrieval stage;

[0055] Figure 3 Schematic diagram of cloud-edge collaborative image anonymous retrieval scenario;

[0056] Figure 4A comparison chart of the time costs of each algorithm under different security parameters and optimization parameters;

[0057] Figure 5 This is a comparison chart of the time costs of various operations under different security parameters in the method of the present invention;

[0058] Figure 6 Schematic diagram of cryptographic parameter scale evaluation for each operation in different vector dimensions in the method of the present invention;

[0059] Figure 7 This is a comparison chart of the encryption and decryption time overheads of the method of the present invention and FHE at different vector dimensions;

[0060] Figure 8 The following is a comparison chart of the time cost of each algorithm in the host and Raspberry Pi environments. DETAILED DESCRIPTION

[0061] To make the objectives, technical solutions and advantages of the present invention more clear, the following will further illustrate an efficient data anonymization and retrieval method based on inner product function encryption proposed by the present invention in conjunction with the accompanying drawings.

[0062] This embodiment proposes an efficient data anonymization and retrieval method based on inner product function encryption, which mainly includes four steps, as follows:

[0063] Step 1: System Initialization This step is performed by the Key Generation Center (Key

[0064] Generation Center (KGC), the input includes the security parameter λ and the maximum length of the vector The output includes the public parameter pp and the master key msk. Specifically, the key generation center KGC first uses the security parameter λ to generate the group parameters corresponding to the asymmetric group through the group generation algorithm: in is a multiplication cyclic group, and q is the order of the group. Then, the key generation center KGC randomly selects the generator of the group Then, for any The key generation center KGC selects a random number and calculate Finally, the key generation center KGC generates the public parameters and master key The key generation center KGC broadcasts the public parameters pp to the public and stores the master key msk in a secure area.

[0065] Step 2: Key generation KeyGen(msk,x)→sk x:The key generation algorithm is executed by the key generation center KGC. The algorithm inputs the master key msk and the user's facial feature data vector x. The algorithm calculates and outputs the corresponding private key sk x Specifically, to generate the vector The corresponding key is calculated by the key generation center KGC according to the following formula:

[0066]

[0067] Step 3: Data encryption Encrypt(pp,y)→C y :The encryption algorithm is executed by the data retrieval party (or retrieval execution party). The algorithm inputs the public parameter pp and the data vector y to be encrypted, and the algorithm outputs the corresponding vector ciphertext C y Specifically, the user first selects a random number And calculate C = g r ,D=h r , The corresponding ciphertext can be expressed as After the ciphertext is generated, it is sent to the retrieval executor for data retrieval.

[0068] Step 4: Data decryption Decrypt(pp,sk x ,C y )→<x,y> :The decryption algorithm is executed by the retrieval execution party. The algorithm inputs the public parameter pp and the private key sk corresponding to the target vector x And the ciphertext vector C to be retrieved y , the algorithm outputs the inner product of vectors x and y<x,y> Specifically, according to the private key sk x =(s x ,t x ) and vector ciphertext C y For example, the search execution party first calculates E' according to the following formula, and finally calculates log g(E') to obtain the inner product of vector x and vector u<x,y> .

[0069]

[0070] The above decryption algorithm involves the calculation of a discrete logarithm, which can be optimized from the following two perspectives:

[0071] 1) If the inner product result<x,y> In the smaller integer interval {0,…,L}, the Pollard kangaroo method can be used for optimization, with O(L 1 / 2 ) to complete the calculation; Pollard's kangaroo method is a mature method for calculating discrete logarithms, with a complete tool set.

[0072] 2) Based on the above method, the time complexity of the decryption algorithm can be further optimized by the idea of ​​pre-computation. Specifically, by pre-computing a scale of O(L 1 / 3 ) table, the time complexity of the decryption process can be reduced to O(L 1 / 3 ). Pre-calculation is common knowledge in the computer field to optimize computational complexity.

[0073] The specific solution optimization process is as follows:

[0074] According to the Euclidean distance calculation formula, the Euclidean distance between two vectors x and y is calculated as follows:

[0075]

[0076] Since x and y belong to different entities and cannot be cross-calculated, the vectors can be transformed as follows:

[0077]

[0078] Therefore, the characteristics based on the inner product operation are:

[0079]

[0080] Note that there are square terms in the inner product operation (respectively and ), the presence of the squared term significantly increases the computational time complexity of the proposed method. This is because encoding requires a sufficiently large number of bits to accommodate the range of possible computational results for each element. This range is determined by the length of the maximum bit length term, and the presence of the squared term causes this length to increase exponentially. To achieve optimal time complexity, the high and low bits of the squared term can be split so that the bit lengths of the split terms are close to those of the other elements in the vector. The specific optimization method is as follows:

[0081] First, let the square term where Ψ (h) and Ψ (l) Represents the square terms The high and low positions of Φ (h) and Φ (1) Represents the square terms The high and low bits of the original vector can then be further expressed as follows based on the vector transformation formula:

[0082]

[0083]

[0084] By calculation<x″,y″> , the equation dist(x,y)=<x″,y″> Still holds true, and based on the IPFE solution calculation<x″,y″> More efficient than direct calculation<x,y> Higher computing efficiency.

[0085] The workflow of the method proposed in this embodiment is divided into two stages, namely, data encryption sharing and ciphertext data retrieval.

[0086] Figure 1 The following shows the execution order of each step in the data encryption sharing phase:

[0087] Step 101: The data owner extracts features of personal data and generates corresponding feature vectors.

[0088] Step 102: The data owner sends the personal data feature vector to the key generation center.

[0089] Step 103: The key generation center executes a key generation algorithm to generate a corresponding key for the data owner's data vector. The key is also regarded as a label of the data.

[0090] Step 104: The key generation center returns the generated data tag (key) to the data owner.

[0091] Step 105: The data owner uploads the data tag (key) to the cloud service provider for subsequent data retrieval.

[0092] Figure 2 The following shows the execution order of each step in the ciphertext data retrieval phase:

[0093] Step 201: The retrieval execution party collects information of the data retrieval party (such as facial information).

[0094] Step 202: The retrieval execution direction key generation center obtains the public parameters used to encrypt data.

[0095] Step 203: The retrieval execution party extracts features of the collected user information and generates corresponding feature vectors.

[0096] Step 204: The retrieval execution party executes the encryption algorithm and uses the public parameters to encrypt the extracted data feature vector.

[0097] Step 205: The search component of the search execution party executes the decryption algorithm and uses the encrypted feature vector to search the tags uploaded by the data owner.

[0098] like Figure 3As shown in the figure, in a cloud-edge collaborative anonymous image data retrieval scenario, users can register by sending the key corresponding to the feature vector of their facial data to the cloud server. The cloud server then sends the key corresponding to the user's facial data to the associated edge node based on the corresponding relationship between the user and the edge node. When a user needs to access a node, the node's facial information collection device first collects the user's facial features, encrypts them, and finally sends the encrypted ciphertext to the edge node's data retrieval component for retrieval. Based on the retrieval results, it determines whether the user has access rights to the current node.

[0099] For this scenario, the proposed method (IPFE-IR) was systematically tested. The experimental environment was set up as follows: This embodiment method is based on the PBC function library and C++-14 implementation. The server configuration used during the test is as follows: CPU parameters are: "2.40GHz Intel Xeon Silver 4314CPU", and the running memory is 32GB; the Raspberry Pi configuration is as follows: CPU parameters are: "2.4GHz Broadcom BCM2712 Cortex-A76 (Arm v8)", and the running memory is 8GB. Both devices run under the Ubuntu 22.04LTS system. To prevent the accuracy of the experimental results from being affected by runtime errors, each algorithm in this embodiment method was executed 10,000 times and the average time overhead was calculated. Among them, since the Setup algorithm is only executed once during system initialization, its time overhead has a negligible impact on the overall system performance. Therefore, the time overhead of this algorithm is not included in the following experimental tests.

[0100] (1) Algorithm optimization evaluation

[0101] Figure 4 The time cost of each algorithm in the method of this embodiment is given. The optimization process of the calculation characteristics of the method of this embodiment is as follows:

[0102]

[0103] The optimization process focuses on reducing the size of the largest element in the bit length of the vector. Therefore, in the implementation process, it is necessary to balance the relationship between the bit length boundary of the elements that the vector can accommodate and the scale of the vector scaling. During the experiment, it was found that the highest bit length corresponding to the extracted image feature vector is 10 bits, so the bit length boundary can be set to 9 for scaling, that is, k=9 in the other two vector transformation formulas (corresponding to Bound in the figure). According to the experimental results in the figure, in the test of various vector dimension sizes under the same security parameters, as the Bound value increases, the time overhead of the key generation and encryption algorithm is not much different, but the time overhead of the decryption algorithm increases significantly. This shows that the optimization of this embodiment is effective. After selecting a smaller Bound value, the time overhead of the decryption operation can be effectively reduced. In addition, this embodiment also evaluates the optimization of the encryption algorithm under different security parameters ( Figure 4 (a) in Figure 4 (b) Figure 4 The corresponding security parameter of (c) is 1024. Figure 4 (d) Figure 4 e) Figure 4 (f in the corresponding security parameter is 2048, Figure 4 (g) Figure 4 (h) Figure 4 (i) in the figure shows the performance of each algorithm when the security parameter is 4096. The results in the figure show that as the security parameter increases, the time overhead of each algorithm increases, but the performance differences between different Bounds remain unchanged.

[0104] (2) Data retrieval performance evaluation

[0105] In order to accurately evaluate the performance of the method of this embodiment in actual scenarios, the performance overhead of different operations in the method of this embodiment is tested. To test the optimal performance of the method of this embodiment, the setting of Bound=9 is selected. In the data vector length range of [2 4 ,2 10 ], the time overhead of each operation in the method of this embodiment was tested. Label generation in the method of this embodiment (i.e., KeyGen algorithm); image acquisition and processing (including Encrypt algorithm and image feature extraction); retrieval operation (including Decrypt algorithm and inner product result comparison operation). The selection of elliptic curve related parameters in this experiment refers to the "secp224r1" curve recommended by the NIST standard. Table 1 shows the specific performance test results of each operation in the method of this embodiment when the security parameter is λ=2048. According to the test results in the table, it can be seen that as the size of the data vector dimension increases, the time overhead of each operation increases linearly with the vector dimension. This experiment further tested the time overhead of each operation in the method of this embodiment under different security parameters, and the test results are as follows Figure 5 As shown in the figure, the experimental results show that as the security parameter increases (the security of the scheme increases), the execution time of each operation increases linearly, which is in line with the general rule.

[0106] Table 1. Time cost test of each operation in the proposed method when the security parameter is 2048 (unit: milliseconds)

[0107]

[0108] (3) Performance comparison with similar methods

[0109] Before making a comparison, we first analyze similar anonymous retrieval methods. Privacy-preserving data retrieval schemes based on homomorphic encryption are limited by the ciphertext expansion and high computational complexity of homomorphic encryption itself, and usually have large computational, communication, and storage overheads. Secure multi-party computation involves frequent interactions between multiple parties, resulting in the protocol itself having high communication complexity, which makes the related anonymous retrieval methods face the problems of large storage overhead and high communication delay. Although the scheme based on data feature summary does not introduce excessive computational, communication, or storage overhead, the hash collision problem it has will have a potential impact on the accuracy of data retrieval.

[0110] In contrast, the data retrieval technology based on inner product function encryption in this embodiment uses the key corresponding to the data feature data vector instead of the vector itself for sharing and retrieval. This not only avoids the heavy storage and communication overhead of each participant in the system, but also effectively ensures the accuracy of data retrieval, thereby achieving a good trade-off between performance and privacy protection. Due to the obvious performance defects of methods based on data feature summaries and secure multi-party computing, they are not suitable for data retrieval in cloud-edge collaboration scenarios. Therefore, this experiment only compares the performance with methods based on homomorphic encryption.

[0111] Specifically, the method of this embodiment is compared with a fully homomorphic encryption (FHE)-based anonymous retrieval technique (CHEON JH, KIM A, KIM M, et al. Homomorphic encryption for arithmetic of approximate numbers [C] / / International Conference on the Theory and Applications of Cryptography and Information Security. Hong Kong, China, 2017: 409-437) from the perspectives of cryptographic parameter scale and algorithm performance. First, regarding cryptographic parameter scale, this experiment compares the sizes of the relevant cryptographic parameters (including public parameters, keys, ciphertext, or secret parameters) involved in the data retrieval process of the two techniques. This experiment is tested on data vectors of different dimensional sizes to evaluate the communication and storage overhead of the two techniques during data anonymous retrieval for data vectors of different sizes. Because the security of fully homomorphic encryption algorithms relies on mathematical problems in high-dimensional linear spaces (such as the LWE problem), their ciphertexts must be large enough to accommodate the increasing noise during the computation and ensure security. Considering the security of the scheme, this experiment evaluated fully homomorphic encryption technology using a minimum vector dimension of 4096. All data vectors with actual dimensions less than 4096 were processed as 4096. This experiment compared two anonymous retrieval techniques across three phases: initialization, registration, and retrieval. The statistical security parameter length of the fully homomorphic encryption technique was set to 128 bits, while the security parameter length of the proposed method was set to 2048 bits, with a Bound of 9. The experiment tested data with vector dimensions of 4096, 8192, and 16384. Table 2 compares the cryptographic parameter scales of the two techniques at different vector dimensions. During the initialization phase, both techniques require the transmission of public parameters for system initialization. During the user registration phase, fully homomorphic encryption requires the transmission of computational parameters (including reline and gal in the table) as well as the ciphertext of the data feature vector, while the proposed method only requires the transmission of the key corresponding to the data feature vector. During the retrieval phase, both techniques require the transmission of ciphertext for retrieval. According to the experimental results, the public parameters and ciphertext scale of the proposed method are larger than those of fully homomorphic encryption technology, while the parameter scale of the registration phase is much smaller than that of fully homomorphic encryption. During a data retrieval process, the public parameters and the ciphertext used for retrieval only need to be transmitted / stored once, while the parameter scale of the registration phase is related to the number of registered users. Therefore, in large-scale data retrieval scenarios, compared with fully homomorphic encryption technology, the method of this embodiment will significantly reduce the communication and storage overhead of the central server and edge nodes.

[0112] In addition, in order to further illustrate the advantages of the method of this embodiment over fully homomorphic encryption technology in processing small-dimensional image data, this experiment tested the cryptographic parameter scale of the proposed method when processing small-dimensional image data. The experimental results are as follows: Figure 6 As shown. This experiment tests the IPFE-IR technology from the three stages of initialization, registration and retrieval, setting Bound=9, and selecting security parameters as 1024, 1536, 2048, 2560, 3072 and 4096 respectively. According to the experimental results, for the public parameters in the initialization stage, the size of the public parameters increases linearly with the increase of security parameters and vector dimensions. In the user registration stage, since the password corresponding to the user image feature vector is only related to the security parameters, under the same security parameter setting, the size of the key is independent of the vector dimension and remains unchanged. For the image retrieval stage, the ciphertext size also increases with the increase of security parameters and vector dimensions. Compared with fully homomorphic encryption technology, the method of this embodiment provides a flexible data processing method, which has a smaller cryptographic parameter scale when processing data vectors of smaller dimensions, thereby bringing less communication and storage overhead.

[0113] Table 2 Comparison of cryptographic parameter scales between the method of this embodiment and FHE at different vector dimensions (unit: KB)

[0114]

[0115]

[0116] In the comparative analysis of algorithm performance, since the KeyGen algorithm is only executed once, this experiment does not involve the performance comparison of this algorithm. The main focus is on the time overhead comparison of the encryption and decryption algorithms in the two technologies. This experiment is conducted under the setting of prime field size of 224 and security parameter λ = 2048. The selected test vector size range is {2 4 -2 10}. The test results are as follows Figure 7 As shown. According to the experimental results, under different test vector dimension scales, although the encryption time of the method of this embodiment is longer, its decryption time is much shorter than the decryption time of the fully homomorphic encryption method. In an image retrieval process, the encryption operation of the retrieved image only needs to be performed once, and the number of retrievals on the data (corresponding to the decryption of the two) is proportional to the scale of the image database, so that the overall overhead of the retrieval process is proportional to the decryption overhead of one time in the relevant scheme. Therefore, the retrieval efficiency of the method of this embodiment in the large-scale image database retrieval scenario is much higher than that of the fully homomorphic encryption scheme with the same security, and has better and obvious performance advantages.

[0117] (4) Performance evaluation on terminal devices

[0118] In order to further demonstrate the wide applicability of the method of this embodiment, the performance of each algorithm of the method of this embodiment is tested on a Raspberry Pi device. In order to comprehensively evaluate the performance of the method of this embodiment, this experiment tests the time overhead of each algorithm of the method of this embodiment on a Raspberry Pi device when the Bound value is 9, 10, and 11. This experiment is carried out under different security parameters (1024, 2048, and 4096) and test data vector size (range 2 4 -2 10 ) was tested. The test results are as follows Figure 8 As shown in the test results, it can be seen that even on a Raspberry Pi device with poor performance, the algorithms in this embodiment method do not have too much performance loss. 10 The single query time overhead for a dimensional data vector does not exceed 0.2 seconds. Furthermore, by comparing the test results with different Bound values, we can see that the method of this embodiment is also effective on a Raspberry Pi device, indicating that the method of this embodiment has good performance in device environments with different performance.

Claims

1. A data encryption sharing method, characterized by: The following steps are involved: The data owner extracts the features of personal data and generates the corresponding personal data feature vector; The data owner sends the personal data feature vector to the key generation center; The key generation center executes the key generation algorithm to generate the corresponding private key for the data owner's personal data feature vector; The key generation center returns the generated private key to the data owner; The data owner uploads the private key to the cloud service provider.

2. The data encryption sharing method according to claim 1, characterized in that: The key generation algorithm comprises the following steps: The key generation center obtains the security parameter λ and the maximum length l of the vector; Use the security parameter λ to generate the group parameters corresponding to the asymmetric group through the group generation algorithm: in is the multiplicative cyclic group, q is the order of the group; The key generation center randomly selects the generator of the group For any i∈{1,…,l}, the key generation center selects a random number and calculate The key generation center generates public parameters and master key The key generation center broadcasts the public parameters pp and stores the master key msk in a secure area; Assuming the data owner's personal data feature vector x, the key generation center generates the vector according to the following formula The corresponding private key:

3. An efficient data anonymization and retrieval method based on inner product function encryption, characterized by: The following steps are involved: Step 1: The key generation center uses a data encryption sharing method to generate public parameters, master keys, and private keys for the personal data of different data owners; and sends the generated private keys to the corresponding data owners; Step 2: The retrieval execution party collects the personal data of the data retrieval party; Step 3: The key generation center obtains public parameters from the retrieval execution direction and encrypts the personal data of the data retrieval party based on the public parameters to form a vector ciphertext; Step 4: The retrieval execution party performs a decryption algorithm on the vector ciphertext based on the private key and public parameters, and retrieves it from the data owner's private key; Wherein, the data encryption sharing method is a data encryption sharing method described in claim 1 or 2.

4. The efficient data anonymization and retrieval method based on inner product function encryption according to claim 3 is characterized by: In step 3, the search execution direction obtains public parameters from the key generation center and encrypts the personal data of the data search party based on the public parameters. The specific operations include: Retrieve the execution direction key generation center to obtain the public parameter pp; Pick a random number And calculate C = g r ,D=h r , According to the following formula, the personal data of the data retrieval party is encrypted to form the corresponding vector ciphertext C y =(C,D,E1,…,E l ).

5. The efficient data anonymization and retrieval method based on inner product function encryption according to claim 4 is characterized in that: In step 4, the retrieval execution party performs a decryption algorithm on the vector ciphertext based on the private key and public parameters, and retrieves the data owner's private key. The specific operations include: Based on the private key and vector ciphertext, the retrieval execution party first calculates the intermediate parameter E' according to the following formula, and finally calculates log g(E') to obtain the inner product result of vector x and vector y<x,y> , to achieve retrieval from the data owner’s private key; Among them, E i The ciphertext parameter obtained by encrypting the vector, x i are the elements in vector x, C and D are random parameters in the ciphertext, s x ,t x are the random parameters in the key respectively.

6. The efficient data anonymization and retrieval method based on inner product function encryption according to claim 5, characterized in that: If the inner product result<x,y> If the decryption algorithm is in the integer interval {0,…,L}, the Pollard kangaroo method is used to optimize the decryption algorithm to O(L 1 / 2 ) to complete the decryption algorithm, and L represents the upper limit of the calculation result range.

7. The efficient data anonymization and retrieval method based on inner product function encryption according to claim 5, characterized in that: By precomputing a size O(L 1 / 3 ) table to optimize the decryption algorithm and reduce the time complexity of the decryption algorithm to O(L 1 / 3 ).

8. The efficient data anonymization and retrieval method based on inner product function encryption according to claim 5 is characterized in that: The execution of the decryption algorithm is optimized according to the following optimization process: According to the Euclidean distance calculation formula, the Euclidean distance between two vectors x and y is calculated as follows: Since x and y belong to different entities and cannot be cross-calculated, the vectors are transformed as follows: Therefore, the characteristics based on the inner product operation are: square term and The high and low bits of are split so that the bit lengths corresponding to the split items are close to those of other elements in the vector. The specific optimization method is as follows: First, let the square term where Ψ (h) and Ψ (l) Represents the square terms The high and low positions of Φ (h) and Φ (l) Represents the square terms high and low; The original vector is then represented as follows: Computation based on encryption scheme of arbitrary inner product function<x″,y″> , by calculating<x″,y″> , the equation dist(x,y)=<x″,y″> Still holds true.

Images