Image retrieval method and device based on polynomial access control, equipment and medium

Abstract

The invention discloses an image retrieval method based on polynomial access control, and the method comprises the following steps: carrying out the image feature extraction of an image data set through employing a convolutional neural network model, and obtaining an image feature vector; performing clustering operation on the image features by using a K-means algorithm according to the similarity of the image features; constructing a role polynomial to configure a polynomial access control strategy of each image; and expanding the corresponding image feature vector by using the coefficient of each image polynomial to obtain an access control index tree. According to the method, the clustering operation of the image features is completed by applying the unsupervised learning K-means algorithm, and the method focuses on related categories during retrieval, so that the retrieval efficiency is improved; according to the method, the range of data which can be accessed by individuals is accurately defined through polynomial setting, and the security and confidentiality of image data are ensured; according to the method, a polynomial access control strategy is combined with an index tree structure, so that the access speed of a ciphertext domain is increased while the image retrieval security is ensured.

Description

Technical Field

[0001] This invention belongs to the technical field of blockchain privacy protection and supervision, specifically relating to an accountable privacy-protecting smart contract implementation method, device, computer equipment, and storage medium based on account blockchain. Background Technology

[0002] With the rapid development of cloud computing technology, data owners with limited resources primarily upload images to cloud servers for storage. Searching for specific images from vast datasets is gaining increasing attention in many practical fields, and image retrieval can conveniently leverage cloud servers to find similar images. However, potential security issues continue to hinder the widespread adoption of image outsourcing to cloud servers. Directly uploading images to cloud servers can easily leak personal information, such as facial attributes and identity details. Traditional encryption schemes cannot guarantee the usability of encrypted images or are unusable on resource-constrained devices, while searchable encryption can protect image data privacy while allowing operations such as image retrieval on cloud servers.

[0003] Currently, most privacy-preserving image retrieval solutions suffer from significant problems in everyday applications. On one hand, they suffer from low retrieval accuracy, making it difficult to accurately retrieve the information users truly want, greatly impacting user experience. On the other hand, retrieval efficiency is also low, failing to meet the demand for rapid searching. Furthermore, they lack fine-grained access control. With the increasing number of large-scale image databases across various fields, the demand for efficient image retrieval services is becoming increasingly strong. Efficient retrieval solutions are of great significance, effectively reducing service response time and improving ease of use. While coarse-grained access control can limit access to outsourced image sets for users with corresponding search keys, it still faces numerous security risks from unauthorized access. To ensure the stable operation of the retrieval system under reasonable load, fine-grained access control technology to regulate image access policies for each user is essential. In summary, given the dual requirements of ensuring privacy and security while meeting efficient retrieval needs, designing an access control image retrieval method that balances privacy protection and retrieval efficiency is particularly urgent, and this is key to meeting current practical application and security requirements. Summary of the Invention

[0004] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide an image retrieval method, apparatus, computer device, and storage medium based on polynomial access control.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention discloses an image retrieval method based on multinomial access control, the image retrieval method comprising the following steps:

[0007] S1. Extract image features from the image dataset to obtain a set of image feature vectors. Cluster the image feature vector set based on the similarity of the image features to obtain the hierarchical structure of the image dataset. Then, encrypt the image dataset using the AES algorithm to obtain an encrypted image.

[0008] S2. Construct a polynomial access control policy for each image in the role polynomial configuration image dataset. Expand the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypt it using an asymmetric dot product encryption algorithm to obtain the access control index tree.

[0009] S3. Extract image features from the image data to be retrieved to obtain the image feature vector. Expand the image feature vector according to the multinomial access control strategy, and then encrypt it using the asymmetric dot product encryption algorithm to obtain a searchable trapdoor.

[0010] S4. Using Euclidean distance as the retrieval benchmark, the searchable trapdoors and access control index tree are calculated to return similar encrypted images. The AES algorithm is used to decrypt the returned encrypted images to obtain similar images in the image dataset.

[0011] Furthermore, in step S1, clustering algorithms and image encryption algorithms are used to obtain the hierarchical structure of the image data, and the image encryption process is as follows:

[0012] S1-1. The RegNet convolutional neural network model is used to extract image features from the image dataset, resulting in a set of image feature vectors. This RegNet model is based on the work of Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollar (see Designing network designspaces, page 1). The model takes an image dataset as input, performs image feature extraction, and outputs a set of image feature vectors. The RegNet neural network model improves performance by increasing network width and depth at different network layers, and the extracted features have strong discriminative power, capable of capturing subtle visual differences, thus returning more relevant search results.

[0013] S1-2. Based on the similarity of image features, the K-means algorithm is used to cluster the feature vectors of the images to obtain the hierarchical structure of the image data, as follows;

[0014] First, select K image feature vectors as centroid vectors; calculate the Euclidean distance from each image feature vector to the centroid vector and add it to the set of nearest centroid vectors; update the centroid vectors to the average of all image feature vectors in the current set, until a fixed number of times or the centroid vectors tend to stabilize; if the number of image feature vectors in the set exceeds K, continue the above operation until the number of image feature vectors in each set does not exceed K, thus obtaining the hierarchical structure of the image data. This hierarchical structure organizes a large-scale image database into a tree-like index, transforming nearest neighbor search from a global linear comparison to a local logarithmic comparison, significantly reducing computational complexity. Simultaneously, this structure effectively maintains the local similarity of the feature space, laying the foundation for subsequent accurate retrieval in the encrypted domain.

[0015] S1-3. The image data to be retrieved is encrypted using the AES algorithm to obtain an encrypted image. The AES algorithm, developed by Joan Daemen and Vincent Rijmen (see The Design of Rijndael: AES - The Advanced Encryption Standard, page 1), is a stream of pixel data generated after the input image file is read. Through obfuscation and diffusion, the image data is completely randomized, and the output is a completely random ciphertext data stream unrelated to the original image. Only an authorized party with the same key can decrypt it using the AES decryption process to restore it to the original image. The AES algorithm has extremely high security and can effectively resist various attacks, ensuring the confidentiality of image content during transmission and storage.

[0016] Furthermore, the process of obtaining the access control index tree through the polynomial access control strategy in step S2 is as follows:

[0017] S2-1, Based on the image data set of the first... Zhang Image Access Role Set , For the first visiting character in the visiting character set, similarly, Let ρ be the t-th visiting role in the visiting role set. The number of elements in the image is used to construct the image. Role polynomial The expression is:

[0018]

[0019] in, For role polynomial The root value, where E is a fixed constant value. Represents the degree of each term in the polynomial. For role polynomial The coefficient corresponding to the j-th term;

[0020] According to the role polynomial Configure a polynomial access control policy for each image in the image dataset. This policy uses the uniqueness of the polynomial root to ensure that the access policy cannot be forged. The square operation ensures that even if the role is authorized repeatedly, it will not affect the verification consistency. The coefficients are generated from the original role set, eliminating the need to maintain a complex permission table and significantly reducing storage and computation overhead.

[0021] S2-2. Using the polynomial coefficients of each image, a vector expansion operation is performed on the corresponding image feature vectors in the hierarchical structure to obtain the expanded hierarchical structure. This is done by embedding discrete permission attributes into a continuous feature space through mathematical transformation, so that the expanded feature vectors simultaneously encode visual content and access rules, achieving deep coupling between access strategies and visual features;

[0022] S2-3. An asymmetric dot product encryption algorithm is used to encrypt the expanded hierarchical structure to obtain an access control index tree. The purpose is to achieve strong privacy protection for node content through linear transformation and noise injection, as detailed below:

[0023] For the node corresponding to the v-th image in the expanded hierarchical structure Using an encryption matrix Encryption is performed to obtain the encrypted node. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution. Each node in the expanded hierarchical structure is encrypted to obtain an access control index tree.

[0024] Furthermore, the process of obtaining a searchable trapdoor in step S3 using a neural network model and an asymmetric dot product encryption algorithm is as follows:

[0025] S3-1. The RegNet convolutional neural network model is used to extract image features from the image data to be retrieved, resulting in a feature vector of the image to be retrieved. The feature vector generated by the above neural network model is not only compact but also effectively captures the key visual content of the image, thereby achieving fast and accurate similarity matching in tasks such as image retrieval.

[0026] S3-2. The feature vector of the query image is extended according to the multinomial access control strategy to obtain the extended feature vector of the image to be retrieved. The extended feature vector of the image to be retrieved encodes the image content and also contains access rule information, ensuring that only users with legitimate permissions can successfully perform feature matching and retrieval. This achieves end-to-end secure access control at the feature level and effectively prevents unauthorized data acquisition.

[0027] S3-3. The expanded feature vector of the image to be retrieved is encrypted using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor, as follows:

[0028] The expanded feature vector of the image to be retrieved Using matrices inverse matrix Encryption is performed to obtain a searchable trapdoor. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution with a mean of 0 and a standard deviation of 1. The searchable trapdoor design ensures that only authorized users can generate valid queries, noise injection prevents statistical inference attacks, inverse matrix transformation protects query privacy, and the retrieval process is fully encrypted, while strictly protecting the sensitive information of both the queryer and the data provider.

[0029] Furthermore, the process of calculating similar images through distance sorting in step S4 is as follows:

[0030] S4-1. Using Euclidean distance as the retrieval benchmark, the distance value is calculated between the searchable trapdoors and the access control index tree, as follows:

[0031] Enter searchable trapdoor and nodes of the access control index tree , obtain the distance value By transforming Euclidean distance into a dot product operation of encrypted vectors through algebraic transformations, efficient similarity calculation in the ciphertext domain is cleverly achieved. The calculation process does not require decryption of sensitive data, and while protecting the privacy of queries and indexes, it directly outputs a numerical result proportional to the plaintext distance, balancing retrieval accuracy and data security.

[0032] S4-2. Sort the distance values ​​from smallest to largest to obtain the encrypted image with the smallest distance value. This sorting mechanism achieves efficient encrypted text retrieval by quantifying similarity. Minimizing the distance directly corresponds to the nearest encrypted image in the feature space, ensuring consistency between the retrieval results and the plaintext domain.

[0033] S4-3. Decrypt the encrypted image with the smallest distance using the AES algorithm to obtain similar images in the image dataset. This step achieves a balance between security and efficiency through symmetric decryption. The AES algorithm can quickly reconstruct the retrieved ciphertext image, ensuring that only authorized users can obtain the final plaintext result. This "end-to-end encryption-partial decryption" mechanism minimizes decryption overhead while maintaining overall data security, achieving a balance between privacy protection and practicality.

[0034] Secondly, the present invention provides an image retrieval apparatus based on polynomial access control for executing an image retrieval method based on polynomial access control, the image retrieval apparatus comprising:

[0035] The image encryption module is used to extract image features from the image dataset to obtain a set of image feature vectors. Based on the similarity of image features, the set of image feature vectors is clustered to obtain the hierarchical structure of the image dataset. Finally, the image dataset is encrypted using an image encryption algorithm to obtain an encrypted image.

[0036] The access control index tree construction module is used to construct a polynomial access control policy for each image in the role polynomial configuration image dataset. It expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree.

[0037] The trapdoor construction module is used to extract image features from the image data to be retrieved, obtain the feature vector of the image to be retrieved, expand the feature vector of the image to be retrieved according to the multinomial access control policy, and then encrypt it using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor.

[0038] The image retrieval module uses Euclidean distance as the retrieval benchmark, calculates the searchable trapdoors and access control index tree, returns similar encrypted images, and uses an image decryption algorithm to decrypt the returned encrypted images to obtain similar images from the image dataset.

[0039] Thirdly, the present invention provides a computer device, including a processor and a memory for storing a processor-executable program, characterized in that, when the processor executes the program stored in the memory, it implements the above-described image retrieval method based on polynomial access control.

[0040] Fourthly, the present invention provides a storage medium storing a program, which, when executed by a processor, implements the above-described image retrieval method based on polynomial access control.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] (1) This invention utilizes transfer learning and pre-trained deep learning models, leveraging convolutional neural networks pre-trained on large-scale image sets. Its network structure, through learning from massive amounts of data, has mastered general image patterns. After fine-tuning in this invention, it can deeply mine the deep semantic features of images, more accurately reflecting the true content of the images. Compared to traditional local and global feature retrieval methods, this invention significantly improves retrieval accuracy and better meets practical application needs.

[0043] (2) This invention adopts a polynomial-based access strategy, constructing a role polynomial definition of access permissions for images and their access role sets. It cleverly integrates access roles into the construction of the polynomial, with each factor corresponding to a specific access role, and uses rigorous mathematical operations to reflect the logical relationship of permissions. When it is necessary to add or remove an access role, simply add or remove the corresponding factor in the polynomial, making the operation convenient. Using the polynomial access control strategy, precise changes in permissions can be easily achieved, ensuring both the reasonable use of resources by different roles and strict control over data security. This achieves fine-grained, efficient, and accurate access control, effectively compensating for many shortcomings of complex attribute strategies.

[0044] (3) This invention uses the K-means clustering algorithm to classify all image feature vectors, grouping images with similar features into the same set, fully exploring the inherent correlation between image features, and further constructing a searchable encrypted index tree. When performing image retrieval, one only needs to start from the root node of the searchable index tree and search downwards along the relevant branches according to the predetermined search strategy to quickly and accurately locate the branch where the image with similar features to the target image is located, thereby obtaining the corresponding image resources. This greatly optimizes the entire image retrieval and access control process, improves the overall efficiency of image data management and use, and brings a better user experience. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the image retrieval method based on multinomial access control in Embodiment 1 of the present invention;

[0047] Figure 2 This is a flowchart of the image retrieval method based on multinomial access control in Embodiment 1 of the present invention;

[0048] Figure 3 This is a schematic diagram of the process of generating the access control index tree in Embodiment 1 of the present invention;

[0049] Figure 4 This is a structural block diagram of the image retrieval device based on polynomial access control in Embodiment 4 of the present invention;

[0050] Figure 5 This is a structural block diagram of the computer device in Embodiment 5 of the present invention. Detailed Implementation

[0051] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0052] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0053] Example 1

[0054] like Figure 2As shown, this embodiment presents a fine-grained privacy-preserving image retrieval method based on searchable encryption, image retrieval, and cloud computing. First, the image owner extracts image features from the image dataset to obtain a set of image feature vectors. These feature vectors are then clustered based on feature similarity to create a hierarchical structure. The image dataset is then encrypted using the AES algorithm to obtain encrypted images, ensuring security during transmission and retrieval. Next, the image owner constructs a role polynomial, configuring a polynomial access control policy for each image in the dataset. The coefficients of each polynomial are used to expand the corresponding image feature vectors in the hierarchical structure, which are then encrypted using an asymmetric dot product encryption algorithm to obtain an access control index tree. Subsequently, the encrypted images and the access control index tree are uploaded to a cloud server. The querying user extracts image features from the image data to be retrieved, obtains the feature vectors of the images to be retrieved, expands these vectors according to the polynomial access control policy, and then encrypts them using an asymmetric dot product encryption algorithm to obtain searchable trapdoors. These searchable trapdoors are then uploaded to the cloud server. The cloud server uses Euclidean distance as the retrieval benchmark, calculating the searchable trapdoors and access control index tree to obtain a set of similar encrypted images in ciphertext state, which is then sent to the querying user. The querying user uses the decryption key sent by the image owner to decrypt the similar encrypted images in ciphertext state, obtaining the desired set of similar images. Notably, calculating the searchable trapdoors and access control index tree significantly reduces the number of feature comparisons during retrieval. Retrieval only requires filtering layer by layer along the branches of the access control index tree to quickly locate the set of candidate images similar to the query features, avoiding the full comparison required by traditional linear retrieval. This provides crucial technical support for large-scale, high-dimensional feature-driven image retrieval tasks.

[0055] This embodiment is described in detail below:

[0056] S1. The image owner extracts image features from the image dataset to obtain a set of image feature vectors. Based on the similarity of the image features, the image feature vector set is clustered to obtain a hierarchical structure of the image dataset. Finally, the image dataset is encrypted using the AES algorithm to obtain an encrypted image. Step S1 is detailed as follows:

[0057] S1-1: The image owner uses the RegNet convolutional neural network model to extract image features from the image dataset, obtaining a set of image feature vectors. This RegNet network model is from the works of Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollar, in the paper "Designing network design spaces," page 1. Through its regularized network architecture design and efficient feature representation capabilities, this network model significantly optimizes the accuracy, efficiency, and generalization of retrieval tasks, making it more suitable for large-scale, cross-scene image retrieval needs.

[0058] S1-2. Using the above set of image feature vectors, the K-means clustering algorithm is used to divide the image feature vectors into different clusters based on the similarity of image features. (Selection) We take random image feature vectors as centroid vectors; calculate the distance from each image feature vector to the centroid vector and add it to the nearest cluster; update the centroid vector to the average of all image feature vectors in the current cluster. We iterate through these operations until a fixed number of iterations or the centroid vectors become stable. After the initial partitioning, each subspace may still have too many feature vectors. In this case, to ensure that the number of vectors in each subspace is less than a fixed value, we perform clustering operations on the subspaces with too many vectors. The process continues until each subspace meets the requirement. At this point, each image feature vector becomes a leaf node, resulting in a hierarchical structure.

[0059] In selection When the value is, if Choosing a value that is too small means classifying image features into too few categories, resulting in clustering results that fail to accurately reflect the differences between image features. Real-world image features often have complex and diverse distributions. Values ​​that are too small cannot accurately reflect the actual distribution for reasonable classification. Excessively large values ​​can easily lead to overfitting in clustering results, meaning that image features are over-subdivided, causing the clustering results to reflect only noise or some accidental feature differences in the data, rather than truly meaningful and stable category divisions. Furthermore, As the value increases, the number of clustering algorithms needs to process more categories, and the computational complexity increases significantly. Therefore, we choose [the appropriate algorithm here]. .

[0060] S1-3. The image data to be retrieved is encrypted using the AES algorithm to obtain an encrypted image, which is then uploaded to the cloud server. The AES encryption algorithm, through the high security and efficiency of symmetric cryptography, provides end-to-end privacy protection for the storage, transmission, and retrieval interaction of image data, while also meeting the real-time and compatibility requirements of the retrieval system.

[0061] S2. The image owner constructs a role polynomial, configures the polynomial access control policy for each image in the image dataset, expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree; this step S2 is detailed as follows:

[0062] S2-1, Based on the image data set of the first... Zhang Image Access Role Set ρ represents the set of accessed roles. The number of elements in the image is used to construct the image. Role polynomial The expression is:

[0063]

[0064] in, For role polynomial The root value, where E is a fixed constant value. , Represents the degree of each term in the polynomial. For role polynomial The coefficients corresponding to each item in the table.

[0065] According to the role polynomial Configure a polynomial access control policy for each image in the image dataset.

[0066] For example, setting an image Access Role Set Fixed constant Then its role polynomial If the user accesses the role Then it can make In other words, this means allowing the user to access the image. .

[0067] S2-2, The dimension is The Feature vector of an image Expand to ,in Characteristic vector The length of the mold, To access the coefficients of the polynomial, It follows a normal distribution. A random vector. Using the polynomial coefficients of each of the above images, a vector expansion operation is performed on the corresponding image feature vectors in the hierarchical structure to obtain the expanded hierarchical structure.

[0068] For example, setting the dimension of the image feature vector in the instance. , Follows a normal distribution Introducing low-variance normally distributed noise can make the intra-class feature vectors form a more compact feature vector distribution around the cluster center without increasing the dispersion of intra-class features. This allows for more accurate differentiation of semantically different images during the retrieval and matching stage, reducing the false detection rate.

[0069] S2-3. For any node in the expanded hierarchy... Using invertible matrices Encryption is performed to obtain the encrypted node. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution. Each node in the expanded hierarchical structure is encrypted to obtain an access control index tree.

[0070] like Figure 3 The diagram shown illustrates the process of generating the access control index tree in this example.

[0071] For example, in this embodiment, an amplification factor is set. , Follows a normal distribution In the processing of image feature vectors, the numerical range of feature values ​​may be small, making them susceptible to noise and computational inaccuracies. A larger scaling factor can expand the numerical range of feature vectors, increasing the complexity and obfuscation of the encryption process. It also makes the distribution of similarity calculation results more dispersed, enabling a more accurate reflection of the differences between feature vectors, thereby improving the retrieval efficiency and accuracy of the access control index tree.

[0072] S3. The query user extracts image features from the image data to be retrieved, obtaining the feature vector of the image to be retrieved. This feature vector is then expanded according to a multinomial access control policy, and finally encrypted using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor. The searchable trapdoor is then uploaded to the cloud server. Step S3 is detailed below:

[0073] S3-1. Use the convolutional neural network model Regnet to extract image features from the image data to be retrieved, and obtain the feature vector of the image data to be retrieved.

[0074] S3-2. Query the user based on the image owner's authorization, for users with access roles of... The image to be retrieved ,Will 3D feature vector of the image to be retrieved Expand to That is, the expanded feature vector of the image to be retrieved, where, It is a random positive integer. normal distribution .

[0075] S3-3, Expanding the feature vector of the image to be retrieved Using matrices inverse matrix Encryption is performed to obtain a searchable trapdoor. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution with a mean of 0 and a standard deviation of 1. The searchable trapdoors are then uploaded to the cloud server.

[0076] S4. The cloud server uses Euclidean distance as the retrieval benchmark to calculate the searchable trapdoors and access control index tree, obtaining similar encrypted images in ciphertext state, and then sends them to the querying user. The querying user uses the decryption key sent by the image owner to decrypt the similar encrypted images in ciphertext state, obtaining the similar images from the image dataset. Step S4 is detailed as follows:

[0077] S4-1. Using Euclidean distance as the retrieval criterion, the aforementioned searchable trapdoors are... and nodes in the access control index tree Perform distance calculations. Select nodes sequentially from the root to the leaf nodes of the access control tree. At each level of the access control index tree, determine the entry node for the next level by calculating the node with the minimum Euclidean distance between its node vector and the searchable trapdoor. The distance calculation expression is:

[0078]

[0079] S4-2. Based on the distance calculation results above, the node with the smallest distance value is selected as the entry node for the next layer. When the query reaches a leaf node, it obtains the identifier of the encrypted image most similar to the query image and adds it to the result list. If the number of elements in the result list is less than the required number, it returns to the sibling node or parent node of the leaf node to find the next nearest neighbor node. This process is repeated until the required number of similar images are found, resulting in a set of similar encrypted images in the ciphertext state.

[0080] S4-3. Send the set of similar encrypted images in the above-mentioned ciphertext state to the querying user. The querying user uses the decryption key sent by the image owner to decrypt the set of similar encrypted images in the ciphertext state to obtain the required set of similar images.

[0081] For example, to further verify the effectiveness of this embodiment, the accuracy of image retrieval will be explained. The Corel-10k image dataset is used as the image data set of the image owners, and then 1000 images are randomly selected from it as the set of images to be retrieved owned by the query user. The Corel-10k image dataset contains 10,000 images, with 100 image categories, and each category contains 100 similar images. As an indicator of accuracy, among them, This indicates the number of similar encrypted images in the ciphertext state returned by the cloud server. Indicates the number of similar images. The closer the value is to 1, the higher the accuracy of the search.

[0082] Table 1. Comparison of image retrieval accuracy with and without polynomial access control strategies

[0083]

[0084] Table 1 shows the number of images returned. The accuracy of image retrieval was compared when the multinomial access control strategy was used, increasing from 10 to 50. Table 1 shows that in this example, using a multinomial access control strategy had no impact on image retrieval accuracy, achieving a perfect decoupling of security control and functional retrieval. This means that image owners can configure strict security policies to the maximum extent possible without sacrificing retrieval accuracy, thus fundamentally ensuring the high practicality and deployability of the solution in real-world scenarios.

[0085] Example 2

[0086] This embodiment follows Figure 2 The execution flow of the image retrieval method based on multinomial access control involves modifications to the image encryption method. The detailed flow is as follows:

[0087] S1. The image owner extracts image features from the image dataset to obtain a set of image feature vectors. Based on the similarity of image features, the image feature vector set is clustered to obtain a hierarchical structure of the image dataset. The image dataset is then encrypted using an image encryption algorithm based on compressed sensing and block DNA coding to obtain an encrypted image. Step S1 is detailed as follows:

[0088] S1-1: Use a convolutional neural network model to extract image features from the image dataset to obtain a set of image feature vectors. Refer to step S1-1 in Example 1 for obtaining the set of image feature vectors.

[0089] S1-2: Using the above set of image feature vectors, a hierarchical structure is obtained by using the K-means clustering algorithm based on the similarity of image features. Refer to S1-2 in Example 1 for the step of obtaining the hierarchical structure.

[0090] S1-3. The image data to be retrieved is encrypted using an image encryption algorithm based on compressed sensing and block DNA encoding to obtain an encrypted image, which is then uploaded to a cloud server. The image encryption method based on compressed sensing and block DNA encoding first obtains the chaotic sequence used in the encryption process and constructs a measurement matrix through an iterative composite chaotic system. Compressed sensing is then used to compress the color image to obtain three measurement value matrices, which are then quantized into integer matrices and scrambled using a confusion method based on the Josephus problem. Subsequently, these scrambled measurement value matrices are divided into sub-blocks. These sub-blocks undergo dynamic DNA encoding, DNA diffusion, and DNA decoding, and are then reassembled to obtain the ciphertext image. This method compresses the plaintext image to obtain a small-sized encrypted image, while ensuring that the encrypted image can resist noise attacks, blocking attacks, statistical attacks, chosen-plaintext attacks, and known-plaintext attacks during transmission.

[0091] S2. The image owner constructs a role polynomial, configures the polynomial access control policy for each image in the image dataset, expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree. Refer to S2 in Example 1 for the steps of obtaining the access control index tree.

[0092] S3. The query user extracts image features from the image data to be retrieved, obtains the feature vector of the image to be retrieved, expands the feature vector of the image to be retrieved according to the multinomial access control policy, and then encrypts it using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor, and uploads the searchable trapdoor to the cloud server. Refer to S3 in Example 1 to obtain the searchable trapdoor step.

[0093] S4. The cloud server uses Euclidean distance as the retrieval benchmark to calculate the searchable trapdoors and access control index tree to obtain similar encrypted images in ciphertext state, and then sends them to the querying user. The querying user uses the decryption key sent by the image owner to decrypt the similar encrypted images in ciphertext state to obtain the similar images in the image dataset. Refer to step S4 in Example 1 for obtaining similar images.

[0094] Example 3

[0095] This embodiment follows Figure 2 The execution flow of the image retrieval method based on multinomial access control involves modifications to the image feature extraction method. The detailed flow is as follows:

[0096] S1. The image owner extracts image features from the image dataset to obtain a set of image feature vectors. Based on the similarity of the image features, the image feature vector set is clustered to obtain a hierarchical structure of the image dataset. Finally, the image dataset is encrypted using the AES algorithm to obtain an encrypted image. Step S1 is detailed as follows:

[0097] S1-1: The image owner uses the DenseNet convolutional neural network model to extract image features from the image dataset, obtaining a set of image feature vectors. This DenseNet network model is from the work of Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kilian Q. Weinberger, in the paper *Densely Connected Convolutional Networks*, page 1. This network model deeply integrates shallow detail textures, mid-level object parts, and deep semantic concepts, enabling the extracted global image features to simultaneously contain detailed local information and high-level semantic information. This allows for more accurate similarity matching when faced with deformation, occlusion, or complex backgrounds.

[0098] S1-2: Using the above set of image feature vectors, a hierarchical structure is obtained by using the K-means clustering algorithm based on the similarity of image features. Refer to S1-2 in Example 1 for the step of obtaining the hierarchical structure.

[0099] S1-3. Encrypt the image data to be retrieved using the AES algorithm to obtain an encrypted image, and then upload the encrypted image to the cloud server. Refer to S1-3 in Example 1 for the step of obtaining the encrypted image.

[0100] S2. The image owner constructs a role polynomial, configures the polynomial access control policy for each image in the image dataset, expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree. Refer to S2 in Example 1 for the steps of obtaining the access control index tree.

[0101] S3. The query user extracts image features from the image data to be retrieved, obtaining the feature vector of the image to be retrieved. This feature vector is then expanded according to a multinomial access control policy, and finally encrypted using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor. The searchable trapdoor is then uploaded to the cloud server. Step S3 is detailed below:

[0102] S3-1. Use the DenseNet convolutional neural network model to extract image features from the image data to be retrieved, and obtain the feature vector of the image data to be retrieved.

[0103] S3-2. Query the user based on the image owner's authorization, for users with access roles of... The image to be retrieved ,Will 3D feature vector of the image to be retrieved Expand to That is, the expanded feature vector of the image to be retrieved, where, It is a random positive integer. normal distribution .

[0104] S3-3, Expanding the feature vector of the image to be retrieved Using matrices inverse matrix Encryption is performed to obtain a searchable trapdoor. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution with a mean of 0 and a standard deviation of 1. The searchable trapdoors are then uploaded to the cloud server.

[0105] S4. The cloud server uses Euclidean distance as the retrieval benchmark to calculate the searchable trapdoors and access control index tree to obtain similar encrypted images in ciphertext state, and then sends them to the querying user. The querying user uses the decryption key sent by the image owner to decrypt the similar encrypted images in ciphertext state to obtain the similar images in the image dataset. Refer to step S4 in Example 1 for obtaining similar images.

[0106] For example, to further verify the effectiveness of this embodiment, the accuracy of image retrieval will be explained. The Corel-10k image dataset is used as the image data set of the image owners, and then 1000 images are randomly selected from it as the set of images to be retrieved owned by the query user. The Corel-10k image dataset contains 10,000 images, with 100 image categories, and each category contains 100 similar images. As an indicator of accuracy, among them, This indicates the number of similar encrypted images in the ciphertext state returned by the cloud server. Indicates the number of similar images. The closer the value is to 1, the higher the accuracy of the search.

[0107] Table 2. Comparison of image retrieval accuracy under different image feature extraction methods

[0108]

[0109] In Table 2, when the number of returned images is... As the value increases from 10 to 50, the accuracy of image retrieval using two image feature extraction methods is compared. DenseNet represents feature extraction from image data using the DenseNet neural network model, while CLD represents extracting the color layout descriptor of the image as its feature vector. Table 2 shows that the retrieval accuracy achieved using the DenseNet image feature extraction method is higher than that achieved using the CLD image feature extraction method. This indicates that, by leveraging a convolutional neural network pre-trained on a large-scale image set and fine-tuning it in this invention, it can deeply mine the deep semantic features of images, more accurately reflecting the true content of the images. Compared to traditional local and global feature retrieval methods, this invention significantly improves retrieval accuracy and better meets practical application needs.

[0110] Example 4

[0111] Reference Figure 4 This embodiment provides an image retrieval device based on polynomial access control. The device includes an image encryption module 401, an access control index tree construction module 402, a trapdoor construction module 403, and an image retrieval module 404 connected sequentially.

[0112] The image encryption module 401 is used to extract image features from the image data set to obtain an image feature vector set, perform clustering operations on the image feature vector set based on the similarity of image features to obtain the hierarchical structure of the image data set, and encrypt the image data set using an image encryption algorithm to obtain an encrypted image.

[0113] The access control index tree construction module 402 is used to construct a polynomial access control policy for each image in the role polynomial configuration image data set. It expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree.

[0114] The trapdoor construction module 403 is used to extract image features from the image data to be retrieved, obtain the feature vector of the image to be retrieved, expand the feature vector of the image to be retrieved according to the polynomial access control strategy, and then encrypt it using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor.

[0115] The image retrieval module 404 uses Euclidean distance as the retrieval benchmark, calculates the searchable trapdoors and access control index tree, returns similar encrypted images, and uses an image decryption algorithm to decrypt the returned encrypted images to obtain similar images in the image dataset.

[0116] The content of the above method embodiments is applicable to the device embodiments, and the specific functions implemented in the device embodiments are the same as those in the above method embodiments.

[0117] Example 5

[0118] This embodiment provides a computer device, which can be a computer, such as... Figure 5 As shown, the system bus 501 connects a processor 502, a memory, an input device 503, a display 504, and a network interface 505. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium 506 and internal memory 507. The non-volatile storage medium 506 stores the operating system, computer programs, and a database. The internal memory 507 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the processor 502 executes the computer programs stored in the memory, it implements the image retrieval method based on polynomial access control proposed in Embodiment 1. The image retrieval method based on polynomial access control includes the following steps:

[0119] S1. Extract image features from the image dataset to obtain a set of image feature vectors. Perform clustering operations on the set of image feature vectors based on the similarity of image features to obtain the hierarchical structure of the image dataset. Then, encrypt the image dataset using an image encryption algorithm to obtain an encrypted image.

[0120] S2. Construct a polynomial access control policy for each image in the role polynomial configuration image dataset. Expand the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypt it using an asymmetric dot product encryption algorithm to obtain the access control index tree.

[0121] S3. Extract image features from the image data to be retrieved to obtain the image feature vector. Expand the image feature vector according to the multinomial access control strategy, and then encrypt it using the asymmetric dot product encryption algorithm to obtain a searchable trapdoor.

[0122] S4. Using Euclidean distance as the retrieval benchmark, the searchable trapdoors and access control index tree are calculated to return similar encrypted images. The returned encrypted images are then decrypted using an image decryption algorithm to obtain similar images from the image dataset.

[0123] Example 6

[0124] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the image retrieval method based on polynomial access control proposed in Embodiment 1 above. The image retrieval method based on polynomial access control includes the following steps:

[0125] S1. Extract image features from the image dataset to obtain a set of image feature vectors. Perform clustering operations on the set of image feature vectors based on the similarity of image features to obtain the hierarchical structure of the image dataset. Then, encrypt the image dataset using an image encryption algorithm to obtain an encrypted image.

[0126] S2. Construct a polynomial access control policy for each image in the role polynomial configuration image dataset. Expand the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypt it using an asymmetric dot product encryption algorithm to obtain the access control index tree.

[0127] S3. Extract image features from the image data to be retrieved to obtain the image feature vector. Expand the image feature vector according to the multinomial access control strategy, and then encrypt it using the asymmetric dot product encryption algorithm to obtain a searchable trapdoor.

[0128] S4. Using Euclidean distance as the retrieval benchmark, the searchable trapdoors and access control index tree are calculated to return similar encrypted images. The returned encrypted images are then decrypted using an image decryption algorithm to obtain similar images from the image dataset.

[0129] The storage medium described in this embodiment can be a disk, optical disk, computer memory, random access memory (RAM), USB flash drive, portable hard drive, etc.

[0130] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0131] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0132] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An image retrieval method based on multinomial access control, characterized in that, The image retrieval method includes the following steps: S1. Extract image features from the image dataset to obtain a set of image feature vectors. Perform clustering operations on the set of image feature vectors based on the similarity of image features to obtain the hierarchical structure of the image dataset. Then, encrypt the image dataset using an image encryption algorithm to obtain an encrypted image. S2. Construct a polynomial access control policy for each image in the role polynomial configuration image dataset. Expand the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypt it using an asymmetric dot product encryption algorithm to obtain the access control index tree. S3. Extract image features from the image data to be retrieved to obtain the image feature vector. Expand the image feature vector according to the multinomial access control strategy, and then encrypt it using the asymmetric dot product encryption algorithm to obtain a searchable trapdoor. S4. Using Euclidean distance as the retrieval benchmark, the searchable trapdoors and access control index tree are calculated to return similar encrypted images. The returned encrypted images are then decrypted using an image decryption algorithm to obtain similar images from the image dataset.

2. The image retrieval method according to claim 1, characterized in that, The hierarchical structure of the image data and the image encryption process obtained in step S1 are as follows: S1-1. Use a convolutional neural network model to extract image features from the image dataset to obtain a set of image feature vectors; S1-2. Based on the similarity of image features, the K-means algorithm is used to cluster the feature vectors of the images to obtain the hierarchical structure of the image data, as follows; First, select K image feature vectors as centroid vectors; calculate the Euclidean distance from each image feature vector to the centroid vector and add it to the set of nearest centroid vectors; update the centroid vectors to the average value of all image feature vectors in the current set, until a fixed number of times or the centroid vectors tend to stabilize; if the number of image feature vectors in the set exceeds K, continue to perform the above operation until the number of image feature vectors in each set does not exceed K, and the hierarchical structure of the image data can be obtained. S1-3. Encrypt the image data to be retrieved using an image encryption algorithm to obtain an encrypted image.

3. The image retrieval method according to claim 1, characterized in that, The process of obtaining the access control index tree in step S2 is as follows: S2-1, Based on the image data set of the first... Zhang Image Access Role Set , For the first visiting character in the visiting character set, Let ρ be the t-th visiting role in the visiting role set. The number of elements in the image is used to construct the image. Role polynomial The expression is: ,in, For role polynomial The root value, where E is a fixed constant value. Represents the degree of each term in the polynomial. For role polynomial The coefficient corresponding to the j-th term; According to the role polynomial Configure the polynomial access control policy for each image in the image dataset; S2-2. Using the polynomial coefficients of each image, perform vector expansion operation on the corresponding image feature vectors in the hierarchical structure to obtain the expanded hierarchical structure. S2-3. Use an asymmetric dot product encryption algorithm to encrypt the expanded hierarchical structure to obtain the access control index tree, as follows: For the node corresponding to the z-th image in the expanded hierarchy Using an encryption matrix Encryption is performed to obtain the encrypted node. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution. Each node in the expanded hierarchical structure is encrypted to obtain an access control index tree.

4. The image retrieval method according to claim 3, characterized in that, The process of obtaining a searchable trapdoor in step S3 is as follows: S3-1. Use a convolutional neural network model to extract image features from the image data to be retrieved, and obtain the feature vector of the image to be retrieved. S3-2. Perform an expansion operation on the feature vector of the query image according to the polynomial access control strategy to obtain the expanded feature vector of the image to be retrieved. S3-3. The expanded feature vector of the image to be retrieved is encrypted using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor, as follows: The expanded feature vector of the image to be retrieved Using matrices inverse matrix Encryption is performed to obtain a searchable trapdoor. The expression is: ,in, As an expansion factor, It is a noise vector that follows a normal distribution with a mean of 0 and a standard deviation of 1.

5. The image retrieval method according to claim 4, characterized in that, The process of obtaining similar images in step S4 is as follows: S4-1. Using Euclidean distance as the retrieval benchmark, the distance value is calculated between the searchable trapdoors and the access control index tree, as follows: Enter searchable trapdoor and nodes of the access control index tree , obtain the distance value ; S4-2. Sort the distance values ​​from smallest to largest to obtain the encrypted image with the smallest distance value; S4-3. Decrypt the encrypted image with the smallest distance using an image encryption algorithm to obtain similar images in the image dataset.

6. An image retrieval apparatus based on polynomial access control, used to execute the image retrieval method based on polynomial access control as described in any one of claims 1 to 5, characterized in that, The image retrieval device includes: The image encryption module is used to extract image features from the image dataset to obtain a set of image feature vectors. Based on the similarity of image features, the set of image feature vectors is clustered to obtain the hierarchical structure of the image dataset. Finally, the image dataset is encrypted using an image encryption algorithm to obtain an encrypted image. The access control index tree construction module is used to construct a polynomial access control policy for each image in the role polynomial configuration image dataset. It expands the corresponding image feature vector in the hierarchical structure using the coefficients of each image polynomial, and then encrypts it using an asymmetric dot product encryption algorithm to obtain the access control index tree. The trapdoor construction module is used to extract image features from the image data to be retrieved, obtain the feature vector of the image to be retrieved, expand the feature vector of the image to be retrieved according to the multinomial access control policy, and then encrypt it using an asymmetric dot product encryption algorithm to obtain a searchable trapdoor. The image retrieval module uses Euclidean distance as the retrieval benchmark, calculates the searchable trapdoors and access control index tree, returns similar encrypted images, and uses an image decryption algorithm to decrypt the returned encrypted images to obtain similar images from the image dataset.

7. A computer device comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the image retrieval method based on polynomial access control as described in any one of claims 1 to 5.

8. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the image retrieval method based on polynomial access control as described in any one of claims 1 to 5.

Images