A blockchain-based method for secure storage, management, and sharing of screen capture images

Abstract

The present invention discloses a method for secure storage, management and sharing of screen capture images based on blockchain, which belongs to the field of image processing and blockchain technology. The method comprises: a screen capture user uploader and a screen capture user visitor register in a blockchain network; dynamically encrypting and feature extracting the screen capture image, and constructing an index of the screen capture image; when the screen capture user visitor sends a request to the screen capture user uploader; the screen capture user visitor uses keyword retrieval to obtain the required screen capture image data by searching in a distributed file system, and obtains the required screen capture image through an image decryption key. The present invention realizes the secure storage, management and sharing of screen capture image data by combining blockchain with smart contracts; destroys the statistical characteristics of the screen capture image by dynamic encryption based on chaotic mapping to ensure the storage security of the image; and realizes high-precision, secure and private image retrieval by using a lightweight network model combined with GMM quantization and searchable encryption technology.

Description

Technical Field

[0001] The present invention relates to a screen capture image management method, specifically a blockchain-based screen capture image secure storage management and sharing method, belonging to the field of image processing and blockchain technology. Background Art

[0002] With the continuous advancement of the internet and computer technology, and in an era of rapid digital information growth, more and more information is presented in the form of screen content, including images, videos, text, web pages, and windows. Screen capture technology has been widely used in various scenarios, such as e-government, electronic contract signing, online education, telemedicine, and financial transactions. A screen capture image is an image captured by a camera (such as a mobile phone or camera) of the content on a display screen. Due to its convenience and ease of use, this method has gained widespread application for information recording and dissemination. However, security issues with screen capture images have also become prominent, primarily manifesting in data tampering, forgery, unauthorized dissemination, and storage security risks. In recent years, the rise of blockchain technology has provided a convenient and efficient solution for the secure storage and management of screen capture images. Leveraging the decentralized and tamper-proof nature of blockchain, it effectively ensures the integrity of screen capture image data and provides a traceable storage management solution to ensure data authenticity. Smart integration allows for the configuration of access rights to ensure that screen capture images are accessible only to authorized users, preventing unauthorized leakage and tampering. At the same time, the use of the InterPlanetary File System (IPFS) combined with encryption technology not only avoids the single point failure problem of traditional centralized storage, improves data availability and anti-attack capabilities, but also further enhances the security protection capabilities of screen capture images and ensures data confidentiality.

[0003] Traditional screen capture image storage management usually adopts a centralized architecture, and user identities and meeting data are managed and verified through centralized servers. Image files are stored on centralized servers or cloud storage platforms, making them easy targets for hacker attacks. Once a data leak occurs on the central server, all stored screen capture images may face security risks. In addition, centralized storage methods may also face problems such as server crashes and data loss, making it difficult to meet high-reliability storage requirements and posing a serious single point failure problem. In addition, traditional encryption technology relies on a key management mechanism, and the storage and distribution of keys remains a security risk. Especially in a decentralized storage environment, how to ensure that keys are not leaked or stolen by malicious attackers remains a difficult problem to solve. Liu et al. (Liu Y, Chang Y. Blockchain-Based Method for Spatial Retrieval and Verification of Remote Sensing Images[J]. Sensors, 2024, 24(7): 2078.) focused on image retrieval on the blockchain, using the Geohash algorithm to linearize the spatial information of remote sensing images and integrating it with the LSM tree for effective retrieval and verification, but the efficiency of constructing the retrieval index is still low.

[0004] While research has made significant progress in integrating blockchain with encrypted image storage and management, personalized system solutions for screen captures remain insufficient. Furthermore, insufficient attention is paid to privacy protection issues involving multiple entities involved in the management process, and insufficient coordination between on-chain and off-chain storage makes it difficult to balance security and cost. Therefore, ensuring efficient and convenient storage and management of screen captures while ensuring image security and the privacy of participating users is a pressing issue. Summary of the Invention

[0005] Purpose of the invention: In response to the above problems, the purpose of the present invention is to provide a blockchain-based method for secure storage, management and sharing of screen capture images.

[0006] Technical solution: The present invention provides a blockchain-based screen capture image secure storage, management, and sharing method, comprising the following steps:

[0007] Screen capture user uploaders and screen capture user visitors are registered in the blockchain network according to the provisions of a dedicated smart contract;

[0008] The screen capture user uploader dynamically encrypts and extracts features from the screen capture image, builds an index of the screen capture image, and then uploads it to the distributed file system, while also building an index list in the distributed file system.

[0009] When a screen capture user visitor wants to obtain the corresponding screen capture image, a request is sent to the screen capture user uploader; if the screen capture user uploader agrees to share, a reply is sent to the screen capture user visitor through a secure channel with the corresponding image decryption key attached;

[0010] Screen capture user visitors use keyword retrieval to obtain the required screen capture image data by searching in the distributed file system, and obtain the required screen capture image through the image decryption key.

[0011] Furthermore, the steps of dynamically encrypting and extracting features of the screenshot image by the screenshot user uploader include:

[0012] A screen capture image encryption algorithm based on hybrid chaotic mapping and dynamic S-box is used to dynamically encrypt screen capture images.

[0013] Furthermore, the screen capture image encryption algorithm based on the hybrid chaotic map and the dynamic S-box includes the following steps:

[0014] Use Logistic mapping to diffuse the pixels of the screen shot image;

[0015] The Henon map is used to dynamically construct the S-box and perform scrambling operations on the pixel positions of the screen shot image.

[0016] Furthermore, the step of diffusing the pixels of the screen shot image using the Logistic map includes:

[0017] Using the initial value of the given Logistic map and control parameters Perform loop iteration to generate chaotic values. The formula is:

[0018] ,

[0019] Where, Indicates the The chaotic value generated by the iteration, the control parameter The range is (0,4);

[0020] Using chaotic values ​​to form chaotic sequences , transform the chaotic sequence s and the image to be encrypted into a one-dimensional sequence Perform XOR operation to obtain a one-dimensional sequence , the formula is:

[0021] ,

[0022] Where, Represents the exclusive OR operation.

[0023] Furthermore, the steps of dynamically constructing an S-box using the Henon map and performing a scrambling operation on the pixel positions of the screen shot image include:

[0024] The Logistic map generates two chaotic sequences, which are used to construct the parameters in the Henon map. and , the formula is:

[0025] ,

[0026] ,

[0027] Where, and represents the initial value of state n in the Henon map; and Indicates the result of the operation in the next state n+1 after one operation in state n;

[0028] A random 16×16 matrix consisting of integers between 0 and 255 is constructed without duplication. The pixels in the matrix are transformed using the Henon map. The S-box is constructed after iteration using the Henon map formula.

[0029] The one-dimensional sequence Restore it to an image, divide the image into four blocks, and use different S-boxes to scramble the pixel positions of each block of the image to obtain the encrypted screen shot image.

[0030] Furthermore, the step of dynamically encrypting and extracting features of the screenshot image by the screenshot user uploader also includes:

[0031] The lightweight CNN model is adjusted, and the adjusted lightweight CNN model is used to perform feature extraction on the screen capture image to obtain a feature vector W of the screen capture image.

[0032] Furthermore, the step of constructing an index of the screen shot image includes:

[0033] The index corresponding to the screen shot image is constructed based on the dynamic quantization method of Gaussian mixture model.

[0034] Furthermore, the step of constructing an index corresponding to the screen shot image using a dynamic quantization method based on a Gaussian mixture model includes:

[0035] Divide the feature vector W into n subspaces and calculate the significance score of each subspace in the attention map through the attention mechanism , according to the significance score Identify the key areas in the screenshot image and assign weights to the subspace containing the key areas. The formula is:

[0036] ,

[0037] Where, represents the weight value of the i-th subspace;

[0038] Train a Gaussian mixture model for each subspace to obtain the quantization center of each subspace, and use the mean vector of the Gaussian mixture model as the quantization center;

[0039] The eigenvector W is quantized using the mean vector of the Gaussian mixture model. The distance between the eigenvector and each quantization center is calculated, and the quantization center with the smallest distance is selected for quantization.

[0040] Combine the quantization results of each subspace to generate the final quantized index , each index corresponds to the corresponding screen capture image data.

[0041] Furthermore, the steps of obtaining the required screenshot image data by searching in the distributed file system using keywords include:

[0042] Step 401: Generate an initialization key:

[0043] The security parameter λ is used as input. The parameter λ determines the group order p and uses a bilinear mapping. , generate cyclic group G, randomly select ,in Is a cyclic group of order p, construct two hash functions , , get the global parameter set , randomly selected , output key pair , where the public key , private key ;

[0044] Step 402: Build an encrypted index:

[0045] Using public keys , randomly selected , for keywords And the quantized index set Encryption, computing , get the ciphertext , output the ciphertext c and encrypted index set corresponding to the searched keyword ;

[0046] Step 403: Generate a trapdoor:

[0047] Input query image feature vector and private key , output trapdoor ;

[0048] Step 404: Search the encrypted index:

[0049] Input generated retrieval trapdoor and encrypted indexes Perform matching. If the match fails, an error symbol is returned. Otherwise, the addresses of the top-N images with the highest similarity are returned. ;

[0050] Step 405, update the encrypted index: Screen capture user uploader using the search trap Update image index data and use the improved product quantization method to construct the index of the newly added image data .

[0051] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0052] 1. Enhanced dynamic encryption strength of screen capture images:

[0053] To address the vulnerability of high-frequency details in screen capture images, such as text and interface icons, to statistical analysis attacks, this paper proposes a dynamic encryption algorithm based on chaotic mapping. This algorithm diffuses pixel values ​​using logistic mapping and combines it with Henon mapping to dynamically generate S-boxes for pixel scrambling. This destroys the statistical characteristics of the image, ensuring that the encrypted image cannot be deduced from its original content through statistical laws. This significantly enhances the protection of sensitive information in screen capture images.

[0054] 2. Privacy retrieval in encrypted state is safe and efficient:

[0055] To address the low efficiency and insufficient precision of traditional retrieval methods in encrypted scenarios, this paper proposes a combination of an improved product quantization GMM model and searchable encryption technology. This model optimizes the quantization center of feature vectors using a Gaussian mixture model, reducing the quantization error of similar screen capture images. Furthermore, this technology, combined with searchable encryption technology, generates a secure trapdoor, allowing users to quickly match encrypted indexes and return high-precision search results without revealing their query intent. This improves both retrieval efficiency and precision while protecting privacy.

[0056] 3. Lightweight on-chain and off-chain collaborative storage architecture:

[0057] To address the contradiction between the high storage cost of blockchain and the insufficient security of decentralized storage, this invention proposes a collaborative mechanism between on-chain lightweight evidence storage and off-chain encrypted storage, storing the hash value and encrypted index of the screenshot image on the blockchain to ensure data integrity and immutability; using IPFS distributed storage to store encrypted image data to avoid the risk of centralized single point failure; and automating the management of retrieval requests, copyright registration and verification processes through smart contracts to reduce manual intervention and improve system scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flowchart of a blockchain-based method for secure storage, management, and sharing of screenshot images.

[0059] Figure 2 A diagram showing the framework for secure storage, management, and sharing of screen capture images.

[0060] Figure 3 Flowchart for dynamic and secure encryption of screen capture images;

[0061] Figure 4 To improve the MobileNetV3 network structure diagram;

[0062] Figure 5 Compare the original image, encrypted image and decrypted image;

[0063] Figure 6 Comparison of grayscale histograms of original image, encrypted image and decrypted image;

[0064] Figure 7 Comparison of retrieval mAP under different feature vector dimensions;

[0065] Figure 8 Comparison curve of top-N retrieval accuracy of three methods. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of this application more clear, this application is further described in detail below with reference to the accompanying drawings and embodiments.

[0067] The flowchart of the blockchain-based screen capture image security storage management and sharing method described in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0068] Step 1: Screen capture user uploaders and screen capture user visitors register in the blockchain network according to the provisions of the dedicated smart contract.

[0069] like Figure 2As shown, the main entities required in the implementation process of the screen capture image security storage management and sharing method described in this embodiment are three entities, namely local users, distributed file systems and distributed blockchain networks. Local users include screen capture user uploaders and screen capture user visitors. Local users have specific local computers on which they process images to generate the final output image. Screen capture user uploaders are as follows: Figure 2 User 1 shown in the figure needs to perform a series of operations in a predefined manner to create an encrypted file for secure and protected upload; Screen capture of user visitor such as Figure 2 As shown in Figure 2, user 2 retrieves the required screenshot image data through access control and secure authentication. The distributed file system IPFS is used to store encrypted images in a manner that avoids including any supplementary metadata that could easily reveal the owner's identity or image details. After uploading, distinct hashes are generated and saved on the blockchain along with the associated data. The blockchain network provides a peer-to-peer network where users can register to share files previously stored on IPFS, which are not directly stored on the blockchain to maintain scalability and reduce latency, while also maintaining a list of shared data. Smart contracts are designed to allow user registration to define access control constraints while simultaneously uploading information to the blockchain. Smart contracts also provide users with a way to search for specific images within the index list based on their requirements. Smart contract provisions refer to the multiple execution functions set by users to access screenshot images or communicate with each other via the blockchain, as shown in Table 1. These provisions can be automated to ensure the complete implementation of each function and communication confirmation. User 1 registers on the blockchain network using a dedicated smart contract already deployed on the network. User details are stored in a structure called the user list. The smart contract ensures that user 1 is not already registered on the platform. User 2, the visitor of the image, is also registered in the blockchain network according to the provisions of the smart contract. His detailed information is also added to the user list and he can access relevant information after verification.

[0070] In step 2, the screen capture user uploader dynamically encrypts and extracts features of the screen capture image, builds an index of the screen capture image, and then uploads it to the distributed file system, and builds an index list in the distributed file system.

[0071] Furthermore, the steps of dynamically encrypting and extracting features of the screenshot image by the screenshot user uploader include:

[0072] A screen capture image encryption algorithm based on hybrid chaotic mapping and dynamic S-box is used to dynamically encrypt screen capture images.

[0073] To further improve the security of raw image data storage, this example proposes a screen capture image encryption algorithm based on a hybrid chaotic map and dynamic S-boxes to securely encrypt screen capture image data. The encryption algorithm consists of two phases: the first, the diffusion phase, uses a logistic map to diffuse image pixels. The second, the scrambling phase, primarily utilizes a Henon map to dynamically construct S-boxes to scramble image pixel positions.

[0074] Combine Figure 3 As shown, the screen capture image encryption algorithm based on the hybrid chaotic map and the dynamic S-box includes the following steps for dynamically encrypting the screen capture image:

[0075] Use Logistic mapping to diffuse the pixels of the screen shot image;

[0076] The Henon map is used to dynamically construct the S-box and perform scrambling operations on the pixel positions of the screen shot image.

[0077] Specifically, the steps of diffusing the pixels of the screen shot image using the Logistic map include:

[0078] Using the initial value of the given Logistic map and control parameters Perform loop iteration to generate chaotic values. The formula is:

[0079] ,

[0080] Where, Indicates the The chaotic value generated by the iteration, the control parameter The range is (0,4);

[0081] Using chaotic values ​​to form chaotic sequences , transform the chaotic sequence s and the image to be encrypted into a one-dimensional sequence Perform XOR operation to obtain a one-dimensional sequence , the formula is:

[0082] ,

[0083] Where, Represents the exclusive OR operation.

[0084] Specifically, the steps of dynamically constructing an S-box using the Henon map and performing a scrambling operation on the pixel positions of the screen shot image include:

[0085] The Logistic map generates two chaotic sequences, which are used to construct the parameters in the Henon map. and , the formula is:

[0086] ,

[0087] ,

[0088] Where, and represents the initial value of state n in the Henon map; and Indicates the result of the operation in the next state n+1 after one operation in state n;

[0089] A 16×16 matrix with no repetition and consisting of integers between 0 and 255 is randomly constructed. The pixels in the matrix are transformed using the Henon map. The S-box is constructed after iteration through the Henon map formula. The essence is a 256-element arrangement. A chaotic sequence is generated through the Henon map. The values ​​of the chaotic sequence are used as permutation indexes to perform multiple rounds of permutation on the S-box to obtain 4 different dynamic S-boxes: S1, S2, S3, and S4. The one-dimensional sequence Restore it to an image, divide the image into four blocks, and use different S-boxes, namely S1, S2, S3, and S4, to scramble the pixel positions of each block of the image to obtain the encrypted screen shot image.

[0090] Furthermore, the step of dynamically encrypting and extracting features of the screenshot image by the screenshot user uploader also includes:

[0091] The lightweight CNN model is adjusted, and the adjusted lightweight CNN model is used to perform feature extraction on the screen capture image to obtain a feature vector W of the screen capture image.

[0092] The lightweight MobileNetV3 CNN model was adopted and specifically tuned. MobileNetV3, with its efficient depthwise separable convolutions and lightweight design, is well-suited for the real-time processing of screenshot images. Input images were uniformly normalized to 224×224 pixels, preserving detailed information about interface elements such as buttons, menu bars, and text areas. This normalization eliminates scale variations caused by varying screen resolutions or shooting angles, ensuring the MobileNetV3 model's sensitivity to local features. Furthermore, redundant deep convolutional layers in MobileNetV3 were removed, retaining the first 10 layers to focus on extracting low-level features such as edges and textures. This adjustment is based on the relatively simple semantic information of screenshot images, as global feature extraction using deep networks can introduce noise. Furthermore, enhancements such as local occlusion (to simulate partially covered screenshots) and brightness adjustment (to simulate screen reflections or low-light capture) were applied to the training data to further improve the MobileNetV3 model's adaptability to screenshot distortion and quality fluctuations in real-world scenarios. Training data was from the publicly available COCO dataset, with 10,000 images selected for the experiments. The network structure used is as follows Figure 4 As shown, the shallow network is retained to focus on local details (such as buttons and text areas) in the captured image, while the last three bottleneck layers and the classification layer of the original network are removed. The input image first passes through a 3×3 convolution before entering the bottleneck layer. Within this bottleneck layer, an expansion layer increases the channel dimension, improving the model's expressiveness. Depthwise convolution processes spatial information channel by channel through 3×3 convolutions, while pointwise convolution adjusts the number of channels through 1×1 convolutions. All three components then undergo batch normalization and activation functions to optimize the training process and enhance model nonlinearity. Finally, a 1×1 convolution is performed to output the final result.

[0093] Before extracting feature vectors, the original screenshot image data must be preprocessed to match the input requirements of the MobileNetV3 network. , use the Min-Max normalization operation to scale the image pixel values ​​from the original range to [0,1]. The normalization formula is:

[0094] ,

[0095] in, is the original image at position The pixel value at represents the normalized pixel value, and The maximum and minimum pixel values ​​of the original image are represented by . Furthermore, the original image data is resized to a fixed pixel size of 224×224, and data augmentation is performed through rotation, flipping, and cropping to generate more samples. These preprocessing operations better match the input of the network model, helping to better extract features and thus improve feature representation. This also further reduces the impact of noise in the screenshot image, making the model more reliable in practical applications.

[0096] Furthermore, the step of constructing an index of the screen shot image includes:

[0097] A dynamic quantization method based on Gaussian mixture model is used to construct the index corresponding to the screen shot image to reduce the quantization error of screen shot images with high similarity.

[0098] Specifically, the steps of constructing the index corresponding to the screen shot image using the dynamic quantization method based on the Gaussian mixture model include:

[0099] Divide the feature vector W into n subspaces and calculate the significance score of each subspace through the attention mechanism , according to the significance score Identify key areas (such as text input boxes and icons) in the screenshot image and assign weights to the subspace containing the key areas. The formula is:

[0100] ,

[0101] Where, represents the weight value of the i-th subspace;

[0102] A Gaussian mixture model (GMM) is trained for each subspace to obtain the quantization center of each subspace, and the mean vector of the Gaussian mixture model is used as the quantization center;

[0103] The eigenvector W is quantized using the mean vector of the Gaussian mixture model. The distance between the eigenvector and each quantization center is calculated, and the quantization center with the smallest distance is selected for quantization. Quantization refers to the process of mapping the originally continuous high-dimensional space vector into a discrete indexable dictionary, and finally representing the vector with one or a group of indices. Selecting the quantization center with the smallest distance is to select the one with the smallest error to generate the corresponding mapping.

[0104] Combine the quantization results of each subspace to generate the final quantized index , each index corresponds to the corresponding screen capture image data.

[0105] During the index construction process, product quantization is used as an efficient vector compression and retrieval method. During the quantization process, a probability-based Gaussian mixture (GMM) model is used instead of the K-means method to optimize the position of the quantization center and further reduce the quantization error.

[0106] The GMM model is constructed based on the expectation maximization algorithm. The parameter mean vector, covariance matrix and weight of each Gaussian component are initialized according to the following formula:

[0107] ,

[0108] Where, Represents the overall probability density function composed of the weighted sum of these K Gaussian components; Represents the mixing weight of k Gaussian components, that is, the contribution ratio of the component to the overall distribution; represents the mean vector of the kth Gaussian, which determines the center position; Represents the covariance matrix, which determines the shape, direction, and variance of the distribution; represents the probability density function of the kth Gaussian component;

[0109] Then repeat the following two steps until the GMM model parameters converge:

[0110] Step 201: Calculate the posterior probability of each data point belonging to each Gaussian component based on the current parameters.

[0111] Step 202: Based on the posterior probability calculated in step 201, update the mean, covariance and weight of each sub-model.

[0112] The uploading user uploads the quantized index through the smart contract. Each search by the user actually obtains the index value of the uploaded image through the product quantization method based on the Gaussian model for comparison.

[0113] Step 3: When the screen capture user visitor wants to obtain the corresponding screen capture image, a request is sent to the screen capture user uploader; if the screen capture user uploader agrees to share, a reply is sent to the screen capture user visitor through a secure channel with the corresponding image decryption key attached.

[0114] When user 2 wants to obtain the corresponding screenshot image data, it sends a request to user 1. If user 1 agrees to share, it replies to user 2 through a secure channel with the corresponding image decryption key.

[0115] Step 4: The screen capture user visitor uses keyword retrieval to obtain the required screen capture image data by searching in the distributed file system, and obtains the required screen capture image through the image decryption key.

[0116] Specifically, the steps of a screen capture user visitor using keyword retrieval to obtain the required screen capture image data by searching in a distributed file system include:

[0117] Step 401: Generate an initialization key:

[0118] The security parameter λ is used as input. The parameter λ determines the group order p and uses a bilinear mapping. , generate cyclic group G, randomly select ,in Is a cyclic group of order p, construct two hash functions , , get the global parameter set , randomly selected , output key pair , where the public key , private key ;

[0119] Step 402: Build an encrypted index:

[0120] Using public keys , randomly selected , for keywords And the quantized index set Encryption, computing , get the ciphertext , output the ciphertext c and encrypted index set corresponding to the searched keyword ;

[0121] Step 403: Generate a trapdoor:

[0122] Input query image feature vector and private key , output trapdoor ;

[0123] The eigenvector When a user wants to perform a search query, they upload the required sample image. The system also performs feature extraction and then uses the index constructed by the vector for retrieval.

[0124] Step 404: Search the encrypted index:

[0125] Input generated retrieval trapdoor and encrypted indexes Perform a match. If the match fails, an error symbol is returned. Otherwise, the addresses of the top-N pictures with the highest similarity are returned. ;

[0126] Step 405, update the encrypted index: Screen capture user uploader using the search trap Update image index data and use the improved product quantization method to construct the index of the newly added image data .

[0127] This invention combines blockchain with smart contracts to achieve secure storage, management, and sharing of screen capture image data. Dynamic encryption based on chaotic mapping destroys the statistical characteristics of screen capture images, ensuring secure image storage. A lightweight network model, combined with GMM quantization and searchable encryption, enables high-precision, secure, and private image retrieval. Furthermore, blockchain and IPFS collaborate to create a low-cost, highly scalable storage architecture. These three elements work together to form a complete closed-loop system for secure storage, management, and sharing of screen capture images, suitable for highly sensitive scenarios such as corporate documents, online education, and confidential meetings.

[0128] To further demonstrate the effectiveness and superiority of the method described in this paper, the gas consumption of each smart contract function throughout the entire process was calculated, as shown in Table 1. Table 2 compares the method described in this paper with other currently studied image storage management solutions and compares multiple security properties of these solutions.

[0129] Table 1 Consumption of each method of smart contract

[0130]

[0131] Table 2 Comparison of security attributes of various solutions

[0132]

[0133] Figure 5 Figures (a), (b), and (c) represent the comparison of the original image, the encrypted image, and the decrypted image, respectively, indicating that the screen capture encryption image algorithm of the present invention can correctly implement image encryption and decryption, and is an accuracy indicator in the image encryption evaluation index.

[0134] In addition, this example also tests the performance of the proposed image encryption algorithm through histogram analysis in statistical analysis. Figure 6 Figures (a), (b), and (c) in the figure represent the grayscale histogram comparisons of the original image, encrypted image, and decrypted image, respectively. The grayscale histogram of the original image shows significant statistical regularity and is vulnerable to statistical analysis attacks, while the grayscale histogram of the encrypted image after encryption is evenly distributed, showing a significant difference from the original image.

[0135] In addition, this example also tests the average precision (mAP) and top-N accuracy and compares them with the k-means-based product quantization method and the amplified local sensitive hashing method. Top-N refers to the proportion of relevant images in the top N results, that is, the proportion of images that are "related" to the query image among the top N results returned by the retrieval system. Figure 7 The average retrieval accuracy results of the three methods under different feature vector dimensions are shown. It can be seen that the higher the dimension of the image feature vector extracted by the present invention, the higher the mAP value will be and the higher the overall accuracy will be.

[0136] Figure 8 The results of retrieving the top-N similar images are shown. The accuracy of the improved Gaussian model-based product quantization method on the top-N is better than the original k-means-based product quantization method and existing similar schemes, which further shows that the retrieval scheme proposed in this example is feasible.

Claims

1. A blockchain-based screen capture image security storage management and sharing method, characterized by: The following steps are involved: Screen capture user uploaders and screen capture user visitors are registered in the blockchain network according to the provisions of a dedicated smart contract; The screen capture user uploader dynamically encrypts and extracts features from the screen capture image, builds an index of the screen capture image, and then uploads it to the distributed file system, while also building an index list in the distributed file system. When a screen capture user visitor wants to obtain the corresponding screen capture image, a request is sent to the screen capture user uploader; If the screen capture user uploader agrees to share, the screen capture user visitor will be replied through a secure channel with the corresponding image decryption key; Screen capture user visitors use keyword retrieval to obtain the required screen capture image data by searching in the distributed file system, and obtain the required screen capture image using the image decryption key; The steps for the screen capture user to upload the screen capture image to dynamically encrypt and extract features also include: Adjust the lightweight CNN model, and use the adjusted lightweight CNN model to perform feature extraction on the screen capture image to obtain a feature vector W of the screen capture image; The steps for constructing an index of screen capture images include: The index corresponding to the screen shot image is constructed based on the dynamic quantization method of Gaussian mixture model; The steps of constructing the index corresponding to the screen shot image using the dynamic quantization method based on the Gaussian mixture model include: Divide the feature vector W into n subspaces and calculate the significance score of each subspace in the attention map through the attention mechanism , according to the significance score Identify the key areas in the screenshot image and assign weights to the subspace containing the key areas. The formula is: ， Where, represents the weight value of the i-th subspace; Train a Gaussian mixture model for each subspace to obtain the quantization center of each subspace, and use the mean vector of the Gaussian mixture model as the quantization center; The eigenvector W is quantized using the mean vector of the Gaussian mixture model. The distance between the eigenvector and each quantization center is calculated, and the quantization center with the smallest distance is selected for quantization. Combine the quantization results of each subspace to generate the final quantized index , each index corresponds to the corresponding screen capture image data; The GMM model is constructed based on the expectation maximization algorithm, and the parameter mean vector, covariance matrix and weight of each Gaussian component are initialized according to the following formula: ， Where, Represents the overall probability density function composed of the weighted sum of these K Gaussian components; Represents the mixing weight of k Gaussian components, that is, the contribution ratio of the component to the overall distribution; represents the mean vector of the kth Gaussian, which determines the center position; Represents the covariance matrix, which determines the shape, direction, and variance of the distribution; represents the probability density function of the kth Gaussian component; Then repeat the following two steps until the GMM model parameters converge: Step 201: Calculate the posterior probability of each data point belonging to each Gaussian component based on the current parameters; Step 202: Based on the posterior probability calculated in step 201, update the mean, covariance and weight of each sub-model.

2. The blockchain-based screen capture image security storage management and sharing method according to claim 1, characterized in that: The steps for the screen capture user to dynamically encrypt and extract features of the screen capture image include: A screen capture image encryption algorithm based on hybrid chaotic mapping and dynamic S-box is used to dynamically encrypt screen capture images.

3. The method for secure storage, management and sharing of screen capture images based on blockchain according to claim 2, characterized in that: The screen capture image encryption algorithm based on hybrid chaotic mapping and dynamic S-box includes the following steps for dynamically encrypting the screen capture image: Use Logistic mapping to diffuse the pixels of the screen shot image; The Henon map is used to dynamically construct the S-box and perform scrambling operations on the pixel positions of the screen shot image.

4. The blockchain-based screen capture image security storage management and sharing method according to claim 3 is characterized in that: The steps of using Logistic mapping to diffuse the pixels of the screen shot image include: Using the initial value of the given Logistic map and control parameters Perform loop iteration to generate chaotic values. The formula is: ， Where, Indicates the The chaotic value generated by the iteration, the control parameter The range is (0,4); Using chaotic values ​​to form chaotic sequences , transform the chaotic sequence s and the image to be encrypted into a one-dimensional sequence Perform XOR operation to obtain a one-dimensional sequence , the formula is: ， Where, Represents the exclusive OR operation.

5. The method for secure storage, management and sharing of screen capture images based on blockchain according to claim 4, characterized in that: The steps of dynamically constructing an S-box using the Henon map and scrambling the pixel positions of the screen shot image include: The Logistic map generates two chaotic sequences, which are used to construct the parameters in the Henon map. and , the formula is: ， ， Where, and represents the initial value of state n in the Henon map; and Indicates the result of the operation in the next state n+1 after one operation in state n; A random 16×16 matrix consisting of integers between 0 and 255 is constructed without duplication. The pixels in the matrix are transformed using the Henon map. The S-box is constructed after iteration using the Henon map formula. The one-dimensional sequence Restore it to an image, divide the image into four blocks, and use different S-boxes to scramble the pixel positions of each block of the image to obtain the encrypted screen shot image.

6. The blockchain-based screen capture image security storage management and sharing method according to claim 1, characterized in that: Screen capture user visitors use keyword retrieval to obtain the required screen capture image data by searching in the distributed file system. The steps include: Step 401: Generate an initialization key: The security parameter λ is used as input. The parameter λ determines the group order p and uses a bilinear mapping. , generate cyclic group G, randomly select ,in Is a cyclic group of order p, construct two hash functions , , get the global parameter set , randomly selected , output key pair , where the public key , private key ; Step 402: Build an encrypted index: Using public keys , randomly selected , for keywords And the quantized index set Encryption, computing , get the ciphertext , output the ciphertext c and encrypted index set corresponding to the searched keyword ; Step 403: Generate a trapdoor: Input query image feature vector and private key , output trapdoor ; Step 404: Search the encrypted index: Input generated retrieval trapdoor and encrypted indexes Perform matching. If the match fails, an error symbol is returned. Otherwise, the addresses of the top-N images with the highest similarity are returned. ; Step 405, update the encrypted index: Screen capture user uploader using the search trap Update image index data and use the improved product quantization method to construct the index of the newly added image data .

Images