Cloud storage security image similar deduplication method and system, medium, equipment and product

Abstract

The invention provides a similar de-duplication method, system, medium, device and product for a cloud storage security image, and relates to the technical field of cloud storage, the method comprises the following steps: carrying out semantic segmentation on an uploaded image, and segmenting the image into image blocks with different semantics; extracting feature values of the image blocks through a local feature extraction algorithm, and encrypting the feature values; in the cloud encryption environment, comparing the encrypted feature value with the encrypted feature value stored in the cloud to obtain a similar image block, and performing image similar de-duplication on the uploaded image; extracting position information from the similar image blocks, encrypting the position information, and rearranging the similar images by the cloud through matching of the encrypted position information; and encrypting the rearranged image, performing difference calculation in an encryption domain, generating difference data in an encryption form, compressing the difference data, and storing the compressed difference data in a cloud end. The utilization efficiency of cloud storage is improved while duplicate removal is performed, and privacy protection of image data in a cloud environment is realized.

Description

Technical Field

[0001] The present invention relates to the field of cloud storage technology, and in particular to a method, system, medium, device, and product for deduplication of secure cloud storage images. Background Art

[0002] With the widespread adoption of smart devices and the rapid development of cloud storage technology, a vast amount of everyday image data is being uploaded to the cloud. However, due to the similarity of shooting scenes and subtle variations in shooting angle, lighting, and other factors, a large amount of redundant image data, similar but not identical, exists in cloud storage. This redundant storage of similar images consumes significant resources and reduces cloud storage efficiency. Therefore, deduplication technology plays a crucial role in image data management, aiming to improve storage utilization and data management efficiency by reducing redundant image data.

[0003] Currently, common image deduplication methods include the following:

[0004] File-level deduplication: Traditional file-level deduplication methods typically identify and remove exact duplicates by calculating a cryptographic hash (such as SHA-1 or MD5) on image files. This method only identifies identical image files and is ineffective against similar images with subtle variations, making it difficult to further reduce storage space usage.

[0005] Deduplication methods based on visual similarity: Visual similarity deduplication techniques detect visually similar but not identical images by generating visual hash tags and compressing similar images through incremental compression. However, this method relies on visual consistency rather than byte-level consistency, which can lead to loss of detail when the compressed image is restored, making it a lossy deduplication method. Furthermore, visual similarity deduplication techniques often lack adequate privacy protection mechanisms, posing a risk of user image data leakage.

[0006] Deduplication methods based on convergent encryption: To improve data privacy, some studies have proposed deduplication methods based on convergent encryption (CE). These methods create CE keys using a hash generated from the image content. However, CE methods are very sensitive to small changes in image content. Slight differences in the pixel level can result in different ciphertexts, reducing deduplication effectiveness and making them vulnerable to security threats such as brute force cracking and side-channel attacks.

[0007] Although existing image deduplication methods have achieved certain results in improving storage efficiency and saving storage space, they still face the following drawbacks:

[0008] Insufficient utilization of local features: Most deduplication methods only target plaintext images and fail to fully utilize the local feature information of the image. The deduplication effect is rough, resulting in insufficient fine-grained deduplication performance.

[0009] Insufficient security: Traditional encryption technology has a certain effect in improving data security, but it is still vulnerable to brute force attacks, side channel attacks, and other attacks, and there is a risk of privacy leakage.

[0010] Position offset problem: In similar images, there may be position offsets between different similar blocks. This offset affects the precise matching of similar blocks, thereby reducing the accuracy and efficiency of deduplication compression. Summary of the Invention

[0011] The purpose of the present invention is to solve the problem of insufficient image matching accuracy due to position offset in image cloud storage, and to propose a cloud storage secure image similarity deduplication method, comprising the following steps:

[0012] S1. Perform semantic segmentation on the uploaded image and divide the image into image blocks with different semantics;

[0013] S2, extracting the feature value of the image block through a local feature extraction algorithm and encrypting it;

[0014] S3. In the cloud encryption environment, the encrypted feature value is compared with the encrypted feature value stored in the cloud to obtain similar image blocks, and the uploaded image is deduplicated based on image similarity.

[0015] S4. Extract location information from similar image blocks and encrypt it. The cloud then rearranges similar images by matching the encrypted location information.

[0016] S5. Encrypt the rearranged image, perform difference calculation on similar image blocks in the encryption domain, generate encrypted difference data, compress the difference data, and save the compressed difference data to the cloud.

[0017] Furthermore, the U-Net model and the improved GrabCut algorithm are used to achieve hybrid semantic segmentation of images;

[0018] The specific steps of the improved GrabCut algorithm are:

[0019] (1) Preprocess the original image;

[0020] (2) Based on the preprocessed image, a binary image and an edge image are generated through adaptive thresholding and edge detection; and the foreground boundary is optimized through contour extraction;

[0021] (3) Initialize the foreground and background models and mark potential background pixels. The background area is further refined by connecting pixels based on color similarity. The segmentation results are iteratively updated until convergence to obtain the segmentation mask.

[0022] (4) Apply the segmentation mask to the original image to extract the foreground part.

[0023] Furthermore, local binary patterns are used to extract features from the segmented image blocks.

[0024] Furthermore, the CKKS homomorphic encryption scheme is used to encrypt the eigenvalues ​​of the image blocks.

[0025] Furthermore, the Umbral proxy re-encryption scheme is used to re-encrypt the ciphertext data.

[0026] Furthermore, random perturbation and Mask strategies are used to process the eigenvalues ​​of image blocks.

[0027] The present invention also proposes a cloud storage secure image similarity deduplication system, comprising:

[0028] Semantic segmentation module, used to perform semantic segmentation on the uploaded image and divide the image into image blocks with different semantics;

[0029] A feature extraction module is used to extract the feature value of the image block through a local feature extraction algorithm and encrypt it;

[0030] The similarity detection module is used to perform similarity detection on the encrypted feature values ​​​​with the encrypted features stored in the cloud in an encrypted environment, obtain similar image blocks, and perform image similarity deduplication on the uploaded images;

[0031] The rearrangement module is used to extract and encrypt position information from similar image blocks. The cloud then rearranges similar images by matching the encrypted position information.

[0032] The difference compression module is used to encrypt the rearranged image, perform difference calculations on similar image blocks in the encryption domain, generate encrypted difference data, compress the difference data, and save the compressed difference data to the cloud.

[0033] The present invention also proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned cloud storage security image similarity deduplication method.

[0034] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned cloud storage security image similarity deduplication method.

[0035] The present invention also proposes a computer program product, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the above-mentioned cloud storage security image similarity deduplication method are implemented.

[0036] The beneficial effects brought about by the technical solution provided by the present invention are:

[0037] The present invention first performs semantic segmentation on the uploaded image, extracts the feature values ​​of the segmented image blocks, and encrypts them; in the cloud encryption environment, the encrypted feature values ​​are compared for similarity with the encrypted features stored in the cloud, and the uploaded image is deduplicated for image similarity. Through image semantic segmentation, the detailed differences of similar images can be better identified, greatly improving the accuracy of similarity detection. Position information is extracted from similar image blocks and encrypted, and the cloud rearranges similar images by matching the encrypted position information; the rearranged images are encrypted, and the difference calculation is performed in the encrypted domain to generate encrypted difference data, which is then compressed. This reduces data storage space while deduplicating and improves the utilization efficiency of cloud storage. The use of multiple encryption mechanisms enables privacy protection of image data in a cloud environment. The present invention achieves more sophisticated image similarity detection and deduplication through the combination of semantic segmentation, local feature extraction, and homomorphic encryption technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for deduplication of secure cloud storage images according to an embodiment of the present invention;

[0039] Figure 2 This is a bar chart comparing the compression ratio of the method of the present invention and the compression ratio of the traditional method (differential compression of similar images under plain text);

[0040] Figure 3 This is a comparison chart of compression rates of various data sets at different similarity thresholds according to an embodiment of the present invention;

[0041] Figure 4 : is a bar chart showing the changes in detection success rates of various data sets under different similarity thresholds according to an embodiment of the present invention. Figure 4 (a) is a comparison of the SIM-1 and SIM-2 data sets. Figure 4 (b) is a comparison of the SIM-3 and SIM-4 data sets;

[0042] Figure 5 FIG. 1 is a schematic diagram showing the effect of the number of cluster centers on the detection success rate of each data set in one embodiment of the present invention. Figure 5 (a) is a comparison of the SIM-1 and SIM-2 data sets. Figure 5 (b) is a comparison of the SIM-3 and SIM-4 data sets;

[0043] Figure 6 It is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0045] The flowchart of the cloud storage security image similarity deduplication method according to the embodiment of the present invention is as follows: Figure 1 , specifically including the following steps:

[0046] S1. Perform semantic segmentation on the uploaded image to divide the image into image blocks with different semantics. The present invention adopts a hybrid semantic segmentation method, using different segmentation algorithms according to the complexity of the image. For complex images, a U-Net model is used for refined segmentation to ensure high-precision separation between foreground and background. For simpler images (such as vector images or cartoon images), an improved GrabCut algorithm is used to perform rapid segmentation based on the automatic selection of the largest bounding box area. The image is divided into different semantic regions.

[0047] The specific steps of the improved GrabCut algorithm are:

[0048] (1) First, preprocess the image to facilitate the extraction of foreground and edge features;

[0049] (2) Then, a binary image and an edge image are generated by adaptive thresholding and edge detection; the foreground boundary is optimized by contour extraction;

[0050] (3) Initialize the foreground and background models and mark potential background pixels. The background region is further refined by connecting pixels based on color similarity. Iteratively update the segmentation results until convergence.

[0051] (4) Apply the segmentation mask to the original image to extract the foreground part.

[0052] By introducing edge detection and color-based refinement, this method significantly improves the foreground segmentation accuracy of simple images without the need for traditional GrabCut manual operations.

[0053] Post-segmentation refinement: To address the issues of blurred and overlapping boundaries in traditional foreground-background segmentation models, a color similarity-based analysis method is proposed. By analyzing similarities pixel by pixel, the foreground image is further refined to ensure that the semantic blocks after segmentation do not overlap, achieving a refined segmentation effect.

[0054] S2. Extract the feature value of the image block through a local feature extraction algorithm and encrypt it.

[0055] We use rotationally invariant local binary patterns (LBP) to extract features from image blocks and add small random perturbations to prevent side-channel attacks. These features preserve the image's local texture information, providing an accurate basis for similarity detection. The extracted features are encrypted using the CKKS homomorphic encryption scheme. While ensuring security, the system enables data matching without decrypting the data, preventing plaintext exposure. Users then upload these encrypted features to the cloud.

[0056] A small random perturbation is added to the feature. The feature vector is processed using random perturbation and Mask strategy. By randomly masking 20% ​​of the input features, the robustness of feature encryption is further enhanced to prevent the risk of side channel attacks.

[0057] S3. In the cloud encryption environment, the encrypted feature value is compared with the encrypted feature value stored in the cloud to obtain similar image blocks, and the uploaded image is deduplicated based on image similarity.

[0058] Specifically, the cosine distance calculation is performed with each cluster center image on the cloud to obtain the clusters where similar image blocks are located.

[0059]

[0060] Where cosineDistance represents the cosine distance, enclbp and enclbp′ are the encrypted LBP feature vectors of the image block (under CKKS encryption), · represents the dot product, and |·| represents the vector modulus (in the encrypted domain).

[0061] Larger cosine distance values ​​indicate greater similarity, ultimately used to match the most similar image clusters. This calculates the encrypted feature similarity between user-uploaded images and existing image cluster centers. This enables image matching and retrieval in encrypted form, minimizing the risk of decryption leaks.

[0062] Then, after the most similar cluster is found through the above calculation, the most similar image block is found through Chi-Square Similarity calculation.

[0063]

[0064] Where k represents the number of image blocks. i with enclbp i ′ They represent the encrypted features of the i-th semantic block of the two most similar image blocks.

[0065] The smaller the difference between samples, the lower the chi-square value, indicating a higher similarity.

[0066] In the cloud, the system uses encrypted features to perform similarity matching with encrypted cluster centers and calculates the distance between encrypted features using the Euclidean distance formula. This method effectively detects similar image blocks without exposing plaintext data, ensuring privacy protection during the detection process.

[0067] The encrypted cluster center is an "encrypted center vector" automatically constructed and maintained in the cloud based on encrypted image features. It is not the center of the original image and does not contain decryptable image data. It represents the encrypted semantic feature center of a cluster of images and is used to determine image similarity in ciphertext, thereby achieving secure deduplication. The cloud stores these encrypted features and uses methods such as KNN nearest neighbor search and cosine similarity in the encrypted domain to identify the most similar images. Each cluster of similar images is represented by an "encrypted center vector," and these "centers" are the encrypted cluster centers. Cluster centers are generated by averaging or updating multiple encrypted features (under homomorphic encryption) and remain encrypted, making them indecryptable by the cloud service provider. User-uploaded encrypted features are compared against existing encrypted cluster centers. If a match is found, the image is assigned to that encrypted cluster, triggering image rescheduling and differential compression. Otherwise, a new encrypted cluster center is created.

[0068] The entire similarity detection process is carried out entirely in the encrypted domain, avoiding the plaintext exposure of image features, thereby effectively protecting the privacy and security of user data.

[0069] S4. Extract location information from similar image blocks and encrypt it. The cloud then rearranges similar images by matching the encrypted location information.

[0070] To improve the efficiency of differential compression, the system extracts location information from similar image blocks, including the coordinates of the upper left corner, width, height, and area, and encrypts and uploads it to the cloud for comparison. This location information can be expressed as: Mapinf = [x, y, w, h, area], where Mapinf represents the spatial position and size of the semantic block in the image, facilitating realignment and differential alignment; x, y represent the pixel coordinates of the semantic block's upper left corner; w, h represent the width and height of the semantic block; and area represents the total number of pixels (area size) contained in the semantic block. This location information is used for position comparison within the ciphertext domain and assists in differential alignment compression.

[0071] The cloud reorders similar images by matching encrypted location information. This data reordering eliminates differences caused by positional offsets, enabling more accurate incremental compression while maintaining encryption. This data reordering and encryption ensures data security and compression accuracy.

[0072] S5. Encrypt the rearranged image. Compare similar image blocks in the encrypted domain to generate encrypted difference data. This difference data is compressed and stored in the cloud. The client only stores the index pointing to the similar image block stored in the cloud. Thus, for similar cloud image block A and client image block B, the cloud stores: a fully encrypted version of image A; the difference between image B and image A, including the alignment position (this position information is also encrypted); and the encrypted difference data. This reduces storage space.

[0073] After similar image blocks are rearranged using encrypted position information, the rearranged image is encrypted using the BFV homomorphic encryption scheme. Difference calculations (such as pixel subtraction) are performed in the encrypted domain to generate encrypted difference data. This difference data is then compressed using run-length encoding (RLE), storing only the unique image and difference data, achieving lossless difference compression.

[0074] Based on the compressed feature matrix of similar images and delta compression technology, difference blocks (such as the difference vector [0,0,0,5,5,0,0]) are compressed to reduce transmission and storage. Run-Length Encoding (RLE) is used to compress the structure of these "ciphertext difference vectors." For example: [0,0,0,5,5,0,0] → [(0,3),(5,2),(0,2)]. This eliminates duplication while reducing data storage space and improving cloud storage utilization efficiency.

[0075] Experiments were conducted on multiple image datasets to evaluate the compression ratio, detection accuracy, and processing speed of the proposed method. The results demonstrate that the proposed method significantly improves similarity detection accuracy and compression efficiency while protecting data privacy. The datasets used in this paper include: SIM-1, SIM-2, SIM-3, SIM-4, and SIM-5.

[0076] SIM-1 dataset

[0077] Source: The Simpsons faces dataset (from Kaggle); Content: 98,770 original images; Construction: 19,754 original images were selected and four similar versions were edited (by rotation, cropping, and color adjustment).

[0078] SIM-2 dataset

[0079] Source: Synthesized by SIM-1 using a generative adversarial network (GAN); Content: 24,754 images; Features: Derived from SIM-1 using deep learning, simulating realistic but similar images (synthetic variants) for testing robustness in synthetic scenarios.

[0080] SIM-3 dataset

[0081] Source: CIFAR-10 dataset; Content: 300,000 images (based on a 60,000-image extension of CIFAR-10); Construction: Four lightly edited versions of each image (such as rotation, cropping, and lighting adjustment) are produced; Features: Low resolution (32×32), multi-category (10 categories), subtle but representative variations, used to test the system's performance on low-resolution images.

[0082] SIM-4 dataset

[0083] Source: 70,000 images (source unspecified, but includes GAN-generated images); Features: Contains 1,000 GAN-generated images, a mixture of real and synthetic images, to test the system's performance and generalization ability under mixed real and synthetic conditions.

[0084] SIM-5 dataset

[0085] Source: PASCAL VOC 2012 dataset (a classic dataset for visual object detection); Content: 10,000 randomly selected images, with four versions of each image containing a random logo added; Features: The images are highly complex, and the addition of logos simulates the "watermarking" or "stickering" scenarios used in social media, testing the system's performance when dealing with occlusion or local changes.

[0086] in, Figure 2 It is a bar chart comparing the compression ratio of the method of the present invention and the compression ratio of the traditional method (differential compression of similar images under plain text). The incremental compression ratio of the method of the present invention is significantly higher than that of the traditional method.

[0087] Figure 3This is a comparison chart of the compression rates of various data sets under different similarity thresholds in an embodiment of the present invention. 0.5, 0.6, 0.7, and 0.8 represent the thresholds for image similarity determination, respectively. This threshold controls the system's criteria for determining whether "the images are similar enough for differential compression." A threshold of 0.5: the most relaxed, as long as the similarity reaches 50%, they are considered compressible; a threshold of 0.8: the most stringent, compression is performed only when the two images reach 80% similarity. Lower thresholds allow the system to identify more "partially similar" images, resulting in higher compression rates; higher thresholds are more conservative, but reduce false recognition. When the threshold is 0.5 or 0.6, the compression rate of all data sets is significantly higher than 1 (approximately 1.8 to 1.9), indicating a significant compression effect. The higher the threshold, the lower the compression rate because the number of pairs identified as "similar images" by the system decreases. This shows that the method of the present invention can significantly improve compression performance at a looser threshold, especially on GAN-generated or structurally complex images such as SIM-2 and SIM-4. Even at the strictest threshold of 0.8, it still maintains a compression ratio of over 1.2, which is better than traditional non-encrypted deduplication solutions.

[0088] Figure 4 is a bar chart showing the change in detection success rate under different similarity thresholds according to an embodiment of the present invention. Figure 4 (a) is a comparison of the SIM-1 and SIM-2 datasets. SIM-2 is slightly better than SIM-1, indicating that the similar structure of the synthetic cartoon images is more obvious and easier to detect. Figure 4 Figure (b) compares the SIM-3 and SIM-4 datasets. SIM-4 (including the GAN image) achieves a higher detection rate, likely because the GAN-generated similarity images retain more semantic block structure. The figure shows that as the similarity threshold decreases (becomes more relaxed), the system's detection success rate continues to improve. This demonstrates that our method has strong recognition capabilities for "partially similar" images and can maintain high detection performance at lower thresholds.

[0089] Figure 5 Schematic diagram of the effect of the number of cluster centers on the detection success rate in an embodiment of the present invention. Figure 5 (a) is a comparison of the SIM-1 and SIM-2 data sets. Figure 5 (b) shows a comparison of the SIM-3 and SIM-4 datasets. The more cluster centers in the cloud-stored images, the more detailed the feature grouping, which improves matching accuracy. The proposed method employs a dynamic clustering update strategy in the encrypted feature space, enabling it to maintain high-precision matching capabilities across large image repositories.

[0090] The present invention also proposes a cloud storage secure image similarity deduplication system, comprising:

[0091] Semantic segmentation module, used to perform semantic segmentation on the uploaded image and divide the image into image blocks with different semantics;

[0092] A feature extraction module is used to extract the feature value of the image block through a local feature extraction algorithm and encrypt it;

[0093] The similarity detection module is used to perform similarity detection on the encrypted feature values ​​​​with the encrypted features stored in the cloud in an encrypted environment, obtain similar image blocks, and perform image similarity deduplication on the uploaded images;

[0094] The rearrangement module is used to extract and encrypt position information from similar image blocks. The cloud then rearranges similar images by matching the encrypted position information.

[0095] The difference compression module is used to encrypt the rearranged image, perform difference calculations on similar image blocks in the encryption domain, generate encrypted difference data, compress the difference data, and save the compressed difference data to the cloud.

[0096] In an exemplary embodiment, a computer-readable storage medium is included, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned cloud storage secure image similarity deduplication method is implemented.

[0097] See also Figure 6 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0098] A computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through a communication bus to execute the above-mentioned cloud storage security image similarity deduplication method.

[0099] In an exemplary embodiment, a computer program product is also included, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the above-mentioned cloud storage security image similarity deduplication method are implemented.

[0100] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cloud storage security image similarity deduplication method, characterized in that: The following steps are involved: S1. Perform semantic segmentation on the uploaded image and divide the image into image blocks with different semantics; S2, extracting the feature value of the image block through a local feature extraction algorithm and encrypting it; S3. In the cloud encryption environment, the encrypted feature value is compared with the encrypted feature value stored in the cloud to obtain similar image blocks, and the uploaded image is deduplicated based on image similarity. S4. Extract location information from similar image blocks and encrypt it. The cloud then rearranges similar images by matching the encrypted location information. S5. Encrypt the rearranged image, perform difference calculation on similar image blocks in the encryption domain, generate encrypted difference data, compress the difference data, and save the compressed difference data to the cloud.

2. A cloud storage secure image similarity deduplication method according to claim 1, characterized in that: Adopting the U-Net model and the improved GrabCut algorithm to achieve hybrid semantic segmentation of images; The specific steps of the improved GrabCut algorithm are: (1) Preprocess the original image; (2) Based on the preprocessed image, a binary image and an edge image are generated through adaptive thresholding and edge detection; and the foreground boundary is optimized through contour extraction; (3) Initialize the foreground and background models and mark potential background pixels. The background area is further refined by connecting pixels based on color similarity. The segmentation results are iteratively updated until convergence to obtain the segmentation mask. (4) Apply the segmentation mask to the original image to extract the foreground part.

3. A cloud storage secure image similarity deduplication method according to claim 1, characterized in that: Local binary patterns are used to extract features from the segmented image blocks.

4. A cloud storage secure image similarity deduplication method according to claim 1, characterized in that: The CKKS homomorphic encryption scheme is used to encrypt the feature values ​​of the image blocks.

5. The method for deduplication of cloud storage secure images according to claim 1, characterized in that: The Umbral proxy re-encryption scheme is used to re-encrypt the ciphertext data.

6. A cloud storage secure image similarity deduplication method according to claim 1, characterized in that: Random perturbation and Mask strategy are used to process the eigenvalues ​​of image blocks.

7. A cloud storage security image similarity deduplication system, characterized by: include: Semantic segmentation module, used to perform semantic segmentation on the uploaded image and divide the image into image blocks with different semantics; A feature extraction module is used to extract the feature value of the image block through a local feature extraction algorithm and encrypt it; The similarity detection module is used to perform similarity detection on the encrypted feature values ​​​​with the encrypted features stored in the cloud in an encrypted environment, obtain similar image blocks, and perform image similarity deduplication on the uploaded images; The rearrangement module is used to extract and encrypt position information from similar image blocks. The cloud then rearranges similar images by matching the encrypted position information. The difference compression module is used to encrypt the rearranged image, perform difference calculations on similar image blocks in the encryption domain, generate encrypted difference data, compress the difference data, and save the compressed difference data to the cloud.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images