An authenticatable client-side watermarking method based on thumbnail encrypted images

Abstract

The application discloses a kind of based on thumbnail encryption image's authenticatable user end watermark method, belong to computer security technical field. Including: image pre-processing, image encryption, key generation and distribution, decryption and watermark embedding, tamper detection and positioning, tamper content recovery, extract watermark lock user identity.The present application is based on the holographic redundancy embedding of thumbnail and the fine authentication based on parity check, with very strong anti-shearing and self-healing ability;Using the disturbance of pair of pixel difference value, all image pixels can be embedded watermark, and the security blind area is eliminated;Encrypted image has clear identifiable thumbnail, and is highly similar to plaintext, which is convenient for cloud management, and the details privacy is effectively protected through high-frequency information confusion;Encryption and watermark embedding are fused in bottom algebra operation, and the calculation efficiency is high.

Description

Technical Field

[0001] This invention relates to the field of computer security technology, and in particular to an authenticable client-side watermarking method based on thumbnail-encrypted images. Background Technology

[0002] With the rapid development of digital imaging technology and the mobile internet, the volume of global image data is exploding. It is reported that by 2025, the global photo capture volume will exceed 1.8 trillion to 2.1 trillion, encompassing sensitive data including biometrics and personal portraits. While local storage inherently offers privacy advantages, limitations in storage capacity and cross-device accessibility have driven the rapid adoption of cloud storage services. However, this convenience also brings multiple security challenges: on the one hand, image data involves personal privacy and trade secrets, requiring encryption to ensure privacy; on the other hand, users need to preview image content to improve usability, but encryption often results in completely blurred images, weakening practicality. Simultaneously, the lossless copying nature of digital images makes them vulnerable to illegal secondary distribution, directly threatening the rights of copyright holders; more insidiously, authorized users may maliciously tamper with decrypted images, eroding the authenticity and integrity of the content. To address these challenges, researchers have proposed various encryption and authentication algorithms to ensure the confidentiality and authenticity of data during transmission and storage, as well as transform-domain-based digital watermarking schemes to protect the integrity of sensitive data such as medical images. Therefore, how to balance privacy, usability, copyright protection, and content integrity in the context of big data in images has become a core challenge for both academia and industry.

[0003] To address the conflict between privacy and usability, Thumbnail-Preserving Encryption (TPE) technology has emerged. TPE allows users to securely preview content by preserving a low-resolution thumbnail of the image after encryption, thus improving user experience while protecting privacy. Since Wright et al. first proposed the TPE concept, subsequent research, such as Tajik et al.'s ideal TPE scheme and Zhao et al.'s multi-pixel TPE scheme, has continuously optimized encryption effectiveness and computational efficiency. However, traditional TPE schemes primarily focus on encryption itself, neglecting copyright protection and content integrity mechanisms, making them ill-equipped to handle the risks of infringement and tampering after distribution.

[0004] In copyright protection, digital watermarking technology is widely used for infringement tracking. Early watermarking schemes often employed an owner-embedded watermark model, such as... Figure 1As shown in (a), the owner pre-generates watermarked image copies for each user. While this approach is robust, it incurs significant computational and storage overhead for the owner when the user base is large, resulting in poor scalability. To address this issue, client-side embedding technology significantly improves system scalability by shifting the watermark embedding process to the user end, such as... Figure 1 As shown in (b). This type of scheme combines cryptographic tools to implicitly embed an identity watermark during decryption, achieving efficient infringement tracing. However, traditional client-side watermarking schemes generally neglect user experience; encrypted images cannot be previewed, and the watermark only serves as a source identifier, unrelated to the integrity of the image content, causing security protection to terminate after decryption.

[0005] In recent years, researchers have attempted to integrate TPE (Transparent Image Processing) with client-side watermarking to simultaneously achieve secure previewing and infringement tracking. For example, Xiao et al. implemented preview functionality and watermark embedding by dividing image blocks into "adjustment areas" and "hidden areas." However, this type of solution has inherent flaws: the "hidden areas," reserved for temporarily storing and restoring information, cannot embed watermarks, creating a blind spot in copyright protection; more importantly, its core design remains limited to infringement tracking, lacking protection for content integrity. Once authorized users obtain clear images, they can maliciously tamper with them locally and then redistribute them. The social harm of such behavior far exceeds copyright infringement, directly impacting the credibility of digital content. Fundamentally, existing solutions do not bind watermarks to content integrity, failing to achieve a secure closed loop throughout the entire lifecycle from publication to use. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention employs a verifiable client-side watermarking framework that covers the entire lifecycle. While maintaining the visual characteristics of encrypted image thumbnails, it utilizes a global holographic redundancy mechanism to achieve highly robust self-repair and eliminates protection blind spots through paired pixel perturbation. This aims to solve the problems in existing technologies that cannot simultaneously achieve "high-quality preview of encrypted images," "full-image watermark protection without blind spots," and "blind recovery under large-scale tampering."

[0007] To achieve the above objectives, the present invention provides an authenticated user-side watermarking method based on thumbnail-encrypted images, comprising the following steps: (1) Image preprocessing: The image owner preprocesses the original image, including thumbnail-based holographic redundancy embedding and parity-based fine-grained authentication. The thumbnail-based holographic redundancy embedding involves generating a thumbnail, quantizing it to obtain a binary recovery bitstream, and repeating it to uniformly fill the entire LSB plane of the original image. The refined authentication based on parity check divides the original image after thumbnail-based holographic redundancy embedding into... Blocks, and then subdivide each block into Perform parity checking on the sub-blocks; (2) Image encryption: The image owner divides the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image. (3) Key generation and distribution: The image owner generates a watermark lookup table containing user information for each authorized user, and generates a decryption lookup table by combining the encryption lookup table with the main encryption key, and distributes it to the authorized user as a decryption key. (4) Decryption and watermark embedding: Authorized users decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding of the authenticated user terminal. (5) Tamper detection and location: For the image to be detected, the parity is checked to determine the tamper location based on the refined authentication based on parity check. (6) Restoration of tampered content: The arbitrator reconstructs the thumbnail from the untampered location to complete the image restoration; (7) Extract watermark to lock user identity. By comparing the recovered image with the original preprocessed image, extract the differential signal, solve the watermark sequence and lock the user identity.

[0008] Furthermore, the thumbnail-based holographic redundancy embedding includes: (1.1) Generate a fixed-size global thumbnail from the original image using a bicubic interpolation algorithm; (1.2) Quantize the global thumbnail and serialize the quantized thumbnail matrix into a binary recovered bit stream; (1.3) The ratio of the total capacity of the LSB plane of the original image to the length of the binary recovery bitstream is used to obtain the maximum number of redundant copies. The binary recovery bitstream is repeated a maximum number of redundant copies to generate the holographic recovery stream, and it is evenly spread across the LSB plane of the entire image. in: The total capacity of the LSB plane of the original image. To recover the bitstream length from binary, The maximum number of redundant replicas, The length and width of the original image; The refined authentication based on parity check includes: calculating the sum of the LSBs of all pixels in each sub-block, and flipping the last pixel to make the sum satisfy a predetermined even parity rule.

[0009] Furthermore, step (2) specifically includes: (2.1) Divide the preprocessed original image into color channels respectively. The image owner generates a block of length locally. Encrypted lookup table Its elements are 0 or 1; (2.2) A secure sequence is generated for the current block using a cryptographically secure pseudo-random number generator based on the standard cryptographic hash function SHA-256; (2.2.1) Generate the pseudo-random sequence seed of the current block based on the master encryption key, the current encryption round, and the sum of the pixels of the previous block. For the first block, the sum of the pixels of the previous block is the sum of the initial pixels of the entire channel. (2.2.2) The pseudo-random sequence seed of the current block is concatenated with a counter that increments from 0 and used as the input of the SHA-256 hash function for each round. Multiple rounds of hashing are performed, and the outputs of each round are concatenated until the total length required for encrypting the current image block is reached. This is then used as the secure sequence of the current block. (2.3) Perform pairwise pixel interpolation perturbation on the current block, including: (2.3.1) Pair the pixels in the current block in order, calculate the pixel sum and pixel difference of the two pixel values ​​of each pair of pixels, and dynamically calculate an encrypted space for the difference of each pixel pair while ensuring that the pixel sum remains unchanged, so as to ensure that the two newly generated pixel values ​​of each pixel pair can strictly fall within the valid range of 0 to 255, and at the same time ensure that they are integers. (2.3.2) For each pair of pixels, locate the index of the original value of the pixel difference between the two pixel values ​​in the encrypted space, and use the encrypted lookup table. The security sequence of this block is used to deeply scramble the original index and generate a new encrypted index. in: It is the first The index of the original value of the pixel difference between two pixel values ​​in the encrypted space. It is the first The new encrypted index in the encrypted space after depth scrambling of each pixel pair. It is the size of the encrypted space. It is the first The index of a pixel pair in an image patch. This is the safe sequence for this block; (2.3.3) For each pair of pixels, according to the new encryption index Select a new difference from the encrypted space. The new difference is algebraically combined with the pixel sum to reconstruct each pixel pair of two encrypted pixel values ​​that are completely different from the original values ​​but whose algebraic relationship is controlled. (2.4) Repeat steps (2.2)-(2.3) until the pairwise pixel difference perturbation of all blocks is completed; (2.5) For each block after pairwise pixel difference perturbation, a global permutation sequence is generated using the master encryption key as a random number seed. The scrambling operation is performed on each pixel block once according to the global permutation sequence to rearrange its spatial position. (2.6) Repeat steps (2.2)-(2.5) until the predetermined encryption round is reached.

[0010] Furthermore, the upper and lower boundaries of the encrypted space are: in: The sum of the pixel values ​​of the two pixels is... Under the constraints, the effective minimum value that the difference can take is... Let S be the maximum effective value that the difference can take under the constraint that the sum is S; To ensure that the reconstructed pixel values ​​are all integers, the parity of the difference must be consistent with the parity of the sum of the two pixel values.

[0011] Furthermore, step (3) specifically involves: (3.1) The owner is each authorized user Allocate a length of Unique binary sequence As its identity identifier; (3.2) Adopt Encoding the original identity sequence Process to generate a binary sequence ; (3.3) The owner generates an encoding matrix locally. One-dimensional codewords are transformed through matrix operations. Mapped and expanded into a two-dimensional watermark lookup table uniquely bound to the user's identity. ; (3.4) Use the watermark lookup table and the encryption lookup table used during encryption The data is then merged to generate a personalized decryption lookup table for the user. ; (3.5) The owner will give the key pair Distributed to authorized users via secure channel .

[0012] Furthermore, step (4) specifically involves: according to the encryption process in step (2), dividing the image to be decrypted into blocks of the same number of blocks as the encryption process according to the color channels, using a decryption lookup table instead of an encryption lookup table, and performing the reverse operation of the same number of rounds as the encryption process on a block basis, thereby embedding the user's identity watermark in the decrypted image.

[0013] Furthermore, based on refined authentication using parity check, the image to be inspected is divided into corresponding sub-blocks, and the parity of the LSBs of each sub-block is determined. If the LSBs violate the predetermined rules, the image is determined to be tampered with.

[0014] Furthermore, the image restoration is achieved by reconstructing the thumbnail: (6.1) Extract LSB data from all image blocks that are determined to be tamper-free and form several complete or incomplete binary recovery bitstream copies according to the binary recovery bitstream laying positions during preprocessing; (6.2) Align the several complete or incomplete binary recovery bitstream copies; (6.3) For each bit, the probability of its value appearing in all copies is counted bit by bit. The value with a probability exceeding the first threshold is taken as the final value of the bit. The reconstructed binary recovered bit stream is obtained. The reconstructed complete binary recovered bit stream is deserialized and dequantized to remap and reconstruct a two-dimensional global thumbnail. (6.4) Upsample this global thumbnail to its original size, extract the pixel block corresponding to the coordinates of the "tampered" image block from the magnified global image, fill it into the tampered area of ​​the image to be detected, and complete the image restoration.

[0015] Furthermore, step (7) specifically includes: (7.1) Divide the restored image and the original preprocessed image into two parts respectively. For each block, pair them up sequentially; for each pixel pair in the corresponding block of the restored image and the original preprocessed image, select one pixel value and calculate the difference as the extracted difference signal. (7.2) Align the differential signals of all blocks. For each differential signal, count the probability of its occurrence in all blocks bit by bit. Take the value whose probability exceeds the second threshold as the final value of that bit. All the final values ​​are compiled into a table, which is the reconstructed watermark lookup table. ; (7.3) Using the coding matrix G used by the system in the key generation stage, the BCH codeword is recovered by solving a system of linear equations. ; (7.4) BCH decoding is performed to ultimately recover the user's original identity sequence. .

[0016] This invention provides an authenticated user-end watermarking system based on thumbnail-encrypted images, and an authenticated user-end watermarking method based on thumbnail-encrypted images according to any one of claims 1-9, characterized in that it includes: an image owner, an authorized user, and an arbitrator; Image owners include: The image preprocessing module is used to preprocess the original image, including a thumbnail-based holographic redundancy embedding layer and a parity-check-based refined authentication layer. The thumbnail-based holographic redundancy embedding layer is used to generate thumbnails, quantize them to obtain binary recovery bitstreams, and repeat them to uniformly fill the LSB plane of the entire original image. The parity-based refined authentication layer is used to divide the original image after thumbnail-based holographic redundancy embedding into [variables]. Blocks, and then subdivide each block into Perform parity checking on the sub-blocks; The image encryption module is used to divide the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image. The key generation and distribution module is used to generate a watermark lookup table containing user information for each authorized user, combine the encrypted lookup table to generate a decryption lookup table, and distribute it to authorized users along with the master encryption key as the decryption key. Authorized users include: The receiving module is used to receive the decryption key issued by the image owner; The decryption and watermark embedding module is used to decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding for the authenticated user terminal. The arbitrators include: The tamper detection and location module is used to check the parity of the image to be detected and determine the tamper location based on the refined authentication based on parity check. The tampered content recovery module is used to reconstruct the thumbnail from the untampered location to complete image recovery; The watermark extraction and user identity locking module is used to extract the differential signal by comparing the recovered image with the original preprocessed image, decipher the watermark sequence, and lock the user's identity.

[0017] Beneficial effects of this invention: 1. This invention has extremely strong anti-shearing and self-healing capabilities: Based on holographic redundancy embedding of thumbnails and refined authentication based on parity check, even if the image is cut and loses 75% of its content, this solution can still recover the full image content with high quality through the holographic redundancy information in the remaining 25% area and the majority voting mechanism, which is superior to existing dedicated recovery solutions.

[0018] 2. Blind-spot-free full-area protection: Compared to existing solutions that require the division of "hidden areas" which prevent watermarks from being embedded in those areas, this invention utilizes paired pixel difference perturbation to enable watermarks to be embedded in all pixels of the image, eliminating security blind spots.

[0019] 3. Balancing privacy and user experience: The encrypted images of this invention have clear and identifiable thumbnails that are highly similar to plaintext, making them easy to manage in the cloud. At the same time, they effectively protect the privacy of details through high-frequency information obfuscation.

[0020] 4. High computational efficiency: Encryption and watermarking are embedded and integrated in the underlying algebraic operations. The total processing time for a single 1024-pixel image is only 38.17 seconds, which is significantly better than the traditional serial "encryption-watermarking" scheme. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the existing watermarking scheme.

[0022] Figure 2 This is a schematic diagram of the process of an authenticable user-end watermarking method based on thumbnail-encrypted images according to an embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the image encryption process according to an embodiment of the present invention.

[0024] Figure 4 This is a schematic diagram of the paired pixel difference perturbation process in an embodiment of the present invention.

[0025] Figure 5 This is a schematic diagram illustrating the image perception effect in an embodiment of the present invention.

[0026] Figure 6 This is a schematic diagram of a pruning attack (75%) and recovery results in an embodiment of the present invention.

[0027] Figure 7 This is a schematic diagram of a pruning attack (50%) and recovery results in an embodiment of the present invention.

[0028] Figure 8 This is a schematic diagram of a pruning attack (25%) and recovery results in an embodiment of the present invention.

[0029] Figure 9 This is a schematic diagram illustrating the effect of recovery from a replication attack according to an embodiment of the present invention.

[0030] Figure 10 This is a schematic diagram comparing the success rate of identity tracing under different tampering rates in an embodiment of the present invention.

[0031] Figure 11 The following are the three-channel histograms of the pepper image encrypted in different rounds according to embodiments of the present invention: (a) R; (b) G; (c) B.

[0032] Figure 12 This is a schematic diagram of the key sensitivity test results in an embodiment of the present invention. Detailed Implementation

[0033] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments.

[0034] like Figure 2 As shown, this invention provides an authenticable user-side watermarking method based on thumbnail-encrypted images. This scheme achieves a full lifecycle security closed loop integrating "encrypted preview," "blind-spot-free copyright protection," and "tamper-proof blind detection and recovery" by constructing a self-healing mechanism based on holographic redundancy embedding and performing pairwise pixel difference perturbation encryption on the owner's side, and implicitly embedding robust watermarks during client decryption. The steps include: S101, Image preprocessing: The image owner preprocesses the original image, including thumbnail-based holographic redundancy embedding and parity-based refined authentication.

[0035] The core task of thumbnail-based holographic redundancy embedding is to construct an image content backup with extreme survivability, ensuring that it retains the information needed to reconstruct the overall semantics even when the carrier image suffers large-area cropping or smearing attacks. Unlike traditional schemes that map local image blocks to specific host blocks, this invention employs a "globally generated, holographically redundant" embedding strategy to solve the "tampering overlap" problem. This includes: (1) Generate a fixed-size global thumbnail from the original image using a bicubic interpolation algorithm.

[0036] For a size of The original image First, a fixed-size global thumbnail is generated using a bicubic interpolation algorithm. This thumbnail provides a high-level summary of the global semantic structure and color distribution of the original image, which is used for subsequent content restoration.

[0037] (2) Quantize the global thumbnail and serialize the quantized thumbnail matrix into a binary recovery bit stream.

[0038] To reduce data volume and support higher redundancy backup rates, we will The color depth was quantized from the standard 8-bit to 4-bit. Subsequently, the quantized thumbnail matrix was serialized into a binary recovery bitstream. .

[0039] (3) The ratio of the total capacity of the LSB plane of the original image to the length of the binary recovery bit stream is used to obtain the maximum number of redundant copies. The binary recovery bit stream is repeated a maximum number of redundant copies to generate a holographic recovery stream, and then it is evenly spread across the LSB plane of the entire image.

[0040] in: The total capacity of the LSB plane of the original image. To recover the bitstream length from binary, The maximum number of redundant replicas, The length and width of the original image.

[0041] Through the aforementioned mechanism, each local region of the image carries complete or partial descriptive information about the entire image. This characteristic ensures that even if more than 75% of the image is maliciously altered or lost, the arbitrator can still aggregate the remaining redundant fragments from the surviving regions to reconstruct the complete global content.

[0042] After the holographic restoration information is embedded, a refined authentication layer based on parity checking is constructed to achieve pixel-level precise positioning of the tampered area. This process utilizes a fragile watermark mechanism to ensure that any minor modifications to the content are sensitively detected. Specifically, the original image after holographic redundancy embedding based on thumbnails is first divided into segments of size [missing information]. The non-overlapping image blocks are further logically divided into blocks of size . A grid composed of tiny sub-blocks. For each sub-block, we calculate the sum of the LSBs of all pixels within it. Define a global parity rule that requires the sum of the LSBs of all sub-blocks to be even. If the LSB of a certain sub-block is even... If the preset parity rule is not met, the system will perform a LSB flip operation on the last pixel of the sub-block. This process essentially embeds a 1-bit message authentication code into each sub-block.

[0043] The final preprocessed image is obtained after the above dual processing. Visually speaking, With the original image Maintain a high degree of consistency ( A highly redundant self-verifying structure has been built within its LSB plane. This structure lays the foundation for subsequent precise location and high-fidelity recovery of tampered images based solely on the tampered image copy itself.

[0044] S102. Image encryption: The image owner divides the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image.

[0045] The image encryption stage is performed by the content owner, and its core objective is to deeply obfuscate high-frequency detail information while keeping the total number of pixels in each block of the image unchanged. For example... Figure 3 As shown, the encryption process uses a size of The system uses non-overlapping image blocks as the basic processing unit, and processes each color channel independently. To achieve high security, an iterative encryption strategy is employed, alternating between the two core operations of pairwise pixel interpolation and intra-block pixel scrambling. This process completely hides the original plaintext information, ultimately generating an encrypted image with the exact same size as the original image. The specific steps are as follows: (1) Divide the preprocessed original image into color channels. The image owner generates a block locally with a length of Encrypted lookup table Its elements are 0 or 1.

[0046] (2) A cryptographically secure pseudo-random number generator based on the standard cryptographic hash function SHA-256 is used to generate a secure sequence for each block.

[0047] To ensure the security and efficiency of the encryption process, each image block will generate a unique, unpredictable security sequence.

[0048] (2.1) Generate the pseudo-random sequence seed of the current block based on the master encryption key, the current encryption round and the sum of the pixels of the previous block. For the first block, the sum of the pixels of the previous block is the sum of the initial pixels of the entire channel.

[0049] To enhance the scheme's resistance to attacks such as differential cryptanalysis, embodiments of this invention establish a chain-like cryptographic dependency between blocks. This design ensures that any minor modification to the plaintext will rapidly propagate through the chain-like dependency and affect the encryption results of all subsequent blocks.

[0050] (2.2) The pseudo-random sequence seed of the current block is concatenated with a counter that increments from 0 and used as the input of the SHA-256 hash function for each round. Multiple rounds of hashing are performed, and the outputs of each round are concatenated until the total length required for encrypting the current image block is reached. This is then used as the secure sequence of the current block.

[0051] This "seed-counter" model is a standardized method for constructing stream ciphers, ensuring that the generated sequence is statistically indistinguishable from a truly random sequence to any attacker with limited computing resources. The final generated sequence... Will be converted An integer within the range, used for subsequent pairwise pixel difference perturbation operations.

[0052] (3) Perform pairwise pixel difference perturbation on the current block. The pairwise pixel difference perturbation process takes the image block as an independent unit, aiming to perform depth perturbation on the pixel value without changing the sum of pixels in each image block, such as... Figure 4 As shown, it includes: (3.1) Pair the pixels in the current block in order, calculate the pixel sum and pixel difference of the two pixel values ​​of each pair of pixels, and dynamically calculate an encrypted space for the difference of each pixel pair while ensuring that the pixel sum remains unchanged, so as to ensure that the two newly generated pixel values ​​of each pixel pair can strictly fall within the valid range of 0 to 255, and at the same time ensure that they are integers.

[0053] For any pair of pixels within the block Its pixel value Calculate its pixels and and pixel difference For each pixel pair, while ensuring Without changing the context, an encryption space is dynamically calculated for the difference between pixel pairs. This space is precisely defined to ensure that the two newly generated pixel values ​​strictly fall within the valid range of 0 to 255, while also guaranteeing that they are integers. The upper and lower boundaries of this encryption space are: in: The sum of the pixel values ​​of the two pixels is... Under the constraints, the effective minimum value that the difference can take is... For in and for S Under the constraints, the difference can take the effective maximum value.

[0054] Assuming encrypted pixel pairs The value is The difference is According to encryption principles, their sum remains the same. Therefore, the reconstructed relation is and Since pixel values ​​are non-negative integers, the following conditions must be met: and .

[0055] Depend on We can obtain: .Depend on We can obtain: At the same time, the pixel value cannot exceed its upper limit of 255, and must meet the following requirements. and .Depend on We can obtain: .Depend on , can be obtained .

[0056] In summary, the difference Must be satisfied simultaneously and Therefore, its effective minimum value is the larger of the two lower bounds, i.e. Its effective maximum value must be the smaller of the two upper bounds, i.e. .

[0057] To ensure that the reconstructed pixel values ​​are all integers, the parity of the difference must be consistent with the parity of the sum of the two pixel values.

[0058] Final difference encryption space It is not a continuous interval, but a discrete ordered set of all integers that satisfy the constraints, which ensures the reversibility of the encryption operation.

[0059] (3.2) For each pair of pixels, locate the index of the original value of the pixel difference between the two pixel values ​​in the encrypted space, and use the encrypted lookup table. The security sequence of the block is used to deeply scramble the original index and generate a new encrypted index.

[0060] in: It is the first The index of the original value of the pixel difference between two pixel values ​​within the encrypted space. It is the first The index of each pixel in the encrypted space after depth scrambling. It is the size of the encrypted space. It is the first The index of a pixel pair in an image patch. This is the safe sequence for this block.

[0061] (3.3) For each pair of pixels, according to the new encryption index A new difference is selected from the encrypted space, and the new difference is algebraically combined with the pixel sum to reconstruct each pixel pair of two encrypted pixel values ​​that are completely different from the original values ​​but whose algebraic relationship is controlled.

[0062] By traversing all pixel pairs within the block, the visual information of the original image block is eliminated and transformed into ciphertext data with uniform statistical properties. Since the sum of each pixel pair remains unchanged, the total pixel sum of the entire block is also preserved, ensuring the stability of the system and its subsequent applications.

[0063] (4) Repeat steps (2)-(3) until the pairwise pixel difference perturbation of all blocks is completed. (5) For each block after pairwise pixel difference perturbation, a global permutation sequence is generated using the master encryption key as a random number seed. The pixel of each block is scrambled once according to the global permutation sequence to rearrange its spatial position. To further enhance encryption strength and effectively resist various statistical attacks, after the difference perturbation is completed, a permutation encryption is performed on all pixels within the block to rearrange their spatial positions. This step applies a deterministic pseudo-random permutation uniquely determined by the master encryption key. The algorithm uses the master encryption key as a globally unique random number seed to generate a global permutation sequence of length equal to the total number of pixels in the block, and uniformly maps and scrambles the same positions of all pixels within the image block that have already undergone difference encryption based on this global permutation sequence.

[0064] (6) Repeat steps (2)-(5) until the predetermined encryption round is reached.

[0065] By alternately performing pairwise pixel difference perturbation and permutation encryption operations, and reaching a preset number of encryption rounds. The plaintext information of the original image blocks is completely hidden, ultimately generating an encrypted image of the exact same size as the original image. .

[0066] S103. Key generation: The image owner generates a watermark lookup table containing user information for each authorized user, combines the encrypted lookup table to generate a decryption lookup table, and distributes it to the authorized user along with the master encryption key as the decryption key.

[0067] (1) The owner is each authorized user Allocate a length of Unique binary sequence As its identity identifier.

[0068] (2) Adopt Encoding the original identity sequence Process to generate a binary sequence .

[0069] To enhance the watermark's error correction capabilities when encountering attacks such as noise and compression, the system adopts... Encoding the original identity sequence This process generates a longer binary sequence containing redundant information. .

[0070] (3) The owner generates an encoding matrix locally. One-dimensional codewords are transformed through matrix operations. Mapped and expanded into a two-dimensional watermark lookup table uniquely bound to the user's identity. .

[0071] (4) Use the watermark lookup table and the encryption lookup table used during encryption The data is then merged to generate a personalized decryption lookup table for the user. .

[0072] (5) The owner will give the key pair Distributed to authorized users via secure channel .

[0073] Ultimately, the complete personalized decryption key is derived from the master encryption key. and decryption lookup table Composition, the owner will give the key pair Distributed to authorized users via secure channel It is important to note that the watermark lookup table is confidential to users; no information about the watermark lookup table can be obtained simply by decrypting it.

[0074] S104. Decryption and watermark embedding: Authorized users decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding of the authenticated user terminal.

[0075] When authorized user Received encrypted image and personalized decryption keys Then, the decryption operation can be performed. The decryption process and the watermark embedding process are completed simultaneously. While restoring the image content, the user's identity is also restored. The corresponding watermark It is then implicitly embedded in the image. Decryption is also performed on a per-image-block basis and requires the same number of rounds as encryption. The reverse operation of the wheel.

[0076] This process is similar to the encryption process in step S102, using a decryption lookup table instead of an encryption lookup table, and performing the reverse operation for the same number of rounds as during encryption, block by block, to obtain a decrypted image that also embeds the user's identity watermark. The specific steps are as follows: (1) Divide the image to be decrypted into color channels. The block.

[0077] (2) Inverse scrambling of pixels within a block, authorized users use the master encryption key As a seed, it is input into the same pseudo-random number generator as during encryption to regenerate a permutation sequence that is exactly the same as during encryption. Then, the inverse sequence of this global permutation sequence is calculated, and the pixels in each block are precisely mapped back to their original positions before the spatial scrambling operation is performed based on this inverse sequence.

[0078] This step aims to restore the pixels that were scrambled during the encryption phase to their original logical positions.

[0079] (3) For the current block, use the master encryption key Regenerate a secure sequence consistent with the one used during encryption. .

[0080] This corresponds to step (2) in step S102.

[0081] (4) Perform the inverse operation of pairwise pixel difference perturbation on the current block, including: (4.1) Pair the pixels in the current block in order, calculate the pixel sum and pixel difference of the two pixel values ​​of each pair of pixels, and reconstruct the difference space consistent with the encryption for each pixel pair using the principle that the pixel sum remains unchanged.

[0082] (4.2) For each pair of pixels, locate the encrypted index of the pixel difference between the two pixel values ​​in the difference space, and use the decryption lookup table value. And the security sequence of the current block, for the cryptographic index Perform the inverse operation to recover the new index.

[0083] in, It is the difference space Size, It is the index of the current pixel pair. It is the first An encrypted index of the pixel difference between two pixel values ​​within the difference space. It is the first The new index is recovered from each pixel pair.

[0084] (4.3) For each pair of pixels, based on the recovered new index A new difference is selected from the difference space, and the new difference is used to recover the finely adjusted pixel pair.

[0085] (5) Repeat steps (2)-(4) until a round consistent with the encryption process is reached.

[0086] The final generated image Not only has the high-quality visual content been restored, but the user's identity watermark has also been subtly and inseparably embedded.

[0087] S105. Tamper detection and location: The arbitrator checks the parity of the image to be detected based on the refined authentication based on parity check to determine the location of the tampering.

[0088] This stage is conducted independently by the arbitrator, and its goal is to accurately identify and locate all tampered areas using only the suspected image copy itself. The specific process is as follows: The image to be inspected is divided into segments identical to those used during preprocessing. indivual The image blocks. For each image block, the arbitrator further logically divides it into multiple... The system calculates the parity of the sum of the LSBs of all pixels within each sub-block according to the parity rules set in the refined parity verification. If any sub-block within a block does not satisfy the preset parity rules, it proves that the LSB plane of that sub-block has been modified. Since even a minor tampering can destroy the original LSB parity structure, the system determines the entire image block as "tampered with".

[0089] S106. Content restoration: The arbitrator reconstructs the thumbnail from the unaltered location to restore the image.

[0090] Since the recovery information is embedded in a holographically redundant manner, the thumbnail can be reconstructed from the LSB data in the untampered location image block.

[0091] (1) Extract LSB data from all image blocks that are determined to be "untampered" and form several complete or incomplete binary recovery bitstream copies according to the binary recovery bitstream laying position during preprocessing.

[0092] (2) Align the several complete or incomplete binary recovery bitstream copies.

[0093] (3) For each bit, the probability of its value appearing in all copies is counted bit by bit. The value with a probability exceeding the threshold (50%) is taken as the final value of the bit. The reconstructed binary recovery bit stream is obtained. The reconstructed complete binary recovery bit stream is deserialized and dequantized to remap and reconstruct a two-dimensional global thumbnail.

[0094] (4) Upsample this global thumbnail to the original size, extract the pixel block corresponding to the coordinates of the "tampered" image block from the enlarged global image, fill it into the tampered area of ​​the image to be detected, and complete the image restoration.

[0095] To eliminate potential bit errors and reconstruct the most accurate recovery information, this scheme introduces a bit-level majority voting mechanism. Specifically, for each bit, its frequency of occurrence in all valid copies is counted, and the value with a probability exceeding 50% is taken as the final value of that bit. Through this process, a high-confidence global thumbnail is reconstructed. Finally, this method ensures that semantic-level recovery of the entire image content can be achieved as long as the remaining valid regions in the image are sufficient to piece together a complete set of information.

[0096] S107. Extract the watermark. By comparing the restored image with the original preprocessed image, extract the differential signal, solve the watermark sequence, and lock the user's identity.

[0097] This stage requires the owner to provide the preprocessed image. The arbitrator analyzes the systematic pixel perturbations caused by the user decrypting the embedded watermark by comparing the differential signals of the preprocessed and restored images. Since the watermark is redundantly embedded in all image blocks, the arbitrator uses a majority voting mechanism to extract signals from all untampered blocks and robustly reconstruct the complete watermark lookup table. Then, the encoding matrix is ​​used. By solving the system of linear equations To recover BCH codewords Finally, regarding BCH decoding is then performed. The robust error correction capabilities of BCH codes can correct minor bit errors introduced by channel noise or slight attacks, ultimately recovering the user's original identity sequence. By matching in the list of authorized users This allows for the identification of users who illegally redistribute or maliciously tamper with content, providing crucial technical evidence for pursuing copyright liability.

[0098] Specifically: (1) Divide the restored image and the original preprocessed image into two parts respectively. For each block, pair them up sequentially; for each pixel pair in the corresponding block of the restored image and the original preprocessed image, select one pixel value and calculate the difference as the extracted difference signal.

[0099] (2) Align the differential signals of all blocks. For each differential signal, count the probability of its value appearing in all blocks bit by bit. Take the value whose probability exceeds the second threshold as the final value of that bit. All the final values ​​form a table, which is the reconstructed watermark lookup table. .

[0100] (3) Using the coding matrix G used by the system in the key generation stage, the BCH codeword is recovered by solving a system of linear equations. .

[0101] (4) BCH decoding is performed to ultimately recover the user's original identity sequence. .

[0102] This invention also provides an authenticated user-side watermarking system based on thumbnail-encrypted images. The authenticated user-side watermarking method based on thumbnail-encrypted images described above includes: the image owner, the authorized user, and the arbitrator. Image owners include: The image preprocessing module is used to preprocess the original image, including a thumbnail-based holographic redundancy embedding layer and a parity-based refined authentication layer.

[0103] The thumbnail-based holographic redundancy embedding layer is used to generate thumbnails, quantize them to obtain a binary recovery bitstream, and repeat the bitstream to uniformly cover the entire LSB plane of the original image.

[0104] The parity-based refined authentication layer is used to divide the original image after thumbnail-based holographic redundancy embedding into [variables]. Blocks, and then subdivide each block into The sub-blocks are parity checked.

[0105] The image encryption module is used to divide the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image.

[0106] The key generation and distribution module is used to generate a watermark lookup table containing user information for each authorized user, combine it with the encryption lookup table to generate a decryption lookup table, and distribute it to authorized users along with the master encryption key as the decryption key.

[0107] Authorized users include: The receiving module is used to receive the decryption key issued by the image owner.

[0108] The decryption and watermark embedding module is used to decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding for the authenticated user terminal.

[0109] The arbitrators include: The tamper detection and location module is used to check the parity of the image to be detected and determine the location of the tampering based on the refined authentication based on parity check.

[0110] The tampered content recovery module is used to reconstruct thumbnails from untampered locations to complete image recovery.

[0111] The watermark extraction and user identity locking module is used to extract the differential signal by comparing the recovered image with the original preprocessed image, decipher the watermark sequence, and lock the user's identity.

[0112] To verify the performance, the embodiments of the present invention verify the scheme from aspects such as perceived quality, tamper recovery function, statistical security, watermark robustness and computational efficiency.

[0113] Dataset: The experiment used the first 500 images from the Helen dataset1. All images were converted to PNG format and resized to 512×512.

[0114] Experimental Platform: Efficiency experiments were conducted on a desktop computer equipped with an Intel(R) Core(TM) i7-10700 CPU and 16GB RAM, while the remaining experiments were conducted on a computer with an AMD Ryzen 9 7945HX CPU with Radeon Graphics and 16GB RAM.

[0115] To objectively evaluate the proposed VCW-TPE scheme's ability to synergistically optimize image privacy and usability in a cloud-sharing scenario, this experiment uses the first 500 color images from the Helen_1 dataset as the test set and selects the textured and color-rich "sailboat" image as a representative sample for visual comparison. The experiment designs a wide range of benchmark comparison schemes, covering four mainstream technical approaches: high-performance, accurate thumbnail-preserving encryption schemes (FTPE, PR3), optical transformation-based TPE schemes (DRPE-Cloud), traditional client-side watermarking schemes (CSSR-W), and client-side watermarking schemes supporting preview (PCS-W, PHS-W). The experiment compares the visual quality and statistical indicators of the encrypted ciphertext under different block sizes, verifying whether the proposed scheme can effectively obfuscate high-frequency detail information of the original image while providing high-quality thumbnail previews, thus establishing its advantage in the trade-off between security and usability.

[0116] 1. Perception effect verification Figure 5 Table 1 shows the qualitative visual effects and quantitative evaluation indicators of different schemes. Figure 5 The visualization results show that the ciphertext generated by the traditional CSSR-W scheme appears as meaningless snowflake-like noise, completely losing its visibility; while the PR3, PCS-W, and PHS-W schemes can retain some information. In contrast, this scheme effectively masks high-frequency privacy details such as leaf texture and sailboat structure, while clearly preserving the coarse-grained contours and main color distribution of the original image. Its visual effect is highly similar to the high-performance FTPE scheme, both presenting highly recognizable block thumbnails. The quantitative data in Table 1 further confirms this conclusion. Under 8×8 block division, the peak signal-to-noise ratio of this scheme reaches 14.9751dB, and the structural similarity is 0.3421, which is significantly better than the PCS-W (PSNR 9.79 dB) and CSSR-W (PSNR 6.26 dB) schemes, and is at the same performance level as the FTPE scheme (PSNR 15.13 dB). In summary, the experimental results show that VCW-TPE achieves excellent preview usability while ensuring content confidentiality, providing efficient technical support for the screening and management of massive media data in the cloud.

[0117] Table 1 To evaluate the watermark concealment of the decrypted image, Table 2 compares the perceptual quality metrics of our proposed solution with existing client-side watermarking solutions (CSSR-W, PCS-W, PHS-W) at different block sizes. The statistics in Table 2 show that our proposed solution exhibits excellent visual fidelity across all test block sizes. Specifically, compared to the PHS-W solution's approximately 30dB PSNR and 0.87 SSIM, our proposed solution maintains a stable PSNR between 39.2dB and 39.4dB, and an SSIM above 0.97. This indicates that the watermarked image generated by VCW-TPE is virtually imperceptible to the human visual system, significantly outperforming the PHS-W solution. It is worth noting that although CSSR-W and PCS-W slightly outperform our proposed solution in some metrics, this is mainly due to the simplicity of their functional design. VCW-TPE simultaneously embeds two layers of information—the "global holographic recovery stream" and the "client identity watermark"—in the same carrier. Despite carrying a larger payload than the comparison scheme, the proposed scheme can still maintain a high-quality image reconstruction of nearly 40dB, which fully demonstrates that the proposed algorithm achieves a better trade-off between embedding capacity and visual quality.

[0118] Table 2 2. Blind tamper detection and recovery function verification 2.1 Quantitative Analysis of Tamper Detection Accuracy To verify the authentication layer's sensitivity to malicious tampering, a performance evaluation experiment based on cropping attacks was conducted. Twelve test images were modified by varying proportions from 10% to 75%, and the average accuracy, recall, precision, and F1 score of the detection algorithm were statistically analyzed. The experimental results in Table 3 demonstrate that the VCW-TPE authentication algorithm exhibits near-perfect detection performance and possesses extremely high forensic credibility.

[0119] Table 3 2.2 Recovery performance stability under different tampering ratios The experiment simulated a shearing attack, altering the proportions. The percentage gradually increased from 10% to 75%. Table 4 details the peak signal-to-noise ratio and structural similarity index of the recovered images from the 12 test images under different attack intensities.

[0120] As can be observed from the statistics in Table 4, for the low-tampering range (10%-30%), the system exhibits excellent recovery capabilities, with an average PSNR maintained above 30dB and an SSIM exceeding 0.85. This indicates that under minor attacks, the holographic redundancy mechanism can almost perfectly reconstruct the damaged area, with visual differences indistinguishable to the naked eye. For the high-tampering range (40%-60%), despite losing more than half of the original pixels, the scheme can still maintain a PSNR of around 27dB. Thanks to the bit-level majority voting mechanism designed in this paper, as long as the remaining untampered area contains at least one complete copy of the global thumbnail, the recovered bitstream can be effectively extracted. For the extreme tampering range (70%-75%), even in the extreme case where 75% of the content is maliciously removed (e.g., ... Figure 6 As shown in the figure, this scheme can still ensure that the average PSNR remains above 25dB and the SSIM is about 0.70. This means that the main semantic structure of the image (such as the facial contours of Mandrill and the edge texture of Peppers) is completely preserved.

[0121] Table 4 2.3 Comparative Analysis of Tamper Recovery Quality To further establish the advantages of this solution in ensuring content integrity, we compared VCW-TPE with two existing solutions. One is a client-side watermarking solution, including the traditional CSSR-W and the preview-enabled PCS-W and PHS-W. The other is a solution specifically for image restoration developed by Li et al. and Ozkaya et al.

[0122] The experiments were conducted at 75% (Table 5). Figure 6 ), 50% (Table 6, Figure 7 ) and 25% (Table 7, Figure 8 This is performed under three different shearing attack intensities.

[0123] 1. Comparison with client-side watermarking schemes: Tables 4 to 6 clearly show that traditional client-side watermarking schemes experience a sharp drop in PSNR to the 6-15dB range after being subjected to a cropping attack. These schemes are designed solely for copyright tracking or previewing and lack self-correction mechanisms. Once an image is cropped, the information in the damaged area is permanently erased, and the resulting noise or black hole areas severely disrupt the image's visual coherence. In contrast, VCW-TPE maintains an average PSNR of around 28dB even under a 75% attack. Figure 6 As shown in the first row, when the Mandrill image is 75% occluded, the contrast scheme can only display the remaining black areas, while this scheme ( Figure 6(The second line) Successfully reconstructed the lost facial features. This proves that this solution not only fills the gap in content recovery technology using client-side watermarking, but also realizes a paradigm shift from "passive tracking" to "active defense".

[0124] 2. Comparison with dedicated recovery schemes: This scheme also demonstrates highly competitive performance when compared to advanced schemes focused on image recovery. Under a 75% attack, the recovered PSNR of the "Blonde" image is 27.7374 dB, slightly lower than that of Ozkaya et al.

[0125] The accuracy is 28.1662 dB, but better than Li et al.'s 24.2931 dB. On some images with complex textures (such as Cameraman), this scheme even surpasses some dedicated restoration algorithms. It is important to emphasize that the schemes of Li et al. and Ozkaya et al. are single-function restoration algorithms, lacking encrypted preview and copyright tracking capabilities. VCW-TPE, on the other hand, achieves restoration using the remaining LSB space while simultaneously satisfying the strong constraints of "thumbnail-preserving encryption" and "client-side watermark embedding." Under such stringent channel capacity limitations, this scheme still achieves restoration accuracy on par with dedicated restoration algorithms, fully demonstrating the efficiency of the proposed pairwise pixel difference perturbation and holographic embedding algorithm in balancing information load.

[0126] Table 5 Table 6 Table 7 Figure 9 Experimental results of an object duplication attack performed on the "Trucks" image are presented. Unlike simple cut-out attacks, this type of attack aims to forge scene semantics by duplicating salient objects (such as vehicles) to confuse human visual inspection, making it highly deceptive. Figure 9 As shown in (b), the attackers attempted to increase the number of trucks in the diagram by using a "copy-and-move" method to create a false scenario. Figure 9 (c) shows that the algorithm accurately located all areas covered by forgery, with a detection accuracy at the pixel level, and no false detections or missed detections occurred. Figure 9 (d) shows the final restoration result. Experimental results show that the restored image successfully removed the three reconstructed trucks, achieving a PSNR of 31.54 dB and an SSIM of 0.8680. This demonstrates that VCW-TPE not only resists physical signal loss but also effectively defends against semantic content forgery, ensuring the authenticity and non-repudiation of digital evidence.

[0127] 2.4 Performance Analysis of Identity Tracing under Different Cutting Ratios This experiment simulated tampering attacks ranging from 10% to 75% on the aforementioned 12 standard test images to evaluate the system's success rate in extracting user identity information under different levels of damage. The experimental results are as follows: Figure 10 As shown. We define "extraction success rate" as the percentage of completely correct user identities successfully recovered from all test images.

[0128] Figure 10 Quantitative data shows that the VCW-TPE scheme exhibits superior resilience. Even under extreme attacks where 75% of the watermarked image is lost (i.e., only 25% of the effective area remains), the identity tracking success rate of this scheme remains consistently at 100%. In contrast, the tracking performance of the three benchmark schemes shows varying degrees of decline as the cropping ratio increases. While CSSR-W performs reasonably well at low tampering rates, its success rate begins to drop significantly when the cropping ratio exceeds 40%. PHS-W and PCS-W, the two schemes that support preview, exhibit poor robustness. In particular, PCS-W's tracking success rate drops to 0% when the cropping ratio reaches 50%. This "cliff-like" drop usually stems from the over-reliance on specific structures (such as hidden regions or specific frequency bands) in its scheme design. Once these regions carrying critical synchronization information or watermark payloads are lost, the entire extraction process fails.

[0129] In summary, the experimental results demonstrate that the VCW-TPE framework not only possesses excellent resilience and survivability, resisting up to 75% of malicious tampering, but also significantly outperforms existing client-side watermarking solutions in terms of recovery quality, and is comparable to dedicated recovery algorithms. This confirms that the solution can provide a closed-loop security guarantee integrating "pre-emptive defense, in-process detection, and post-event recovery" for digital media in a cloud environment.

[0130] 3. Performance analysis against common attacks Differential Attack Analysis: Differential attack tests were conducted using "pepper," "baboon," and "flowers" as test images. A single pixel was changed in each block of the original plaintext image, and then encrypted using the same key. The block size was 64×64. Finally, the NPCR and UACI between the two ciphertexts were calculated. Experimental results show that the NPCR values ​​all exceeded 95%. This indicates that any tiny perturbation to the plaintext is amplified and spread throughout the ciphertext during encryption, creating an avalanche effect that prevents attackers from obtaining any useful information by analyzing the input-output differences. Therefore, this scheme exhibits strong resistance to differential attacks.

[0131] Table 8 Statistical attack: The experiment involved performing histogram analysis on the image in each channel after a certain number of encryption rounds. This is because for a secure TPE algorithm, after a certain number of encryption rounds, the histogram in each channel tends to stabilize and differs from the histogram of the original image. Figure 11 Examples of histograms of the encrypted images for each channel after different rounds of encryption of the pepper image are shown. Experiments show that after a certain number of rounds, the histograms of each channel of the encrypted image tend to stabilize and differ from the original image, thus resisting statistical attacks. In this experiment, all images exhibited similar results.

[0132] Key Space Value and Key Sensitivity Analysis: A secure encryption scheme must be highly sensitive to the key; that is, any tiny change to the key should lead to a significant difference in the encryption result and cause decryption to fail completely. To verify the key sensitivity of this scheme, a small change was made to the key value. The encryption and decryption images of the two keys are shown below. Figure 12 As shown in Table 9, the values ​​of key1 and key2 differ by 1. First, the same original image is encrypted using both keys, resulting in two visually completely different ciphertexts. To quantitatively evaluate their difference, the NPCR and UACI values ​​between the two ciphertext images are calculated. As shown in Table 9, the NPCR values ​​of both ciphertexts exceed 0.98 on all channels, reaching a maximum of 0.9952. Simultaneously, the UACI values ​​also exceed 0.22, reaching a maximum of 0.2871. Subsequently, attempts to decrypt the ciphertext using an incorrect key result in a completely meaningless noisy image; only using the correct key can successfully recover a high-quality image. In conclusion, the proposed VCW-TPE scheme exhibits extremely high key sensitivity. Even a small change in the key is sufficient to trigger an avalanche effect, generating statistically completely different ciphertexts and causing complete decryption failure.

[0133] Table 9 4. Efficiency Verification The encryption schemes compared are the ideal TPE and TPE2, as well as the F-TPE (the vector length used in this experiment is 3). This is because they are also precise thumbnail-preserving encryptions and can achieve multi-round substitution-permutation encryption like the scheme of this invention. The time spent on a single round of encryption for each scheme is shown in Table 10, and it can be found that the scheme of this invention has a lower average time cost than other schemes.

[0134] Table 10 Finally, the efficiency and performance of the proposed solution were evaluated and compared with related technologies. Table 11 shows that the preprocessing and watermark extraction modules account for a relatively low proportion of the total time. This indicates that the mechanism for embedding the vulnerable authentication code and extracting the traceable watermark is lightweight and highly efficient, ensuring smooth processes during content publishing and arbitration. The key distribution time remains at the millisecond level across all image sizes and is unaffected by the increase in image data size. This result strongly demonstrates that the VCW-TPE framework inherits the inherent advantages of the client-side embedding mode, namely, the high controllability of computational and communication overhead on the owner side as the number of users expands.

[0135] Table 11 As shown in Table 12, the overall execution time of VCW-TPE is significantly lower than that of general client-side watermark embedding and TPE schemes. This result confirms that VCW-TPE achieves the underlying fusion of the TPE and watermark embedding processes through a pairwise pixel difference perturbation encryption mechanism, effectively eliminating redundant calculations caused by the serialization of functional modules.

[0136] Across all test resolutions, VCW-TPE consistently delivers significantly shorter runtimes than traditional encryption benchmarks (TDES and AES-S-128), demonstrating the inherent efficiency of its LUT-based and algebraic constraint-based architecture in image processing. Furthermore, compared to solutions focused solely on a single TPE objective, VCW-TPE maintains highly competitive performance even after integrating authentication, recovery, and traceability functionalities. As image size increases, VCW-TPE's runtime exhibits a stable growth trend, similar to Celik et al.'s approach, indicating that the framework possesses good scalability and can effectively address the sharing needs of high-resolution digital images.

[0137] In summary, the VCW-TPE solution achieves a complete security loop encompassing secure preview, infringement tracing, content authentication, and blind recovery. Through efficient integration at the algorithm level, it demonstrates significant computational efficiency advantages and excellent scalability, meeting the high-performance requirements of trusted digital media distribution systems. It is worth noting that all methods were implemented in Matlab under the same hardware conditions. While this objectively captures the algorithmic behavior under standardized experimental settings, Matlab's computational architecture naturally favors matrix operations (such as LUT-based encryption) rather than hybrid and permutation operations (as required by DES and AES).

[0138] Table 12 5. Functional Analysis Table 13 presents a comprehensive comparison of the key functional characteristics of this solution with four existing representative technical solutions. While traditional client-side watermarking solutions solve the scalability problem during large-scale distribution and possess copyright traceability capabilities, their encrypted text lacks visual usability and does not support content recovery. Advanced TPE solutions focus on balancing privacy and preview, but often lack copyright protection and robust recovery mechanisms. Dedicated tamper detection and recovery solutions, while excelling in authentication accuracy and self-healing capabilities, are typically limited by the owner-side embedding mode and cannot resist robust attacks for rights confirmation. Existing client-side watermarking solutions supporting previews attempt to accommodate multiple functions, but due to structural blind spots, they cannot achieve full-domain protection and content recovery. In contrast, the VCW-TPE solution, through underlying algorithm fusion, successfully breaks down the barriers between the above functions. It is currently the only comprehensive solution capable of simultaneously achieving efficient client-side embedding, encrypted image visual preview, copyright traceability, high-precision tamper detection, and robust blind recovery, effectively filling the gap in existing technologies for the entire lifecycle security of digital media.

[0139] Table 13 In summary, the solutions of this invention have superior performance compared to existing technologies in terms of perceived quality, tamper recovery function, statistical security, watermark robustness, and computational efficiency.

[0140] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for authenticating user-side watermarking based on thumbnail-encrypted images, characterized in that, Includes the following steps: (1) Image preprocessing: The image owner preprocesses the original image, including thumbnail-based holographic redundancy embedding and parity-based fine-grained authentication. The thumbnail-based holographic redundancy embedding involves generating a thumbnail, quantizing it to obtain a binary recovery bitstream, and repeating it to uniformly fill the entire LSB plane of the original image. The refined authentication based on parity check divides the original image after thumbnail-based holographic redundancy embedding into... Blocks, and then subdivide each block into Perform parity checking on the sub-blocks; (2) Image encryption: The image owner divides the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image. (3) Key generation and distribution: The image owner generates a watermark lookup table containing user information for each authorized user, and generates a decryption lookup table by combining the encryption lookup table with the main encryption key, and distributes it to the authorized user as a decryption key. (4) Decryption and watermark embedding: Authorized users decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding of the authenticated user terminal. (5) Tamper detection and location: For the image to be detected, the parity is checked to determine the tamper location based on the refined authentication based on parity check. (6) Restoration of tampered content: The arbitrator reconstructs the thumbnail from the untampered location to complete the image restoration; (7) Extract watermark to lock user identity. By comparing the recovered image with the original preprocessed image, extract the differential signal, solve the watermark sequence and lock the user identity.

2. The method for authenticable user-side watermarking based on thumbnail-encrypted images according to claim 1, characterized in that, The thumbnail-based holographic redundancy embedding includes: (1.1) Generate a fixed-size global thumbnail from the original image using a bicubic interpolation algorithm; (1.2) Quantize the global thumbnail and serialize the quantized thumbnail matrix into a binary recovered bit stream; (1.3) The ratio of the total capacity of the LSB plane of the original image to the length of the binary recovery bitstream is used to obtain the maximum number of redundant copies. The binary recovery bitstream is repeated a maximum number of redundant copies to generate the holographic recovery stream, and it is evenly spread across the LSB plane of the entire image. in: The total capacity of the LSB plane of the original image. To recover the bitstream length from binary, The maximum number of redundant replicas, The length and width of the original image; The refined authentication based on parity check includes: calculating the sum of the LSBs of all pixels in each sub-block, and flipping the last pixel to make the sum satisfy a predetermined even parity rule.

3. The method for authenticable user-end watermarking based on thumbnail-encrypted images according to claim 1, characterized in that, Step (2) specifically involves: (2.1) Divide the preprocessed original image into color channels respectively. The image owner generates a block locally with a length of Encrypted lookup table Its elements are 0 or 1; (2.2) A secure sequence is generated for the current block using a cryptographically secure pseudo-random number generator based on the standard cryptographic hash function SHA-256; (2.2.1) Generate the pseudo-random sequence seed of the current block based on the master encryption key, the current encryption round, and the sum of the pixels of the previous block. For the first block, the sum of the pixels of the previous block is the sum of the initial pixels of the entire channel. (2.2.2) The pseudo-random sequence seed of the current block is concatenated with a counter that increments from 0 and used as the input of the SHA-256 hash function for each round. Multiple rounds of hashing are performed, and the outputs of each round are concatenated until the total length required for encrypting the current image block is reached. This is then used as the secure sequence of the current block. (2.3) Perform pairwise pixel interpolation perturbation on the current block, including: (2.3.1) Pair the pixels in the current block in order, calculate the pixel sum and pixel difference of the two pixel values ​​of each pair of pixels, and dynamically calculate an encrypted space for the difference of each pixel pair while ensuring that the pixel sum remains unchanged, so as to ensure that the two newly generated pixel values ​​of each pixel pair can strictly fall within the valid range of 0 to 255, and at the same time ensure that they are integers. (2.3.2) For each pair of pixels, locate the index of the original value of the pixel difference between the two pixel values ​​in the encrypted space, and use the encrypted lookup table. The security sequence of this block is used to deeply scramble the original index and generate a new encrypted index. in: It is the first The index of the original value of the pixel difference between two pixel values ​​within the encrypted space. It is the first The new encrypted index in the encrypted space after depth scrambling of each pixel pair. It is the size of the encrypted space. It is the first The index of a pixel pair in an image patch. This is the safe sequence for this block; (2.3.3) For each pair of pixels, according to the new encryption index Select a new difference from the encrypted space. The new difference is algebraically combined with the pixel sum to reconstruct each pixel pair of two encrypted pixel values ​​that are completely different from the original values ​​but whose algebraic relationship is controlled. (2.4) Repeat steps (2.2)-(2.3) until the pairwise pixel difference perturbation of all blocks is completed; (2.5) For each block after pairwise pixel difference perturbation, a global permutation sequence is generated using the master encryption key as a random number seed. The scrambling operation is performed on each pixel block once according to the global permutation sequence to rearrange its spatial position. (2.6) Repeat steps (2.2)-(2.5) until the predetermined encryption round is reached.

4. The method for authenticable user-end watermarking based on thumbnail-encrypted images according to claim 3, characterized in that, The upper and lower boundaries of the encrypted space are: in: The sum of the pixel values ​​of the two pixels is... Under the constraints, the effective minimum value that the difference can take is... Let S be the maximum effective value that the difference can take under the constraint that the sum is S; To ensure that the reconstructed pixel values ​​are all integers, the parity of the difference must be consistent with the parity of the sum of the two pixel values.

5. The method for authenticable user-side watermarking based on thumbnail-encrypted images according to claim 1, characterized in that, Step (3) specifically involves: (3.1) The owner is each authorized user Allocate a length of Unique binary sequence As its identity identifier; (3.2) Adopt Encoding the original identity sequence Process to generate a binary sequence ; (3.3) The owner generates an encoding matrix locally. One-dimensional codewords are transformed through matrix operations. Mapped and expanded into a two-dimensional watermark lookup table uniquely bound to the user's identity. ; (3.4) Use the watermark lookup table and the encryption lookup table used during encryption The data is then merged to generate a personalized decryption lookup table for the user. ; (3.5) The owner will give the key pair Distributed to authorized users via secure channel .

6. The method for authenticable user-side watermarking based on thumbnail-encrypted images according to claim 1, characterized in that, The specific steps (4) are as follows: according to the encryption process in step (2), the image to be decrypted is divided into blocks of the same number of blocks as the encryption process according to the color channels. The decryption lookup table is used instead of the encryption lookup table. The reverse operation of the same number of rounds as the encryption is performed by block. The user's identity watermark is then embedded in the decrypted image.

7. The method for authenticable user-end watermarking based on thumbnail-encrypted images according to claim 1, characterized in that: Based on parity check-based refined authentication, the image to be inspected is divided into corresponding sub-blocks, and the parity of the LSB of each sub-block is determined. If the LSB violates the predetermined rules, it is determined to be tampered with.

8. The method for authenticable user-side watermarking based on thumbnail-encrypted images according to claim 1, characterized in that, The image restoration is achieved by reconstructing the thumbnail: (6.1) Extract LSB data from all image blocks that are determined to be tamper-free and form several complete or incomplete binary recovery bitstream copies according to the binary recovery bitstream laying positions during preprocessing; (6.2) Align the several complete or incomplete binary recovery bitstream copies; (6.3) For each bit, the probability of its value appearing in all copies is counted bit by bit. The value with a probability exceeding the first threshold is taken as the final value of the bit. The reconstructed binary recovered bit stream is obtained. The reconstructed complete binary recovered bit stream is deserialized and dequantized to remap and reconstruct a two-dimensional global thumbnail. (6.4) Upsample this global thumbnail to its original size, extract the pixel block corresponding to the coordinates of the "tampered" image block from the magnified global image, fill it into the tampered area of ​​the image to be detected, and complete the image restoration.

9. The method for authenticable user-end watermarking based on thumbnail-encrypted images according to claim 5, characterized in that, The specific steps (7) are as follows: (7.1) Divide the restored image and the original preprocessed image into two parts respectively. For each block, pair them up sequentially; for each pixel pair in the corresponding block of the restored image and the original preprocessed image, select one pixel value and calculate the difference as the extracted difference signal. (7.2) Align the differential signals of all blocks. For each differential signal, count the probability of its occurrence in all blocks bit by bit. Take the value whose probability exceeds the second threshold as the final value of that bit. All the final values ​​are compiled into a table, which is the reconstructed watermark lookup table. ; (7.3) Using the coding matrix G used by the system in the key generation stage, the BCH codeword is recovered by solving a system of linear equations. ; (7.4) BCH decoding is performed to ultimately recover the user's original identity sequence. .

10. A user-end watermarking system based on thumbnail-encrypted images, based on the user-end watermarking method based on thumbnail-encrypted images according to any one of claims 1-9, characterized in that, include: Image owner, authorized user, and arbitrator; Image owners include: The image preprocessing module is used to preprocess the original image, including a thumbnail-based holographic redundancy embedding layer and a parity-check-based refined authentication layer. The thumbnail-based holographic redundancy embedding layer is used to generate thumbnails, quantize them to obtain binary recovery bitstreams, and repeat them to uniformly fill the LSB plane of the entire original image. The parity-check-based refined authentication layer is used to divide the original image after thumbnail-based holographic redundancy embedding into [variables]. Blocks, and then subdivide each block into Perform parity checking on the sub-blocks; The image encryption module is used to divide the preprocessed original image into... The image blocks are processed and an encrypted lookup table is generated. The master encryption key is used to perform multiple rounds of encryption on each image block using an encryption algorithm based on dynamic difference space to generate an encrypted image. The key generation and distribution module is used to generate a watermark lookup table containing user information for each authorized user, combine the encrypted lookup table to generate a decryption lookup table, and distribute it to authorized users along with the master encryption key as the decryption key. Authorized users include: The receiving module is used to receive the decryption key issued by the image owner; The decryption and watermark embedding module is used to decrypt the image to be decrypted according to the decryption key, and at the same time complete the watermark embedding for the authenticated user terminal. The arbitrators include: The tamper detection and location module is used to check the parity of the image to be detected and determine the tamper location based on the refined authentication based on parity check. The tampered content recovery module is used to reconstruct the thumbnail from the untampered location to complete image recovery; The watermark extraction and user identity locking module is used to extract the differential signal by comparing the recovered image with the original preprocessed image, decipher the watermark sequence, and lock the user's identity.

Images