A blockchain-based secondary creation picture traceability and copyright protection method and system

Abstract

The application provides a blockchain-based secondary creation picture traceability and copyright protection method and system, which comprises the following steps: obtaining an original picture and a secondary creation picture thereof, and obtaining aligned reference data after preprocessing; performing frequency domain decomposition on the reference data, extracting editing traces according to the decomposition result, and performing saliency enhancement and original content interference suppression to generate a copyright fingerprint; generating a first key according to registration time information and a content identifier of the original brightness picture, performing encryption processing on the copyright fingerprint according to the first key, generating integrity check information according to the encryption result, and writing the integrity check information into a blockchain; extracting a feature vector of a current copyright state from the encrypted fingerprint data and converting the feature vector into a current state identifier, combining the first key and a traceability chain code of a historical copyright state to generate a traceability chain code of the current copyright state, and writing the traceability chain code into the blockchain. Through the method, the security of fingerprint encryption is enhanced, and the protection of copyright pedigree integrity is realized.

Description

Technical Field

[0001] This invention relates to the field of blockchain technology, specifically to a blockchain-based method and system for tracing the source and protecting the copyright of derivative works. Background Technology

[0002] With the widespread adoption of digital content creation tools and the booming development of social media platforms, derivative works based on original works have become an important part of the digital cultural ecosystem. Users continuously generate a large number of new digital works by cropping, color correcting, applying filters, sharpening, blurring, retouching, and compressing / exporting images across platforms. However, during the dissemination of these derivative works, the copyright ownership and creative lineage are often difficult to trace, leading to frequent copyright disputes. Therefore, establishing a verifiable and tamper-proof link between derivative works and their original counterparts has become a crucial issue that urgently needs to be addressed.

[0003] The existing technical solutions have obvious shortcomings in the following aspects:

[0004] (1) Traditional methods usually perform global hashing on the entire image or directly extract spatial difference features. The generated fingerprint is easily mixed with the semantic content of the original image, which poses a risk of information leakage. At the same time, when the image is compressed, scaled or partially edited, the stability of the fingerprint decreases significantly, making it difficult to meet the needs of practical application scenarios.

[0005] (2) Existing frequency domain fingerprint extraction techniques fail to effectively distinguish between the frequency domain response generated by the natural structure of an image (such as strong edges and textures) and the traces introduced by editing operations. If all frequency domain information is retained without distinction, the fingerprint will carry too much original image content and cannot accurately represent the operational characteristics of secondary creation.

[0006] (3) Existing encryption schemes mostly use fixed keys, which can easily lead to key reuse issues between different works registrations. In addition, the encryption process fails to bind the registration timestamp and the work source identifier at the same time, lacking the security constraints of the spatiotemporal dimension, making it difficult to resist replay attacks and source forgery.

[0007] (4) Conventional verification methods only perform independent hash calculations on single registration data, which cannot depict the derivative relationships and state transitions between different copyright statuses, and are difficult to support the tracing and integrity verification of the entire life cycle of a work in complex scenarios such as secondary creation and authorization transfer.

[0008] In summary, existing technologies have significant shortcomings in fingerprint privacy protection, key security management, chained representation of copyright lineage, and off-chain data integrity verification. There is an urgent need for an improved blockchain-based solution for tracing the source and protecting the copyright of derivative images. Summary of the Invention

[0009] In response, this invention provides a blockchain-based method and system for tracing the source and protecting the copyright of derivative works.

[0010] In a first aspect, the present invention provides a blockchain-based method for tracing the origin and protecting the copyright of derivative works, comprising:

[0011] Obtain the original image and its derivative image, perform content-aware preprocessing on both, and obtain aligned reference data, which includes the aligned derivative brightness map. Align the original brightness map with the shrinking effective area mask;

[0012] The reference data is decomposed in the frequency domain. Based on the decomposition results, the editing traces of the secondary creation image relative to the original image are extracted. The editing traces are then enhanced for saliency and the interference of the original content is suppressed to generate a copyright fingerprint.

[0013] Obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result and write it into the blockchain;

[0014] The feature vector of the current copyright status is extracted from the encrypted fingerprint data and converted into a current status identifier. The traceability chain code of the current copyright status is generated by combining the first key and the traceability chain code of the historical copyright status and written into the blockchain.

[0015] In some embodiments, after extracting the feature vector of the current copyright status based on the encrypted fingerprint data and converting it into a current status identifier, generating the traceability chain code of the current copyright status based on the first key and the traceability chain code of the historical copyright status, and writing it into the blockchain, the method further includes:

[0016] During verification, the integrity verification information and the traceability chain code of the current copyright status are read from the blockchain. The integrity verification information and the traceability chain code of the current copyright status are recalculated based on the data to be verified, and compared with the information on the blockchain. The verification is then determined based on the comparison result.

[0017] In some embodiments, obtaining the original image and its secondary creation image, performing content-aware preprocessing on both to obtain aligned reference data includes:

[0018] Scaling the original image and the derivative images to a uniform base resolution;

[0019] Spatial registration is performed on the scaled original image and the re-created image to obtain the aligned original image and the re-created image.

[0020] An effective region mask is generated based on the effective overlap range after spatial registration, and the effective region mask is then shrunk at the boundary to obtain a shrunk effective region mask.

[0021] The aligned original image and the aligned secondary creation image are converted to a unified brightness representation space to obtain the original brightness image and the secondary creation brightness image.

[0022] In some embodiments, generating a copyright fingerprint includes:

[0023] Multi-scale and multi-directional frequency domain decomposition is performed on the reference data to obtain the sub-band response of the secondary creation brightness map and the original brightness map at each scale and in each direction. The shrinkage effective region mask is scaled to the sub-band spatial size corresponding to each scale to obtain the effective calculation range of each sub-band.

[0024] Within the effective calculation range of each sub-band, the directional residuals are extracted by checking the response of each directional sub-band of the secondary creation brightness map through directional high-frequency residual filtering, and the editing traces are significantly enhanced based on the local kurtosis coefficient to obtain the residual significance map.

[0025] Within the effective calculation range of each sub-band, the original content local variance map is calculated based on the sub-band response in each direction of the original brightness map. A content suppression coefficient map is generated based on the original content local variance map. The content suppression coefficient map is used to suppress the residual significance map to obtain a content suppression residual significance map.

[0026] The content suppression residual significance maps of all scales and directions are unified in size and stitched together to form a copyright fingerprint.

[0027] In some embodiments, the steps of obtaining registration time information, generating a first key based on the registration time information and the content identifier of the original brightness map, encrypting the copyright fingerprint based on the first key, generating integrity verification information based on the encryption result, and writing it into the blockchain include:

[0028] Obtain registration time information and generate a first key based on the registration time information and the content identifier of the original brightness map;

[0029] The copyright fingerprint is encrypted using the first key to obtain encrypted fingerprint data;

[0030] A binary Merkle tree is constructed based on the encrypted fingerprint data to obtain the Merkle root commitment as integrity verification information, and the Merkle root commitment and the first key ciphertext after asymmetric encryption are written into the blockchain.

[0031] In some embodiments, the step of extracting the feature vector of the current copyright status based on the encrypted fingerprint data and converting it into a current status identifier, combining the first key and the historical copyright status traceability chaincode to generate the traceability chaincode of the current copyright status and writing it into the blockchain includes:

[0032] Complex amplitudes are taken from the encrypted fingerprint data and aggregated into a single-channel amplitude map. After two-dimensional Fourier transform, the data is divided into several frequency band sectors. The logarithmic energy of each sector is extracted and assembled into a feature vector, which is used to characterize the frequency domain distribution of editing traces in the current copyright status.

[0033] A fixed projection matrix is ​​generated based on the first key, and the feature vector is converted into a current state identifier based on the fixed projection matrix;

[0034] Generate a traceability chain code for the current copyright status based on the first key, the traceability chain code of the historical copyright status, and the current status identifier, and write the traceability chain code of the current copyright status into the blockchain.

[0035] In some embodiments, the step of constructing a binary Merkle tree based on the encrypted fingerprint data to obtain a Merkle root commitment as integrity verification information, and writing the Merkle root commitment and the first key ciphertext after asymmetric encryption into the blockchain includes:

[0036] The encrypted fingerprint data is serialized and segmented into multiple fingerprint data blocks;

[0037] Perform a secure hash calculation on each fingerprint data block to obtain the data block hash value;

[0038] Construct a binary Merkle tree from the hash values ​​of all fingerprint data blocks to obtain the Merkle root commitment;

[0039] Integrity verification information and the first key ciphertext are written into the blockchain.

[0040] In some embodiments, encrypting the copyright fingerprint using the first key to obtain encrypted fingerprint data includes:

[0041] Initialize the chaotic mapping based on the first key and generate a pseudo-random phase matrix;

[0042] By using the normalized copyright fingerprint tensor as the amplitude tensor and performing complex-domain phase rotation using a pseudo-random phase matrix, encrypted fingerprint data is obtained.

[0043] Secondly, this invention provides a blockchain-based system for tracing the source and protecting the copyright of derivative works, used to run the aforementioned blockchain-based method for tracing the source and protecting the copyright of derivative works, including:

[0044] The image alignment module is used to acquire the original image and its secondary creation image, perform content-aware preprocessing on the two, and obtain aligned reference data. The reference data includes the aligned secondary creation brightness image, the aligned original brightness image, and the shrinkage effective area mask.

[0045] The copyright fingerprint generation module is used to perform frequency domain decomposition on the reference data, extract the editing traces of the secondary creation image relative to the original image based on the decomposition results, and perform saliency enhancement and original content interference suppression on the editing traces to generate a copyright fingerprint.

[0046] An encrypted fingerprint data and integrity verification information generation module is used to obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result, and write it into the blockchain;

[0047] The traceability chain code generation module is used to extract the feature vector of the current copyright status based on the encrypted fingerprint data and convert it into the current status identifier, combine the first key and the traceability chain code of the historical copyright status to generate the traceability chain code of the current copyright status and write it into the blockchain.

[0048] In some embodiments, the system further includes a verification module, which, during verification, reads the integrity verification information and the traceability chain code of the current copyright status from the blockchain, recalculates the integrity verification information and the traceability chain code of the current copyright status based on the data to be verified, compares them with the information on the blockchain, and determines whether the verification passes based on the comparison result.

[0049] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages:

[0050] (1) This invention uses local kurtosis coefficient to adaptively distinguish between natural textures and editing traces, so that heavy tail areas such as compressed block boundaries and sharpened ringing receive higher response weights, thus solving the problem of easily confusing the original image content with editing traces in traditional methods.

[0051] (2) The present invention generates a content suppression coefficient by using the local variance of the sub-band image of the original work, and performs pixel-by-pixel suppression on the residual significance map, thereby reducing the fingerprint contribution of strong texture regions and avoiding the leakage of the original work's visual information by copyright fingerprints.

[0052] (3) The present invention uses the on-chain registration timestamp and the original work perception hash to derive the first key, so that the encryption key is simultaneously bound to the registration time and the source of the work, preventing key reuse and enhancing encryption security.

[0053] (4) This invention uses chaotic recursive gating to irreversibly bind the current state to the previous chain code, generating chain code that depends on the complete genealogical history, so that single-point tampering propagates along the link and causes all subsequent chain code verification to fail, thereby achieving copyright genealogical integrity protection. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0055] Figure 1 A schematic diagram illustrating a blockchain-based method for tracing the source and protecting the copyright of derivative works, provided in an embodiment of the present invention.

[0056] Figure 2 A data flow and system functional architecture diagram provided for an embodiment of the present invention;

[0057] Figure 3 A flowchart of a method for extracting fingerprints from a secondary creation image provided in an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of a fingerprint encryption and integrity verification information generation process provided in an embodiment of the present invention;

[0059] Figure 5 A schematic diagram of a binary Merkle tree provided for an embodiment of the present invention;

[0060] Figure 6 This is a schematic diagram of a traceability chain code generation method provided in an embodiment of the present invention;

[0061] Figure 7 This is a schematic diagram of the binary coefficient distribution of a partially fixed projection matrix provided in an embodiment of the present invention;

[0062] Figure 8 A current state identifier bitmap in the form of black and white stripes is provided as an embodiment of the present invention;

[0063] Figure 9 A bitmap for comparing the reordering of code in a pre-sequence tracing chain is provided in this embodiment of the invention;

[0064] Figure 10 Scatter plot of normalized Hamming distance between the recalculated traceability chaincode and the on-chain record traceability chaincode of multiple test cases provided in the embodiments of the present invention;

[0065] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0067] Original image: refers to digital content that is created for the first time and has originality and integrity, including but not limited to original photographs, design drafts or illustrations, which serve as the basis for subsequent secondary creations.

[0068] Secondary creative images: Based on original photographic works, design drafts, or illustrations, new creative images are continuously generated through operations such as cropping, color correction, adding filters, sharpening, blurring, local retouching, and multiple compression and exporting between different platforms.

[0069] Effective region mask: refers to a binarized mask used to mark the difference between the secondary creation image and the original image. Figure 2 Each region has a valid pixel location. In the mask, a region with a value of 1 indicates that the location participates in subsequent fingerprint extraction, while a region with a value of 0 indicates that the location does not participate in fingerprint extraction due to cropping, missing pixels, or boundary effects. In this invention, the valid region mask uses... express.

[0070] Morphological erosion: A fundamental morphological operation in digital image processing used for boundary manipulation of binary images. Its core operation can be described as: using a structuring element of fixed shape and size (such as...) A square kernel slides across the binary image. The pixel corresponding to the center point of the structuring element is kept as 1 if and only if the structuring element completely covers the white area (the area with a value of 1); otherwise, the pixel is set to 0. For the mask, the erosion effect is to shrink the boundary of the effective area with a value of 1 inward.

[0071] Controllable Oriented Pyramid: A multi-scale, multi-directional image frequency domain decomposition tool that decomposes an image into sub-band responses at multiple scales (frequency levels) and in multiple directions (angles). It is approximated by combining a Laplacian pyramid with a directional filter bank, or by using complex wavelets, directional wavelets, Gabor directional filter banks, etc., enabling the separation of frequency components at different scales and directions.

[0072] Directional high-frequency residual filter kernel: This refers to a two-dimensional filter kernel corresponding to a specific scale and direction, used to enhance local frequency domain variations in a specified direction and scale. It can be implemented using a directional Gaussian difference kernel, a directional second-order difference kernel, or a directional Gabor high-pass kernel.

[0073] Secure hashing: refers to the process of mapping input data of arbitrary length to a fixed-length hash value through a one-way hash function. It is irreversible and is used to generate keys and build Merkle trees.

[0074] Pseudo-random phase matrix: refers to a multidimensional matrix with the same size as the fingerprint tensor, generated by the chaotic mapping driven by the first key, whose elements are from 0 to... A range of pseudo-random phase values ​​are used to perform complex-domain phase rotation encryption on the fingerprint tensor.

[0075] Complex-domain phase rotation encryption: refers to an encryption method that uses the fingerprint tensor in the real domain as a complex amplitude and performs element-wise complex phase rotation on it using a pseudo-random phase matrix.

[0076] Merklegen commitment: This refers to serializing the encrypted fingerprint data, dividing it into multiple data blocks, constructing a binary Merkle tree, and writing the root node hash value as the integrity verification information of the encrypted fingerprint into the blockchain to verify whether the encrypted fingerprint stored off-chain has been tampered with.

[0077] Bit permutation index: refers to the index from 0 to 1 generated by the iterator key. A non-repeating random permutation sequence is used to specify the rearrangement order of bits in the preorder tracing chain code.

[0078] The following example demonstrates the entire process of this invention, from secondary creation image registration to closed-loop verification, through a complete application scenario.

[0079] Photographer Zhang created a photograph on March 15, 2024. A forest landscape photo (including leaf textures and tree trunk edges). On August 20, 2024, designer Li downloaded and cropped the image (cropping the top left 224×224 area), increased the contrast, and lightly sharpened the tree trunk edges. The exported image has the following dimensions: They intend to register the copyright for secondary creations.

[0080] like Figure 1 , Figure 2 As shown in the figure, this invention proposes a blockchain-based method for tracing the source and protecting the copyright of derivative works, including the following steps:

[0081] S1. Obtain the original image and its secondary creation image, perform content-aware preprocessing on both, and obtain aligned reference data, including the aligned secondary creation brightness map. Align the original brightness map with the shrinking effective area mask.

[0082] This step involves obtaining the original image and its derivative image, performing content-aware preprocessing on both, and obtaining the aligned brightness map of the derivative image. The original brightness map, along with the mask for the shrunken effective region, provides spatially consistent and brightness-uniform input data for subsequent frequency domain decomposition and editing trace extraction. Specifically, this includes the following sub-steps:

[0083] S11 scales the original image and derivative images to a uniform base resolution.

[0084] This sub-step is used to eliminate frequency band position shifts caused by size differences.

[0085] In practice, the secondary creation image is obtained. and the original image If the two dimensions are different, scale their shorter sides to the same target pixel value (e.g., scale them both to 1024 pixels) while keeping their aspect ratios unchanged. Use bicubic interpolation for scaling. If the two dimensions are the same, skip this step.

[0086] In this embodiment, the original image size Size of derivative works The system scales the shorter sides of both images to 1024 pixels, while keeping the original image unchanged (the shorter side is now 1024). The re-created image is then processed using bicubic interpolation from... Stretch to The content of the image is magnified evenly.

[0087] S12, perform spatial registration on the scaled original image and the secondary creation image to obtain the aligned original image and secondary creation image.

[0088] This sub-step is used to eliminate geometric offsets caused by translation, rotation, clipping, or perspective changes, so that both are in the same spatial coordinate system, ensuring that the same spatial position corresponds to the same semantic region during subsequent frequency domain decomposition.

[0089] In practice, scale-invariant or speed-up robust feature points are extracted from both the scaled original image and the re-creation image. Candidate matching point pairs are obtained through feature descriptor matching, and erroneous matching point pairs are eliminated using a random sampling consensus algorithm. The geometric transformation matrix is ​​then estimated. If only translation, scaling, and slight rotation exist between the two images, affine transformation matrix is ​​used for registration; if there are changes in viewpoint or perspective distortion, homography transformation matrix is ​​used for registration; if there is local stretching or local modification, optical flow thinning is used to correct local displacement after global registration, resulting in an aligned re-creation image. Align the original image .

[0090] After alignment, if the effective pixel range of the secondary creation image is smaller than the size of the original image, the pixel positions outside the effective range are filled with a preset pixel value (e.g., 0) to make it have the same spatial size as the original image. Subsequently, this filled area is marked as invalid (with a value of 0) by the effective area mask in step S13 and does not participate in subsequent fingerprint extraction, thereby avoiding the spurious frequency domain response generated by the filled pixels from polluting the copyright fingerprint.

[0091] S13. Generate an effective region mask based on the effective overlap range after spatial registration, and perform boundary shrinking on the effective region mask to obtain a shrunken effective region mask.

[0092] This sub-step is used to mark the locations where both have valid pixels, preventing cropping boundaries or missing areas from being mistakenly identified as editing marks.

[0093] In practice, the original image size is used as a reference, and the projection range of the secondary creation image in the original coordinate system is calculated according to the transformation matrix. The intersection area is marked as the effective area mask. The positions with a value of 1 are included in the subsequent fingerprint extraction, while the positions with a value of 0 are not included.

[0094] In this embodiment, a binary effective region mask is generated based on the registered overlap range. The 224×224 area in the upper left corner (filled area) is marked as 0 (invalid), and the remaining overlapping areas are marked as 1 (valid).

[0095] Subsequently, the effective area mask was applied. Perform morphological etching operation, and determine the etching core size. The effective shrinkage area mask is obtained. If the registration error is large, the size of the etched core can be increased to [a larger value]. If the registration accuracy is high, it can be reduced to... Subsequent frequency domain decomposition content suppression and fingerprint tensor construction both occur within a shrinking effective region mask. Execute within the corresponding region.

[0096] In this embodiment, the following is adopted: The effective area of ​​nuclear corrosion is masked to obtain The effective region boundary is shrunk inward by 3 pixels to eliminate artifact interference that may be caused by registration and interpolation boundaries.

[0097] S14. Convert the aligned original image and the aligned secondary creation image to a unified brightness expression space to obtain the original brightness image and the secondary creation brightness image.

[0098] This sub-step is used to reduce the impact of color grading and encoding format differences on frequency domain trace extraction, allowing subsequent processing to focus on structural changes rather than color changes.

[0099] In practice, the aligned secondary creation image will be used. Align the original image Convert to a uniform brightness representation space. If the input is a color image, convert to... Color space and extract aisle: ; If the input is a grayscale image, the grayscale values ​​are directly used as the brightness values. The brightness values ​​are normalized to the range of 0 to 1 to obtain the secondary creation brightness map. and the original brightness map .

[0100] The output of steps S11 to S14 above is: Aligned secondary creation brightness map Align the original brightness map and shrinking effective area mask Together, they constitute the standardized registration data (in this embodiment, the dimensions of all three are...). ).

[0101] S2, perform frequency domain decomposition on the reference data, extract the editing traces of the secondary creation image relative to the original image based on the decomposition results, and perform saliency enhancement and original content interference suppression on the editing traces to generate a copyright fingerprint.

[0102] Secondary creative operations typically manifest as localized directional high-frequency changes in the frequency domain. For example, sharpening creates directional ringing, heavy compression creates blocky boundary responses, blurring causes high-frequency energy attenuation, and local retouching creates localized discontinuous residuals. However, the strong edges, fabric textures, hair strands, leaves, and periodic patterns of the original image itself can also produce strong frequency domain responses. If the natural content response and the editing operation response are not distinguished, the extracted creative fingerprint may contain semantic information from the original image.

[0103] This step involves performing frequency domain decomposition on the reference data, extracting editing traces of the secondary creation image relative to the original image, and generating a copyright fingerprint after saliency enhancement and original content interference suppression. For example... Figure 3 As shown, the specific steps include the following:

[0104] S21, perform multi-scale and multi-directional frequency domain decomposition on the reference data to obtain the sub-band response of the secondary creation brightness map and the original brightness map at each scale and in each direction, and scale the shrinkage effective region mask to the sub-band spatial size corresponding to each scale to obtain the effective calculation range of each sub-band.

[0105] This sub-step decomposes the secondary creation brightness map and the original brightness map into sub-band responses of multiple scales and multiple directions, so that different types of editing traces can be extracted in the corresponding scales and directions, and at the same time, the effective regions of each sub-band are generated, which limits the calculation range for subsequent feature statistics.

[0106] The specific implementation includes the following steps:

[0107] 1. A controllable directional pyramid was used to adjust the brightness of the secondary creation image. and the original brightness map Decompose the image to obtain a secondary creation sub-band image. and original sub-band diagram ,in For scale indexing, For direction index.

[0108] In this embodiment, the total number of scales ( ), total number of directions ( A total of 32 sub-band maps were obtained. The controllable directional pyramid can be composed of a Laplace pyramid and a directional filter bank, or it can be approximated by a complex wavelet or Gabor directional filter bank. First, the brightness map is bandpass decomposed, and then the directional response is extracted using filters with different directional angles.

[0109] 2. The secondary creation sub-band image and the original sub-band image are matched one-to-one according to (s,d), and stored in a fixed order of "scale first, then direction" to ensure that the channel order is consistent during subsequent splicing.

[0110] 3. For each scale The effective area mask will be shrunk. By scaling to the corresponding subband spatial size using nearest neighbor interpolation, the effective subband shrinkage mask at that scale is obtained. For direction index The effective area mask of this sub-band shrinkage The effective calculation range of all directional subbands under the same scale is jointly constrained. That is, the position with a value of 1 indicates that the corresponding position in the subband response of that scale and direction participates in the subsequent feature statistics, and the position with a value of 0 does not participate.

[0111] Through the above processing, the subband response at each scale and in each direction obtains a shrinkage effective region mask consistent with its spatial resolution, so as to obtain the effective calculation range of the subband, which is used to eliminate the interference of invalid regions on subsequent kurtosis statistics, variance statistics and fingerprint stitching.

[0112] S22, within the effective calculation range of each sub-band, the directional residuals are extracted by verifying the response of each directional sub-band of the secondary creation brightness map through directional high-frequency residual filtering, and the editing traces are significantly enhanced according to the local kurtosis coefficient to obtain the residual significance map.

[0113] This sub-step is used to extract edit trace residuals in subbands of each direction, and to adaptively enhance edit traces and suppress natural textures through local kurtosis coefficients. The residual distribution of natural textures approximates a Gaussian distribution (kurtosis ≤ 3), while edit traces such as compressed block boundaries and sharpening ringing exhibit a heavy-tailed distribution (kurtosis significantly greater than 3) with a small number of strong residuals and a large number of weak residuals, which can be used to distinguish between the two.

[0114] The specific implementation includes the following steps:

[0115] S221: Within the effective calculation range of each sub-band, construct a directional high-frequency residual filter kernel and extract non-negative directional residual response maps from the directional sub-band responses of the secondary creation brightness map.

[0116] Specifically, masking is applied to the effective shrinkage area of ​​each sub-band. Within the limited effective calculation range, for each secondary creation sub-image Construct a high-frequency residual filter kernel corresponding to this scale and direction. According to the direction index Determine the direction angle ( By sequentially selecting 0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, and 157.5°, a Gaussian smoothing kernel stretched along this direction and a lateral smoothing kernel perpendicular to this direction are generated and subtracted to obtain the directional high-frequency residual filter kernel. Core size Or 9×9.

[0117] Directional high-frequency residual filter kernel With sub-band diagram Perform two-dimensional convolution (using mirror padding for boundary processing), and obtain the non-negative direction residual response map after taking the absolute value. ,in The position corresponding to the value of 0 Set the value to 0.

[0118] It should be noted that since this filter kernel is obtained by subtracting the smoothing kernel in the target direction from the kernel in its perpendicular direction, its response intensity reflects the frequency domain difference between the target direction and the perpendicular direction. For direction-specific editing traces (such as ringing or block boundaries extending along a specific direction), this difference is significant, and the filter response is strong; for isotropic natural textures, the responses in the two directions are similar, and the filter responses cancel each other out and are suppressed.

[0119] S222: Within the effective calculation range of each sub-band, a local statistical window is established with each position in the non-negative direction residual response map as the center, and the local kurtosis coefficient of the response value distribution within the window is calculated to characterize whether the local region has heavy-tailed distribution characteristics.

[0120] Specifically, using the residual response diagram in the non-negative direction Each position Establish for the center The local kurtosis calculation window is used to calculate the ratio of the fourth-order central moment to the square of the variance of the response within the window, thus obtaining the local kurtosis coefficient. :

[0121]

[0122] in, Represents the local kurtosis calculation window Internal non-negative direction residual response diagram The set of response values, This represents the row index in the sub-band graph. This represents the column index in the sub-band diagram. , Representing windows respectively The mean and standard deviation of the residual response values ​​in the non-negative direction.

[0123] when When the value is less than or equal to 3, the residual distribution in this local region approximates a Gaussian distribution, corresponding to natural textures or a stable background; when... When the value is greater than 3, there are a small number of strong residual responses in this local area, corresponding to the boundaries of compressed blocks, sharpened ringing, or local modification boundaries.

[0124] S223: Within the effective calculation range of each sub-band, generate a local kurtosis significance weighted map based on the local kurtosis coefficient.

[0125] Based on local kurtosis coefficient Generate a local kurtosis significance weighted map :

[0126]

[0127] in This represents the kurtosis gain coefficient. This represents the upper limit of the significance weight of local kurtosis. Preferably, Take 0.5, Take 3.

[0128] When the local kurtosis coefficient is less than or equal to 3, the weight is 1 (no additional enhancement is given to natural texture areas); when it is greater than 3, the weight increases as the local kurtosis coefficient increases (editing trace areas are magnified).

[0129] S224: Perform nonlinear enhancement on the non-negative direction residual response map, and then multiply it element-by-element with the local kurtosis significance weight map to obtain the residual significance map.

[0130] Specifically, for the residual response diagram in the non-negative direction Perform nonlinear enhancement and multiply element-wise with the local kurtosis significance weight map to obtain the residual significance map. :

[0131]

[0132] in, This represents the residual nonlinear enhancement index, used to amplify the intensity difference between weak editing traces, with a value ranging from 1.2 to 1.6; This indicates element-wise multiplication.

[0133] This processing allows for a more pronounced response to compression block effects, sharpening ringing, local filtering boundaries, and local modification boundaries, while maintaining a lower response for smooth backgrounds and natural continuous textures.

[0134] In this embodiment, taking a sub-band with a scale index of 2 and an orientation of 0° (horizontal direction) as an example, in the natural texture area of ​​leaves, the residuals approximate a Gaussian distribution, with local kurtosis coefficients. (less than 3), local kurtosis significance weight (No enhancement). Directional ringing (heavy-tailed distribution) is present at the sharpened tree trunk edges. Weight (Significantly enhanced).

[0135] S23, within the effective calculation range of each sub-band, calculate the original content local variance map based on the sub-band response of each direction of the original brightness map, generate a content suppression coefficient map based on the original content local variance map, and use the content suppression coefficient map to perform content suppression on the residual significance map to obtain a content suppression residual significance map.

[0136] Strong edges and complex textures in the original image also generate significant responses in the directional subbands, and these responses do not necessarily originate from secondary creation operations. If these responses are directly preserved, the secondary creation fingerprint tensor may carry semantic information from the original image.

[0137] This sub-step is used to calculate the local variance of the original content from the original image sub-band image, and generate a continuous content suppression coefficient based on the local variance of the original content, thereby reducing the fingerprint contribution of regions with strong natural textures.

[0138] The specific implementation includes the following steps:

[0139] S231: Within the effective calculation range of each sub-band, calculate the local variance map of the original content based on the sub-band response in each direction of the original brightness map.

[0140] Specifically, masking is applied to the effective shrinkage area of ​​each sub-band. Within the limited effective computational range, for each original sub-band map , for each position Establish a local variance calculation window centered on the local variance. For example, the window size is The local variance of the subband response within the calculation window is first calculated in the local variance calculation window. The mean of the sub-band response of the original brightness map is calculated, and then the average of the squared differences between the response of each sub-band within the window and the mean is calculated. This average is then used as the position value. The original content local variance value is used to obtain the original content local variance map. .in, Positions with a value of 0 are not included in the calculation, and the corresponding variance value is set to 0.

[0141] Local variance plot of original content This refers to the original graph in the 1st... The first scale and the first The local undulation intensity map in each direction indicates that the larger the local variance value of the original content, the more obvious the natural texture or semantic edge of the original image is near that location; the smaller the local variance value of the original content, the closer the location is to a flat area or a weak texture area.

[0142] S232: Normalize the original content local variance plot and generate a content suppression coefficient plot using an exponential decay method.

[0143] Specifically, for the local variance plot of the original content Perform normalization to generate a normalized local variance plot of the original content. A content suppression coefficient map was generated using an exponential decay method. :

[0144]

[0145] in Take 2.0, The value ranges from 0 to 1, with smaller values ​​indicating stronger inhibition.

[0146] S233: Multiply the residual significance map and the content suppression coefficient map element by element to obtain the content suppression residual significance map.

[0147] Specifically, the residual significance plot Content suppression coefficient graph Element-wise multiplication yields the significance plot of content-suppressed residuals. :

[0148]

[0149] In this embodiment, for the sub-band with a scale index of 2 and an orientation of 0° (horizontal direction), the normalized local variance is calculated in the region of strongly textured leaves. Content suppression coefficient (Strong suppression to prevent the original leaf texture from entering the fingerprint). Normalized local variance in flat sky areas. Suppression coefficient (Almost preserved).

[0150] S24, unifies the size of the content suppression residual significance map across all scales and directions and stitches them together to form a copyright fingerprint.

[0151] This sub-step is used to unify the size of the content suppression residual significance maps of all scales and orientations and stitch them together to form a copyright fingerprint tensor.

[0152] In practice, the residual significance map for each content will be suppressed. Use bicubic interpolation to uniformly interpolate to size( , (Height and width of the original image) Set the outer position to 0. Following the fixed "scale first, direction second" order in step S21, concatenate all uniform-scale feature maps along the channel dimension to obtain the copyright fingerprint tensor. The size is .

[0153] Specific ( ), It has 32 channels, with channels 1 through 8 corresponding to the 8 directions at the first scale, and so on. Finally, for... Perform amplitude normalization, normalizing all channel values ​​to the range of 0 to 1, to obtain the normalized copyright fingerprint tensor. This is used for subsequent encryption processing.

[0154] The output of steps S21 to S24 above is a normalized copyright fingerprint tensor. Each channel carries editing traces under a combination of scale and direction, and the response of strong semantic texture regions has been suppressed, thereby reducing the risk of copyright fingerprints leaking the original image content.

[0155] In this embodiment, the significance of the content suppression residuals of 32 enhanced and suppressed images are interpolated back. Stacking them in scale-direction order yields the normalized copyright fingerprint tensor. The size is .

[0156] S3, obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result and write it into the blockchain.

[0157] This step generates a first key based on the registration time information and the content identifier of the original brightness map, encrypts the copyright fingerprint to hide its spatial structure, and then generates integrity verification information based on the encrypted fingerprint data and writes it to the blockchain. For example... Figure 4 As shown, the specific steps include the following:

[0158] S31, Obtain registration time information and generate a first key based on the registration time information and the content identifier of the original brightness map.

[0159] This sub-step is used to bind the first key to both the registration time and the original graph source, preventing key reuse.

[0160] The specific implementation includes the following steps:

[0161] S311: Obtain the on-chain registered timestamp and convert it into a uniform string format.

[0162] Specifically, obtain the on-chain registration timestamp. On-chain registration timestamp This can be obtained from the block time, transaction time, or time recorded within the contract of the blockchain system. The precision can be in the second or millisecond range, with millisecond precision being preferred. This precision is then converted into a uniform string format (such as a Coordinated Universal Time (UTC) millisecond timestamp string). This string format needs to be fixed in the algorithm version to ensure consistency between the key derivation results during the registration and verification phases.

[0163] S312: Calculate the perceptual hash of the original brightness map.

[0164] First, the original brightness map Scaling to The grayscale image is then subjected to a discrete cosine transform, resulting in a coefficient matrix of the same size. The coefficients in this matrix are arranged in ascending order of frequency.

[0165] Top left corner: Low-frequency coefficients, corresponding to the overall brightness, outline, and slowly changing areas of the image;

[0166] Bottom right corner: High-frequency coefficients, corresponding to image details, edges, and rapidly changing areas.

[0167] Extract the top left corner In the low-frequency coefficient region, binarization is performed using the median value of this region as a threshold: coefficients above the median value are recorded as 1, and coefficients below or equal to the median value are recorded as 0, resulting in a 64-bit binary sequence as a perceptual hash. This method is tolerant of slight compression, scaling, and brightness variations, while still maintaining the overall visual structure of the original brightness map.

[0168] S313: Concatenate the three factors and perform a secure hash calculation to generate the first key.

[0169] Register timestamps on the blockchain Original brightness map perceptual hash and algorithm version number Concatenate the following data in a fixed format (e.g., "algorithm version number | on-chain registration timestamp | original brightness map perceptual hash"), input it into a 256-bit secure hash function, and obtain the first key. .

[0170] Understandably, since the first key depends on the registration time, the content of the original brightness map, and the algorithm version, different registration times or different original brightness maps will generate different first keys, thereby reducing the risk of key reuse.

[0171] S32, the copyright fingerprint is encrypted using the first key to obtain encrypted fingerprint data.

[0172] This sub-step is used to hide the spatial structure of the copyright fingerprint through chaotic phase rotation, while preserving its amplitude statistical characteristics, so that nodes without the key cannot recover the spatial distribution of editing traces.

[0173] The specific implementation includes the following steps:

[0174] S321: Initialize the chaotic mapping based on the first key and generate a pseudo-random phase matrix.

[0175] According to the first key Initialize the chaotic mapping, generate and normalize the copyright fingerprint tensor. Uniformly sized pseudo-random phase matrix Its element values ​​range from 0 to 1. .

[0176] Specifically, the first key The binary data is split into multiple binary segments. Two segments are normalized to the range of 0 to 1 and used as the initial state for the coupled Logistic mapping. Chaos control parameters are then set. The subsequent iterations generate a chaotic sequence, discard the first 100 iterations to reduce the correlation of the initial state, and then fill the subsequent sequences to the target size in a fixed order.

[0177] The chaotic mapping can be implemented using Logistic mapping, Henon mapping, or Tent mapping; alternatively, a cryptographically secure pseudo-random number generator can be used instead of chaotic mapping, provided that the same first key is maintained. Same algorithm version number Able to deterministically generate the same pseudo-random phase matrix That's all.

[0178] S322: Using the normalized copyright fingerprint tensor as the amplitude tensor, perform complex-domain phase rotation using a pseudo-random phase matrix to obtain encrypted fingerprint data.

[0179] Normalized copyright fingerprint tensor As amplitude tensor .because The amplitude has been nonnegated and normalized, and can be used directly as a complex amplitude value.

[0180] Using pseudo-random phase matrix Perform a phase rotation in the complex field to obtain the encrypted fingerprint data. :

[0181]

[0182] in, For amplitude tensors, The imaginary unit, For element-wise multiplication, the pseudo-random phase matrix The value range is from 0 to 2.

[0183] Understandably, encrypted fingerprint data It is a complex fingerprint tensor encrypted with chaotic phase, with a size of This is used for secure storage in authorized nodes and for subsequent on-chain integrity commitments and traceable chaincode generation. Due to the pseudo-random phase matrix... The value range is from 0 to 2. Coverage from 0 to The phase angle range. This phase rotation process does not change the fingerprint amplitude tensor. The amplitude value is shown, but the spatial phase structure of the fingerprint tensor is hidden.

[0184] This operation only changes the phase information without changing the amplitude value, thus preserving the intensity distribution of the edit traces while hiding the spatial structure of the fingerprint. Nodes that do not hold the first key cannot reproduce the pseudo-random phase matrix and therefore cannot recover the spatial location information of the edit traces.

[0185] S33, construct a binary Merkle tree based on the encrypted fingerprint data to obtain the Merkle root commitment as integrity verification information, and write the Merkle root commitment and the first key ciphertext after asymmetric encryption into the blockchain.

[0186] This sub-step is used to make an integrity commitment to the cryptographic fingerprint, so that if the cryptographic fingerprint data stored off-chain is tampered with, it can be detected, while avoiding the storage burden caused by writing the complete cryptographic fingerprint data on-chain.

[0187] The specific implementation includes the following steps:

[0188] S331: Serialize the encrypted fingerprint data and divide it into multiple fingerprint data blocks.

[0189] Specifically, encrypted fingerprint data The real and imaginary parts are arranged in a 32-bit floating-point format according to a fixed order of "channel, row, column" and a fixed byte order (such as little-endian) to obtain the encrypted fingerprint byte sequence. Calculate based on a fixed single-block byte length (e.g., 4096 bytes or 8192 bytes). Divided into multiple fingerprint data blocks If the length of the last data block is less than the fixed single-block byte length, it is padded with zeros to the fixed length before being included in the hash.

[0190] Serialization rules and zero-padding rules need to be specified in the algorithm version number. The byte sequence and block results are fixed in the middle to ensure that different computing environments obtain consistent byte sequences and block results.

[0191] S332: Perform a secure hash calculation on each fingerprint data block to obtain the data block hash value.

[0192] Specifically, the fingerprint data block Convert the data into a deterministic byte sequence, then input it into a 256-bit secure hash function to obtain the data block hash value. .

[0193] S333: Construct a binary Merkle tree from the hash values ​​of all fingerprint data blocks to obtain the Merkle root commitment.

[0194] Specifically, first, according to the fingerprint data block index The order in which the data block hash values ​​are processed. The hash value of the leaf node is then used as the basis for concatenating the hash values ​​of two adjacent data blocks and performing a 256-bit secure hash again to obtain the hash value of the parent node. If the number of nodes in a certain layer is odd, the last node is copied to participate in the pairing. This process is repeated until a unique root node is obtained, which is the Merklegen commitment. The Merkelgen pledged to provide encrypted fingerprint data. The integrity verification information is such that any tampering with any data block will lead to a recalculation of the Merkle root and the on-chain evidence. Inconsistent. See also Figure 5 This shows an example of a binary Merkle tree.

[0195] S334: Write the integrity verification information and the first key ciphertext into the blockchain.

[0196] Specifically, using the copyright owner's public key For the first key Perform asymmetric encryption (such as elliptic curve integration encryption schemes) to obtain the first key ciphertext. Nodes that do not possess the corresponding private key cannot recover the first key, and therefore cannot reproduce the pseudo-random phase matrix.

[0197] Merkelgen's promise First key ciphertext On-chain registration timestamp Algorithm version number and secondary creation image logo Write the encrypted fingerprint data into a blockchain smart contract. The data is securely stored off-chain by authorized nodes. Through this process, any tampering with a data block will result in a recalculation of the Merkle root and the on-chain notarization. Inconsistencies are detected during the verification phase.

[0198] S4. Extract the feature vector of the current copyright status based on the encrypted fingerprint data and convert it into the current status identifier. Combine the first key and the traceability chain code of the historical copyright status to generate the traceability chain code of the current copyright status and write it into the blockchain.

[0199] This step involves extracting the feature vector of the current copyright status from the encrypted fingerprint data and converting it into a current status identifier. Then, using chaotic recursive gating, the current status identifier is irreversibly bound to the traceability chaincode of previous historical copyright statuses. This generates a traceability chaincode that depends on the entire genealogy and writes it to the blockchain. Through this recursive traceability chaincode generation method, any change in historical copyright status propagates along the genealogy, thus achieving chain-based traceability. Figure 6 As shown, the specific steps include the following:

[0200] S41, take the complex amplitude of the encrypted fingerprint data and aggregate it into a single-channel amplitude map. After two-dimensional Fourier transform, divide it into several frequency band sectors, extract the logarithmic energy of each sector and assemble it into a feature vector, which is used to characterize the frequency domain distribution of editing traces in the current copyright status.

[0201] This sub-step is used to extract a fixed-length feature vector from the cryptographic fingerprint, representing the distribution of editing trace intensity at different frequencies and directions in the current copyright status.

[0202] In specific implementation, the first... Encrypted fingerprint data of secondary copyright status After taking the complex amplitude, the amplitude of all channels is averaged and aggregated to obtain the single-channel amplitude map. For single-channel amplitude plots Perform a two-dimensional Fourier transform and shift the zero frequency to the center to obtain the frequency domain amplitude diagram. Taking the center of the spectrum as the origin, Divided into A logarithmic polar coordinate frequency band sector. The amplitude value within each sector is squared and then processed. Compress and average to obtain the eigenvalues. Arranged in a fixed order as 3D feature vector In this embodiment, the frequency radius is divided into 6 rings at logarithmic intervals, and the direction angle is evenly divided into 12 sectors, totaling... Each sector.

[0203] S42, Generate a fixed projection matrix based on the first key, and convert the feature vector into a current state identifier based on the fixed projection matrix.

[0204] This sub-step is used to project continuous feature vectors into fixed-length binary state identifiers.

[0205] In specific implementation, the feature vector Perform mean centering and standard deviation normalization to obtain standardized eigenvectors. .

[0206] According to the first key and algorithm version number Generate a fixed projection matrix Specifically, using the first key and algorithm version number Together they serve as the seed for a cryptographically secure pseudorandom number generator, generating A fixed projection matrix is ​​obtained by mapping random bits 0 to -1 and random bits 1 to +1. (size ,in (The number of bits for the status identifier, set to 256).

[0207] Fixed projection matrix With standardized feature vectors After multiplication, check the sign of each element: record 1 for greater than or equal to 0, and 0 for less than 0, to obtain the current state identifier. This process essentially involves partitioning the feature space using multiple random hyperplanes determined by the first key; similar editing lineages will yield state identifiers with smaller Hamming distances. See also... Figure 7 This diagram illustrates the binary coefficient distribution of a portion of the fixed projection matrix. Matrix elements take values ​​of +1 (red) or -1 (blue), deterministically generated by the first key and algorithm version number. Random projection converts continuous feature vectors into binary state identifiers. The horizontal axis represents the frequency band sector index, and the vertical axis represents the projection vector index.

[0208] See Figure 8 The current state identifier bitmap is displayed in black and white stripes. The bit length is the same as the number of bits in the traceability chain code, and the horizontal axis is the bit position index.

[0209] S43, Generate the traceability chain code of the current copyright status based on the first key, the traceability chain code of the historical copyright status and the current status identifier, and write the traceability chain code of the current copyright status into the blockchain.

[0210] This sub-step is used to bind the current state identifier to the preceding recursive tracing chain code, forming a recursive tracing chain code that depends on the complete lineage history.

[0211] In practical implementation, if This indicates that the current state is the initial registration state of the copyright lineage, and the current state identifier is obtained by directly extracting and projecting the current copyright status data. As the initial traceability chaincode Write it to the blockchain. If First, the traceability chain code of the previous historical copyright status is read from the blockchain. ; the first key As the key for the HMAC-SHA256 algorithm, the copyright status index is used. With preceding source chaincode Concatenate the data into a fixed format as the message to be authenticated, input it into the HMAC-SHA256 function, and generate an iterative subkey. .

[0212] Subsequently, using the iterative subkey As the seed, a pseudo-random number generator is used to generate a number of lengths. The random permutation is used as the bit permutation index; or, the iterative subkey is used. The initial state of the chaotic mapping is generated with a length of... The chaotic sequence is sorted numerically in ascending order, and its original index order is used as the bit permutation index. The source chain code of the previous historical copyright status is then rearranged according to this bit permutation index. The bit sequence is used to obtain the traceability chain code of the historical copyright status before rearrangement. .

[0213] Finally, the current state identifier will be displayed. and XORing the bits yields the source code for the current copyright status. :

[0214]

[0215] The source code of the current copyright status Write it to the blockchain. If the encrypted fingerprint data or state identifier of any historical state is tampered with, it will recursively propagate and change all subsequent traceability chaincode, thus being detected during verification.

[0216] See Figure 9 The diagram shows a comparison of the rearrangement of the preorder tracing chain code, comparing the bit arrangement of the preorder tracing chain code in its original order with that after chaotic rearrangement, as well as the bit difference positions between the two. The horizontal axis of the three sub-graphs represents the bit position index.

[0217] In one alternative implementation, after step S4, the method further includes:

[0218] S5. During verification, the integrity verification information and the traceability chain code of the current copyright status are read from the blockchain. The integrity verification information and the traceability chain code of the current copyright status are recalculated based on the data to be verified, and compared with the information on the blockchain. The verification is then determined based on the comparison result.

[0219] This step involves reading integrity verification information and traceability chaincode from the blockchain during verification, recalculating the integrity verification information and traceability chaincode based on the data to be verified, and comparing them with the corresponding information on the blockchain. The verification success is determined based on the comparison result. Specifically, it includes the following sub-steps:

[0220] S51 reads the integrity verification information and the traceability chain code of the current copyright status from the blockchain.

[0221] This sub-step is used to obtain the on-chain benchmark data required for verification.

[0222] In practice, the verification node uses the secondary creation image identifier and copyright status index of the copyright status to be verified. Read the corresponding Merklegen commitment from the blockchain smart contract First key ciphertext On-chain registration timestamp On-chain traceability chaincode that records the current copyright status and algorithm version number and according to Choose processing rules consistent with those used in the registration phase to avoid verification failures due to version inconsistencies.

[0223] S52, Receive the off-chain stored encrypted fingerprint data to be verified provided by the licensor, or the suspected derivative image and its corresponding original image provided by the licensor, and re-execute steps S1 to S3 to generate the encrypted fingerprint data to be verified.

[0224] In practice, if the verification objective is to prove that the encrypted fingerprint stored off-chain has not been tampered with or whether it belongs to a certain copyright status, then the verification node directly uses the encrypted fingerprint data provided by the licensor. .

[0225] If the verification objective is to determine whether a suspected derivative work belongs to a certain copyright status, the verification node needs to re-perform spatial alignment, directional subband decomposition, residual enhancement, content suppression, fingerprint tensor construction, and chaotic phase encryption on the suspected derivative work, and use the same on-chain registered timestamp. Original graph-aware hashing Algorithm version number and mixed keys Ensure the results are reproducible.

[0226] S53, recalculate the integrity verification information based on the encrypted fingerprint data to be verified, and compare it with the integrity verification information on the blockchain.

[0227] This sub-step is used to verify the integrity of the encrypted fingerprint data to be verified, confirming that it has not been tampered with and is consistent with the on-chain evidence.

[0228] The verification node is based on the encrypted fingerprint data to be verified obtained in step S52. Process according to the same serialization, block division, and hashing rules as in step S33. Constructing a Merkle tree yields the recalculated Merkle root. .like On-chain If they match, the integrity verification passes and S54 continues; otherwise, the encrypted fingerprint data to be verified is determined to have been tampered with or does not match the on-chain evidence, and the verification fails.

[0229] S54. After the integrity verification passes, the traceability chain code of the current copyright status is recalculated based on the encrypted fingerprint data to be verified, and a consistency comparison is performed with the traceability chain code of the current copyright status on the blockchain. The verification is then determined based on the comparison result.

[0230] This sub-step is used to verify the legitimacy of the current copyright status within the entire genealogy.

[0231] In practice, the licensor uses the copyright owner's private key to decrypt the first key ciphertext. Recover the first key .

[0232] The verification node recalculates the source chaincode of the current copyright status according to steps S41 to S43. ,Will With on-chain records Bit-by-bit comparison to calculate the normalized Hamming distance :

[0233]

[0234] in, For the traceability chain code bit length, if Less than or equal to the Hamming distance threshold If the result is positive, the verification passes; otherwise, the verification fails.

[0235] Specifically, the number of bits in the traceability chain code. , Take a value of 0.05 to 0.10, preferably 0.08.

[0236] It should be noted that the normalized Hamming distance determination does not require the chaincode to be completely identical bit by bit. Instead, it allows for slight bit differences due to compression, scaling, floating-point calculation errors, or the process of re-extracting fingerprints. The motivation for this design is that derivative images may undergo multiple platform compressions and format conversions during actual distribution. Using a threshold determination can improve the usability of practical verification while maintaining tamper resistance.

[0237] In this embodiment, the derivative image is slightly compressed using JPEG (quality factor 75) and then re-uploaded for verification of copyright ownership. Calculation On-chain Normalized Hamming distance: Only 9 bits out of 256 are inconsistent. . determination: Verification passed. The system confirms that the compressed image originated from this registered derivative work.

[0238] See Figure 10 This presentation displays a Hamming distance determination graph for copyright lineage verification. The graph, presented as a scatter plot, shows the normalized Hamming distance between the recalculated source chaincode and the on-chain record source chaincode for multiple test cases, comparing it to a Hamming distance determination threshold. Most legitimate matching cases have Hamming distances below the threshold, while tampered or erroneous cases significantly exceed the threshold, thus completing the closed-loop verification. The horizontal axis represents the test case number, and the vertical axis represents the normalized Hamming distance, a dimensionless proportional value.

[0239] This invention also proposes a blockchain-based system for tracing and protecting the copyright of derivative works, used to run the aforementioned blockchain-based method for tracing and protecting the copyright of derivative works, including:

[0240] The image alignment module is used to acquire the original image and its secondary creation image, perform content-aware preprocessing on the two, and obtain aligned reference data. The reference data includes the aligned secondary creation brightness image, the aligned original brightness image, and the shrinkage effective area mask.

[0241] The copyright fingerprint generation module is used to perform frequency domain decomposition on the reference data, extract the editing traces of the secondary creation image relative to the original image based on the decomposition results, and perform saliency enhancement and original content interference suppression on the editing traces to generate a copyright fingerprint.

[0242] An encrypted fingerprint data and integrity verification information generation module is used to obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result, and write it into the blockchain;

[0243] The traceability chain code generation module is used to extract the feature vector of the current copyright status based on the encrypted fingerprint data and convert it into the current status identifier, combine the first key and the traceability chain code of the historical copyright status to generate the traceability chain code of the current copyright status and write it into the blockchain.

[0244] In one embodiment, the system further includes: a verification module, used to read the integrity verification information and the traceability chain code of the current copyright status from the blockchain during verification, recalculate the integrity verification information and the traceability chain code of the current copyright status according to the data to be verified, compare them with the information on the blockchain, and determine whether the verification is successful based on the comparison result.

[0245] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0246] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0247] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A blockchain-based method for tracing the source and protecting the copyright of derivative works, characterized in that, include: Obtain the original image and its derivative image, perform content-aware preprocessing on both, and obtain aligned reference data, which includes the aligned derivative brightness map. Align the original brightness map with the shrinking effective area mask; The reference data is decomposed in the frequency domain. Based on the decomposition results, the editing traces of the secondary creation image relative to the original image are extracted. The editing traces are then enhanced for saliency and the interference of the original content is suppressed to generate a copyright fingerprint. Obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result and write it into the blockchain; The feature vector of the current copyright status is extracted from the encrypted fingerprint data and converted into a current status identifier. The traceability chain code of the current copyright status is generated by combining the first key and the traceability chain code of the historical copyright status and written into the blockchain.

2. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 1, characterized in that, After extracting the feature vector of the current copyright status from the encrypted fingerprint data and converting it into a current status identifier, generating the traceability chain code of the current copyright status based on the first key and the traceability chain code of the historical copyright status, and writing it into the blockchain, the method further includes: During verification, the integrity verification information and the traceability chain code of the current copyright status are read from the blockchain. The integrity verification information and the traceability chain code of the current copyright status are recalculated based on the data to be verified, and compared with the information on the blockchain. The verification is then determined based on the comparison result.

3. The blockchain-based method for tracing the source and protecting the copyright of secondary creation images according to claim 1, wherein obtaining the original image and its secondary creation image, performing content-aware preprocessing on both to obtain aligned reference data, is characterized in that... include: Scaling the original image and the derivative images to a uniform base resolution; Spatial registration is performed on the scaled original image and the re-created image to obtain the aligned original image and the re-created image. An effective region mask is generated based on the effective overlap range after spatial registration, and the effective region mask is then shrunk at the boundary to obtain a shrunk effective region mask. The aligned original image and the aligned secondary creation image are converted to a unified brightness representation space to obtain the original brightness image and the secondary creation brightness image.

4. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 1, characterized in that, The generation of the copyright fingerprint includes: Multi-scale and multi-directional frequency domain decomposition is performed on the reference data to obtain the sub-band response of the secondary creation brightness map and the original brightness map at each scale and in each direction. The shrinkage effective region mask is scaled to the sub-band spatial size corresponding to each scale to obtain the effective calculation range of each sub-band. Within the effective calculation range of each sub-band, the directional residuals are extracted by checking the response of each directional sub-band of the secondary creation brightness map through directional high-frequency residual filtering, and the editing traces are significantly enhanced based on the local kurtosis coefficient to obtain the residual significance map. Within the effective calculation range of each sub-band, the original content local variance map is calculated based on the sub-band response in each direction of the original brightness map. A content suppression coefficient map is generated based on the original content local variance map. The content suppression coefficient map is used to suppress the residual significance map to obtain a content suppression residual significance map. The content suppression residual significance maps of all scales and directions are unified in size and stitched together to form a copyright fingerprint.

5. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 1, characterized in that, The steps of obtaining registration time information, generating a first key based on the registration time information and the content identifier of the original brightness map, encrypting the copyright fingerprint using the first key, generating integrity verification information based on the encryption result, and writing it into the blockchain include: Obtain registration time information and generate a first key based on the registration time information and the content identifier of the original brightness map; The copyright fingerprint is encrypted using the first key to obtain encrypted fingerprint data; A binary Merkle tree is constructed based on the encrypted fingerprint data to obtain the Merkle root commitment as integrity verification information, and the Merkle root commitment and the first key ciphertext after asymmetric encryption are written into the blockchain.

6. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 1, characterized in that, The step of extracting the feature vector of the current copyright status based on the encrypted fingerprint data and converting it into a current status identifier, combining the first key and the historical copyright status traceability chaincode to generate the traceability chaincode of the current copyright status and writing it into the blockchain includes: Complex amplitudes are taken from the encrypted fingerprint data and aggregated into a single-channel amplitude map. After two-dimensional Fourier transform, the data is divided into several frequency band sectors. The logarithmic energy of each sector is extracted and assembled into a feature vector, which is used to characterize the frequency domain distribution of editing traces in the current copyright status. A fixed projection matrix is ​​generated based on the first key, and the feature vector is converted into a current state identifier based on the fixed projection matrix; Generate a traceability chain code for the current copyright status based on the first key, the traceability chain code of the historical copyright status, and the current status identifier, and write the traceability chain code of the current copyright status into the blockchain.

7. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 5, characterized in that, The step of constructing a binary Merkle tree based on the encrypted fingerprint data to obtain the Merkle root commitment as integrity verification information, and writing the Merkle root commitment and the first key ciphertext after asymmetric encryption into the blockchain includes: The encrypted fingerprint data is serialized and segmented into multiple fingerprint data blocks; Perform a secure hash calculation on each fingerprint data block to obtain the data block hash value; Construct a binary Merkle tree from the hash values ​​of all fingerprint data blocks to obtain the Merkle root commitment; Integrity verification information and the first key ciphertext are written into the blockchain.

8. The blockchain-based method for tracing the source and protecting the copyright of derivative works as described in claim 5, characterized in that, The step of encrypting the copyright fingerprint using the first key to obtain encrypted fingerprint data includes: Initialize the chaotic mapping based on the first key and generate a pseudo-random phase matrix; By using the normalized copyright fingerprint tensor as the amplitude tensor and performing complex-domain phase rotation using a pseudo-random phase matrix, encrypted fingerprint data is obtained.

9. A blockchain-based system for tracing the source and protecting the copyright of derivative works, used to run the blockchain-based method for tracing the source and protecting the copyright of derivative works as described in any one of claims 1-8, characterized in that, include: The image alignment module is used to acquire the original image and its secondary creation image, perform content-aware preprocessing on the two, and obtain aligned reference data. The reference data includes the aligned secondary creation brightness image, the aligned original brightness image, and the shrinkage effective area mask. The copyright fingerprint generation module is used to perform frequency domain decomposition on the reference data, extract the editing traces of the secondary creation image relative to the original image based on the decomposition results, and perform saliency enhancement and original content interference suppression on the editing traces to generate a copyright fingerprint. An encrypted fingerprint data and integrity verification information generation module is used to obtain registration time information, generate a first key based on the registration time information and the content identifier of the original brightness map, encrypt the copyright fingerprint based on the first key, generate integrity verification information based on the encryption result, and write it into the blockchain; The traceability chain code generation module is used to extract the feature vector of the current copyright status based on the encrypted fingerprint data and convert it into the current status identifier, combine the first key and the traceability chain code of the historical copyright status to generate the traceability chain code of the current copyright status and write it into the blockchain.

10. The blockchain-based secondary creation image traceability and copyright protection system according to claim 9, characterized in that, Also includes: The verification module is used to read the integrity verification information and the traceability chain code of the current copyright status from the blockchain during verification, recalculate the integrity verification information and the traceability chain code of the current copyright status based on the data to be verified, compare them with the information on the blockchain, and determine whether the verification passes based on the comparison result.

Images