Remote sensing image sensitive target security distribution method based on zero watermark and block chain

Abstract

The application relates to a remote sensing image sensitive target security distribution method based on zero watermark and a blockchain, and relates to the technical field of remote sensing image processing. The method comprises the following steps: extracting a sensitive target image from an original remote sensing image according to a mask, and repairing a sensitive area to generate a release image. The release image is normalized, and an effective area is determined to extract a robust feature map. The sensitive target image and the robust feature map are scrambled respectively and then XORed to generate a zero watermark image. The zero watermark image, effective area coordinates and repair algorithm parameters are packaged into a record file and encrypted, and then the release image and the zero watermark image are uploaded to distributed storage to obtain an index. The key of the record file is split, and the index and the key split information are written into a blockchain for storage. When the sensitive target needs to be recovered, an authorized user obtains data and a key according to the blockchain record, reproduces a feature extraction process from a to-be-verified image according to the effective area coordinates and the algorithm parameters, and restores the sensitive target image through inverse operation combined with the zero watermark image.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and more specifically, to a method for securely distributing sensitive targets in remote sensing images based on zero watermarking and blockchain. Background Technology

[0002] Remote sensing images often contain sensitive targets that require protection during public sharing, data release, and collaborative applications. To meet the dual requirements of open sharing and security compliance, it is usually necessary to process sensitive areas and release usable versions that do not disclose sensitive information, while retaining the ability to restore sensitive targets in authorized scenarios.

[0003] Existing technologies primarily employ methods for handling sensitive targets, including irreversible sensitive area hiding, embedded or reversible watermarking schemes, and recovery through external credentials. These methods achieve sensitive content hiding through techniques such as blurring, occlusion, patching, or information embedding, but each has its limitations.

[0004] However, the aforementioned existing technical solutions have significant drawbacks. Embedded watermarks introduce additional distortion into published images, affecting the usability and quantitative analysis of remote sensing images. Irreversible concealment fails to meet authorization recovery requirements, while external credential solutions lack verifiable binding and unified recording with published images. Furthermore, these methods lack robustness to common processing such as compression, noise, and scaling, and inconsistent management of coordinates and parameters leads to difficulties in reproduction, making it difficult to achieve full-scenario forensics and audit evidence preservation. Summary of the Invention

[0005] This invention provides a method for secure distribution of sensitive targets in remote sensing images based on zero watermarking and blockchain, in order to improve at least one of the aforementioned technical problems.

[0006] The present invention provides a secure distribution method for sensitive targets in remote sensing images based on zero watermarking and blockchain, comprising steps S1 to S6.

[0007] S1. Extract sensitive target images from the original remote sensing images based on the mask, repair sensitive areas, and generate published images.

[0008] S2. Normalize the published image to a uniform size and determine the effective region, from which robust feature maps are extracted.

[0009] S3. After scrambling the sensitive target image and the robust feature map respectively, XOR them to generate a zero-watermark image.

[0010] S4. Package the zero-watermark image, effective region coordinates, and repair algorithm parameters into a record file and encrypt it. Then upload it together with the published image and the zero-watermark image to the distributed storage to obtain an index.

[0011] S5. Segment the key of the record file and write the index and key segmentation information into the blockchain for evidence storage.

[0012] S6. When it is necessary to recover sensitive targets, authorized users obtain data and keys based on blockchain records, reproduce the feature extraction process from the image to be verified based on the valid area coordinates and algorithm parameters, and recover the sensitive target image by inverse operation in combination with the zero-watermark image.

[0013] As a further aspect of the present invention, S1 specifically includes the following steps.

[0014] Acquiring remote sensing images Masks for sensitive areas .

[0015] If the mask is empty, the original remote sensing image will be used directly as the published image.

[0016] If sensitive areas exist, use a mask to determine the bounding box of the sensitive areas and crop out the sensitive target image. If there are multiple sensitive areas, each area will be cropped separately.

[0017] Using image inpainting algorithms to fill in the original remote sensing image The sensitive area is made continuous with the background to obtain the published image. The image inpainting algorithm is selected from inpainting based on partial differential equations, inpainting based on sample block matching, or inpainting based on deep generative models.

[0018] As a further aspect of the present invention, S2 specifically includes the following steps.

[0019] Images will be published. Normalized to standard dimensions Image for The width. for The height.

[0020] Reading sensitive target images Size for The width. for The height.

[0021] Calculation satisfies and , Scale factor ,in This is the base size for the singular value decomposition sub-blocks.

[0022] by The center point is taken as a reference to calculate a rectangular range as the effective area coordinate Coord, and a boundary processing rule is set. The boundary processing rule includes truncation and / or mirror padding and / or zero padding.

[0023] According to the effective area coordinate Coord, the effective area is cropped from .

[0024] As a further scheme of the present application, S2 further comprises the following steps.

[0025] The effective area is subjected to a two-level discrete wavelet transform to extract a low-frequency subband , and the low-frequency subband is divided into subblocks.

[0026] The singular value decomposition is performed on each subblock to extract a maximum singular value , and a threshold is generated by taking the mean value of all the maximum singular values .

[0027] .

[0028] In the formula, and are the row index and the column index of the subblock, respectively.

[0029] Bit expansion is performed on to generate a robust feature map of the same size as the sensitive target image.

[0030] As a further scheme of the present application, S3 specifically comprises the following steps.

[0031] A scrambling sequence is generated based on a chaotic mapping parameter. The chaotic mapping parameter includes a mapping type, an initial value, a control parameter, an iteration number, or a sequence length.

[0032] The generated scrambling sequence is used to perform a scrambling operation on the sensitive target image and the robust feature map , respectively, to obtain an encrypted image and an encrypted feature map .

[0033] An exclusive OR operation is performed to generate a zero-watermark image :

[0034] .

[0035] In the formula, represents a scrambling encryption function based on chaotic mapping. represents a bitwise XOR operation.

[0036] As a further scheme of the present application, the scrambling operation employs a Henon chaotic map.

[0037] As a further scheme of the present application, S4 specifically comprises the following steps.

[0038] The zero-watermark image, the valid area coordinates and the repair algorithm parameters are packaged into a record file.

[0039] A random symmetric key Key is generated to encrypt the record file.

[0040] The published image, the zero-watermark image and the encrypted record file are uploaded to a distributed storage to obtain an index.

[0041] Preferably, the record file contains the following fields:

[0042] A standardization rule field is used to record the rule for unifying the image to the same size.

[0043] A coordinate field records the valid area coordinates Coord, which are used to crop the valid area from the standardized image.

[0044] A repair metadata field records the repair algorithm type, the key parameters and the random seed used when generating the repaired carrier image, which are used to reconstruct the environment when forensics is performed.

[0045] A feature extraction parameter field records the algorithm identification and the binarization strategy of feature extraction.

[0046] A scrambling parameter field records the scrambling algorithm type and the parameters.

[0047] As a further scheme of the present application, S5 comprises the following steps.

[0048] A Shamir secret sharing algorithm with a threshold of is used to divide the key Key into secret shares. The minimum number of secret shares required to reconstruct the key is

[0049] The secret shares are distributed to a hybrid storage layer. The hybrid storage layer includes private cloud nodes, notary nodes and IPFS nodes.

[0050] As a further scheme of the present application, S6 specifically comprises the following steps.

[0051] The encrypted record file is downloaded from the distributed storage system, and enough secret shares are collected to reconstruct the key. The record file is decrypted using the key, and the valid area coordinates, the scrambling parameters and the feature extraction strategy are parsed.

[0052] Using the effective region coordinates obtained from the analysis, the image to be verified is normalized and cropped to extract the effective region, and a test feature map is generated according to the feature extraction strategy obtained from the analysis. .

[0053] The test feature map is analyzed using the scrambling parameters obtained from the analysis. Encryption is performed to obtain And combined with the obtained zero-watermark image Computational encryption sensitive targets :

[0054] .

[0055] In the formula This represents a scrambling encryption function based on chaotic mapping. This indicates a bitwise XOR operation.

[0056] Using the scrambling parameters and key pair obtained from parsing Perform the reverse scrambling operation to recover the sensitive target image. .

[0057] As a further aspect of the present invention, it also includes an infringement evidence collection step for the leaked images.

[0058] Obtain suspected infringing images, calculate their hash values, and retrieve evidence records on the blockchain to obtain the original zero-watermark and record files.

[0059] The sensitive areas of suspected infringing images are masked using the valid region coordinates in the log file.

[0060] The repair algorithm and parameters in the log file are used to re-perform the repair process on the occluded area in order to reconstruct the verification environment.

[0061] Features are extracted from the reconstructed image and combined with the on-chain zero-watermark image to perform blind restoration. If a clear and sensitive target can be restored, it is determined to be an infringement.

[0062] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0063] The present application brings significant benefits in the safe distribution and copyright protection of sensitive targets in remote sensing images. First, by adopting zero watermark technology, the zero watermark image is not embedded in the published image, effectively avoiding the secondary distortion problem that may be introduced by traditional watermark embedding, thereby ensuring the high quality and usability of the published image, while realizing the reliable recovery of sensitive targets in authorized scenarios. Second, the present application combines standardized coordinate system and robust feature extraction methods (such as discrete wavelet transform and singular value decomposition), enhancing the adaptability to common image processing such as compression, noise and scaling, and improving the robustness of the recovery process. In addition, by combining key parameter recording with blockchain notarization, the unforgeable binding of hash, content identifier and timestamp is realized, facilitating third-party auditable verification and traceability, enhancing the credibility and transparency of the entire system. Finally, this method has simplicity in engineering landing level, the published image generation and zero watermark registration process are decoupled, and the verification end only needs to reproduce the coordinates and feature flow to complete the recovery, reducing the system complexity and coupling, and improving the practical efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the specific embodiments of the present application. It should be understood that the following drawings only show some specific embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0065] Figure 1 is a flowchart of a remote sensing image sensitive target safe distribution method.

[0066] Figure 2 is a zero watermark generation full flowchart.

[0067] Figure 3 is a sensitive target recovery flowchart.

[0068] Figure 4 is a system interaction and storage distribution timing diagram. DETAILED DESCRIPTION

[0069] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application.

[0070] Please refer to Figures 1 to 4 , the present application provides a remote sensing image sensitive target safe distribution method based on zero watermark and blockchain, which can be executed by a remote sensing image sensitive target safe distribution device based on zero watermark and blockchain (hereinafter referred to as: sensitive image distribution device). In particular, it is executed by one or more processors in the sensitive image distribution device to realize steps S1 to S6.

[0071] S1, extract sensitive target image from original remote sensing image according to mask, and repair sensitive area to generate publishing image.

[0072] S2, normalize publishing image to uniform size and determine effective area, and extract robust feature map therefrom.

[0073] S3, exclusive-OR after scrambling sensitive target image and robust feature map respectively to generate zero-watermark image.

[0074] S4, package zero-watermark image, effective area coordinates and repair algorithm parameters into record file and encrypt, and then upload to distributed storage together with publishing image and zero-watermark image to obtain index.

[0075] S5, split key of record file, and write index and key split information into blockchain for storage.

[0076] S6, when sensitive target needs to be recovered, authorized user obtains data and key according to blockchain record, reproduces feature extraction process from to-be-verified image according to effective area coordinates and algorithm parameters, and recovers sensitive target image by inverse operation combined with zero-watermark image.

[0077] The embodiment realizes authorized recoverable hiding of sensitive target under the premise of not embedding watermark in publishing image and maintaining quality of publishing image as much as possible, that is, authorized party can recover sensitive target image from "publishing image + zero-watermark + record parameter". The robustness to common processing: still stable recovery to compression, noise, scaling and other disturbances.

[0078] On the basis of the above embodiment, an optional embodiment of the present application is shown in Figure 1 , S1 specifically includes steps S11 to S14.

[0079] S11, obtain remote sensing image and mask of sensitive area .

[0080] S12, if the mask is empty, directly take the original remote sensing image as the publishing image.

[0081] Specifically, the remote sensing image owner (RSI Owner) obtains remote sensing image and mask of sensitive area . The mask may be generated by manual annotation, rule detection or deep segmentation model. If is empty (no sensitive area), directly output , and take "no sensitive area" as record state, and the process ends.

[0082] S13. If a sensitive region exists, use a mask to determine the bounding box of the sensitive region and crop out the sensitive target image. If there are multiple sensitive areas, each area will be cropped separately.

[0083] Specifically, the RSI Owner bases on Calculate the bounding box of the sensitive region, and from The sensitive target image is obtained by cropping the middle. If multiple sensitive targets exist, a corresponding [database name] can be generated for each target. The target number and location coordinates are recorded in the record.

[0084] S14. Fill the original remote sensing image using an image inpainting algorithm. The sensitive area is made continuous with the background to obtain the published image. The image inpainting algorithm is selected from inpainting based on partial differential equations, inpainting based on sample block matching, or inpainting based on deep generative models.

[0085] Specifically, the RSI Owner has... The sensitive areas are processed using an image inpainting algorithm to generate the repaired carrier image. This makes the sensitive area visually continuous with the background.

[0086] Patching algorithms may include, but are not limited to: PDE-based patching, PatchMatch-based sample block matching patching, and deep generative model-based patching (GAN / diffusion model, etc.).

[0087] Based on the above embodiments, in an optional embodiment of the present invention, such as Figure 1 As shown, S2 specifically includes S21 to S27.

[0088] S21, Images to be published Normalized to standard dimensions Image for The width. for The height. Specifically, the repaired carrier image. Normalized to standard size (For example, 1024×1024) .

[0089] S22. Read the image of the sensitive target. Size for The width. for The height.

[0090] S23, Calculation satisfies and , Scale factor ,in This determines the base size for the singular value decomposition sub-blocks. This step is used to determine the robust feature extraction region Coord and ensure reproducibility at the validation end.

[0091] Reading sensitive target images Size Then, the scaling factor is calculated according to the feature extraction constraints. . It must be a positive even number, and must simultaneously satisfy the following feature extraction effectiveness constraints and boundary constraints.

[0092] Feature Capacity Constraint: To ensure that the effective region still has enough pixels in the low-frequency subband to support partitioning after performing a two-level discrete wavelet transform (DWT). 3 non-overlapping SVD sub-blocks (each sub-block being at least [size missing]) (pixels). The scaling factor k must satisfy: .in The base size for singular value decomposition sub-blocks (recommended value) That is, requirements Preferred ,Right now ).

[0093] Boundary Constraint: To ensure that the extracted effective region does not exceed the boundary of the standardized image, the following must be satisfied: and Preferably, the preferred sub-block size is set. (correspond ), and calculate the maximum allowable multiple that satisfies the boundary constraints. .like Then take .like ,Pick .like If so, the fallback strategy described in step (2.3) will be executed.

[0094] like Then, a fallback strategy will be implemented, adjusting the standardized dimensions. Alternatively, boundary padding (mirroring / zero padding) can be applied to the effective region to make subsequent feature extraction available, and a fallback strategy can be written into the record.

[0095] by The center point is taken as a reference to calculate a rectangle Coord=(x0, y0, x1, y1), and a boundary processing rule (truncation / mirror padding / zero padding) is agreed to ensure that the extraction area is reproducible.

[0096] S24, taking the center point as a reference, calculating a rectangle range as the effective area coordinate Coord, and agreeing on a boundary processing rule. The boundary processing rule includes truncation and / or mirror padding and / or zero padding. The center point is taken as a reference to calculate a rectangle range as the effective area coordinate Coord, and a boundary processing rule is agreed. The boundary processing rule includes truncation and / or mirror padding and / or zero padding.

[0097] S25, according to the effective area coordinate Coord, cropping from to obtain the effective area .

[0098] Specifically, according to Coord, the effective area is cropped from . If an out-of-boundary occurs, the effective area is generated according to the agreed boundary processing rule.

[0099] S25, performing two-level discrete wavelet transform on the effective area to extract a low-frequency sub-band , and dividing the low-frequency sub-band into sub-blocks.

[0100] Specifically, two-level discrete wavelet transform (DWT) is performed on the effective area to extract a low-frequency sub-band containing the main energy of the image, to filter high-frequency noise interference and compress the data dimension. Then, according to the target size of the sensitive target image , the low-frequency sub-band is divided into non-overlapping sub-blocks. According to the constraint in step (2.2) of Embodiment 2, the size of each sub-block is at least (e.g. pixels).

[0101] S26, performing singular value decomposition on each sub-block to extract the maximum singular value , and taking the mean of all maximum singular values as a threshold to generate a binary feature matrix .

[0102] .

[0103] In the formula, and are the row index and column index of the sub-block, respectively.

[0104] Specifically, singular value decomposition (SVD) is first performed on each sub-block to extract the maximum singular value of each sub-block . Then, the average of the maximum singular values of all sub-blocks is calculated as an adaptive threshold . If , it is determined that the position feature is 1. Otherwise, it is 0, and a binary feature matrix is generated.

[0105] S27, bit expansion is performed on to generate a robust feature map of the same size as the sensitive target image .

[0106] When he, bit expansion is performed on : if , it is expanded to pixel value 255. If , it is expanded to pixel value 0. An expanded feature map of the same size and bit depth as the sensitive target is generated .

[0107] Based on the above embodiments, an optional embodiment of the present application includes the following steps, as shown in Figure 1 .

[0108] A scrambling sequence is generated based on a chaotic mapping parameter. The chaotic mapping parameter includes a mapping type, an initial value, a control parameter, an iteration number or a sequence length.

[0109] The generated scrambling sequence is used to perform a scrambling operation on the sensitive target image and the robust feature map , respectively, to obtain an encrypted image and an encrypted feature map . Preferably, the scrambling operation uses Henon chaotic mapping.

[0110] An exclusive OR operation is performed to generate a zero-watermark image :

[0111] .

[0112] In the formula, represents a scrambling encryption function based on chaotic mapping. represents a bitwise exclusive OR operation.

[0113] Specifically, a scrambling sequence is generated based on a chaotic mapping parameter (mapping type, initial value, control parameter, iteration number / sequence length, etc.). Then, the Henon scrambling operation is performed on the sensitive target image and the expanded feature map using the scrambling sequence to scramble the pixel positions and obtain an encrypted image With encrypted feature map . Finally, the exclusive or operation is performed to obtain the zero watermark : .

[0114] On the basis of the above-mentioned embodiments, in an optional embodiment of the present application, as shown in Figure 4 S4 specifically includes S41 to S43.

[0115] S41, the zero watermark image, the effective area coordinates and the repair algorithm parameters are packaged into a record file record.

[0116] Preferably, the record file contains the following fields:

[0117] The standardization rule field is used for recording the rule of unifying the image to the same size.

[0118] The coordinate field records the effective area coordinates Coord, which is used for cutting the effective area from the standardized image.

[0119] The repair metadata field records the repair algorithm type, the key parameters and the random seed used when generating the repaired carrier image, which is used for environment reconstruction when forensics is performed.

[0120] The feature extraction parameter field records the algorithm identification and the binarization strategy of feature extraction.

[0121] The scrambling parameter field records the scrambling algorithm type and the parameters.

[0122] Specifically, the record file record (for example, record.json) contains at least the following fields (examples):

[0123] The record structure version number (schema_version) is used for parsing the compatibility and subsequent upgrade (when the record is read by the verification / forensics end, it is used for selecting the corresponding field interpretation rule).

[0124] The standardization rule (standardize) is used for unifying the image to the same size (width / height), so as to ensure that the verification end can reproduce the features and the coordinate system.

[0125] The effective area coordinates (feature extraction / recovery coordinates) (coord) give the cutting frame (x0, y0, x1, y1) in the coordinate system specified by the coord_system, which is used for cutting the effective area from the standardized image to extract robust features.

[0126] ​Inpainting Meta: Record the type of inpainting algorithm (e.g. PatchMatch / GAN), key parameters and random seed (RandomSeed) used when generating the published image.

[0127] Pasteback Coord (Reconstruction Coord): Used to form a visual reconstruction image by pasting the recovered sensitive target back to its corresponding position in the published image (e.g. {xMin, yMin, xMax, yMax}).

[0128] Feature Profile ID: Used to verify that the feature extraction implementation on the verifier side is consistent with that on the registrar side. Default not to transfer detailed parameters one by one.

[0129] Feature Binarize / Quantization Strategy and Parameters (Method, Threshold )(feature_binarize): Used to enhance the consistency of XOR operation. This is a mandatory field in the present embodiment.

[0130] Scramble Algo: Includes the algorithm type (Henon / Logistic / Arnold, etc.) and the scrambling parameter K.

[0131] S42, a random symmetric key Key is generated to encrypt the record file.

[0132] Specifically, the RSI Owner generates a random symmetric key Key (e.g. AES-256 Key). Then, the record file containing the sensitive target recovery parameters is encrypted using the key Key to generate a ciphertext file record.enc.

[0133] S43, the published image, the zero-watermark image and the encrypted record file are uploaded together to the distributed storage to obtain an index.

[0134] Specifically, the desensitized published image , the zero-watermark and the ciphertext record.enc are uploaded to the IPFS. Return CIDs: Obtain carrier_cid, watermark_cid and meta_cid (pointing to record.enc).

[0135] On the basis of the above embodiments, an optional embodiment of the present application includes S51 to S52 as shown in Figure 4 .​

[0136] S51, using a threshold of Shamir's Secret Sharing algorithm, the key Key is divided into secret shares. The minimum number of secret shares required to reconstruct the key.

[0137] S52, the secret shares are distributed to the hybrid storage layer. The hybrid storage layer includes private cloud nodes, notary nodes and IPFS nodes.

[0138] First, key splitting (SSS Sharding) is performed. The RSI Owner uses a threshold Shamir's Secret Sharing algorithm to divide the key Key into secret shares (Shares). For example, set =3, =2, that is, 3 shares are generated, and any 2 shares can reconstruct the original record.

[0139] Then, distribute the shares (Distribute Shares): distribute the generated Shares to different nodes of the hybrid storage layer (Hybrid Storage Layer) to eliminate single point failure and enhance security.

[0140] Node 1: Private Cloud (RSI Owner's private cloud, fully controlled).

[0141] Node 2: Notary Node (third-party notary node / supervision node).

[0142] Node 3: IPFS (encrypted public storage, or stored in decentralized confidential storage).

[0143] Finally, register metadata (Register Metadata): write carrier_cid, watermark_cid, meta_cid, and the storage index address of each Share to the smart contract.

[0144] On the basis of the above embodiment, in an optional embodiment of the present application, as shown in Figure 4 S6 specifically includes S61 to S64.

[0145] S61, download the encrypted record file from the distributed storage system, and collect enough key shares to reconstruct the key, use the key to decrypt the record file, and parse to obtain the effective area coordinates, scrambling parameters and feature extraction strategy.

[0146] First, metadata query: RSIBuyerinitiates a query request to the smart contract (e.g., query remote sensing images of a specific area or specific ID). The smart contract returns the return content identifier (CID) and index of the image (Return CIDs&Indexes), including carrier_cid, watermark_cid, meta_cid, and index information of recovery key shares.

[0147] Then, asset download: download , and encrypted record file (record.enc): RSIBuyer downloads the published image , zero watermark , encrypted record file (record.enc) from the IPFS network according to the returned CID.

[0148] Then, share request: RSIBuyerinitiates a request to obtain shares from each node of the hybrid storage layer (Private Cloud, Notary Node, IPFS). The storage nodes (especially Private Cloud and Notary Node) will verify the purchase credentials (NFT / Token) or authorization signature of the user (Buyer). Only after verification, the corresponding Share data is returned.

[0149] Finally, reconstruct the key (Reconstruct "Key"): RSIBuyer collects at least valid shares (e.g., 2). Decrypt record.enc with K to get the original record parameter package. Parse record to get key information such as standardized size, effective area coordinates Coord, feature extraction parameters, and scrambling key reference.

[0150] S62, use the effective area coordinates obtained by parsing to normalize and crop the to-be-verified image, extract the effective area, and generate test feature maps according to the feature extraction strategy obtained by parsing .

[0151] Specifically, is normalized to the same size WxH according to the standardization rule in the record file record, and is obtained. Then, Coord is reproduced according to the rule in record and the boundary processing rule, and is cropped according to Coord to obtain Extracting the effective region Then, according to the feature extraction algorithm, the binarization strategy and the scrambling parameter recorded in the record file, the effective region is processed in the same way as the RSIOwner side: Then, the is subjected to two-level discrete wavelet transform (DWT) and block singular value decomposition (SVD), the maximum singular value is extracted, and a binary feature matrix is generated according to the recorded strategy (such as adaptive mean threshold). Finally, the binary feature matrix is subjected to bit expansion operation (logical 1 is expanded to pixel value 255, and logical 0 is expanded to pixel value 0), and a test feature map with the same bit depth as the sensitive target is generated.

[0152] S63, the test feature map is encrypted using the parsed scrambling parameter to obtain , and the obtained zero-watermark image is combined to calculate the encrypted sensitive target :

[0153] .

[0154] In the formula, represents a scrambling encryption function based on chaotic mapping. represents a bitwise XOR operation.

[0155] Specifically, according to the scrambling parameter in the record, the is subjected to Henon scrambling to obtain the encrypted test feature map . Then, the XOR operation is performed: .

[0156] S64, the scrambling parameter and the key obtained by parsing are used to perform inverse scrambling operation on to restore the sensitive target image .

[0157] Specifically, the is subjected to inverse scrambling to restore the sensitive target image using the scramble_algo parameter recorded in the record. Then, according to the pasting coordinates contained in the record, the is pasted to the corresponding position of for visual reconstruction.

[0158] On the basis of the above embodiments, in an optional embodiment of the present application, the remote sensing image sensitive target security distribution method based on zero watermark and blockchain further includes the infringement evidence collection step for the leaked image.

[0159] Get the suspected infringement image, calculate its hash value and retrieve the record on the blockchain, get the original zero watermark and record file.

[0160] Use the effective area coordinates in the record file to mask and shield the sensitive area of the suspected infringement image.

[0161] Use the repair algorithm and parameters in the record file to re-execute the repair processing of the shielded area to reconstruct the verification environment.

[0162] Extract features from the reconstructed image and perform blind recovery in combination with the on-chain zero watermark image. If the clear sensitive target can be recovered, it is determined to be infringement.

[0163] This embodiment realizes full-scene forensics and auditable record storage: not only can it deal with the release of stolen images, but also can deal with the traceability of leaked original sensitive images. The key coordinates, repair parameters and hash / CID information are standardized and can be chained to solidify, which is convenient for third-party verification of consistency and timestamp.

[0164] Specifically, the suspected infringement image is obtained . Calculate SHA-256 digest and perceptual hash (pHash), and stamp with trusted timestamp.

[0165] Note that the infringement sample may belong to the following two categories. The stolen release image category: Approximately the release image (May experience compression / scaling / cropping and other propagation processing), and the sensitive area is the background after repair. The leak recovery category: It is an image with "recovered sensitive target and pasted back" or a directly leaked original image I, and its sensitive area contains real target content, which is significantly different from the release image .

[0166] Use the perceptual hash or related business ID of to retrieve on the blockchain, locate the record storage, and obtain the index of the original zero watermark and record (and decrypted parameters).

[0167] Verification environment reconstruction: if the sample is determined to be "leak recovery category", the sensitive area needs to be masked and shielded using the coordinates Coord in the record, and the repair algorithm and parameters recorded in the record are used to regenerate the repair area, so that its background features are consistent with the release image .

[0168] Blind extraction and recovery: without knowing the original sensitive target, use the processed and the on-chain public , record, execute the recovery process in embodiment 4.

[0169] Result determination: if the clear sensitive target (highly semantically consistent with the explicit target or original recorded target in ) can be recovered, it constitutes direct infringement evidence.

[0170] Report generation: output the complete judicial authentication report containing "on-chain evidence hash + off-chain recovery result + traceable identity ID".

[0171] In summary, when it is necessary to publish remote sensing images containing sensitive targets to the outside world and retain authorized recovery capabilities, the processing steps of the present application are as follows:

[0172] I. Zero watermark generation and registration step: according to the sensitive area mask, the sensitive target image is cropped from the original remote sensing image , and the repair processing is performed on the sensitive area to obtain the repaired carrier image (as the published image). The is normalized to a uniform size and the effective area coordinates Coord are determined based on the center point, and the robust feature map is extracted from the effective area .The and are respectively encrypted by chaotic scrambling, and then XORed to obtain the zero watermark image . Finally, the , Coord, repair algorithm parameters (used for environment reconstruction) and scrambling parameters are packaged and saved to form a reproducible record (Record).

[0173] II. Evidence binding step: encrypt the record file containing Coord and repair parameters (generate ciphertext record.enc), upload , and record.enc to distributed storage to obtain CID. The decryption key is stored in slices or mixed. Finally, the image hash, CID and key index information are written into the blockchain transaction to obtain the on-chain timestamp and verifiability.

[0174] III. Sensitive target recovery step: the image to be verified is normalized according to the same rules as the registration stage and Coord is determined, the test features are extracted and scrambled to obtain . . And the inverse scrambling is performed to obtain the recovered sensitive target image .

[0175] Four, infringement evidence collection and closed loop verification steps: calculate the hash of the suspected infringement image and retrieve the on-chain evidence record. For leaked images containing real sensitive content (i.e. "leak recovery type"), use the recorded repair algorithm parameters and coordinates to perform environment reconstruction (i.e. mask the sensitive area and force the use of the same parameters to re-repair, to reproduce the background features at the time of registration). Then use the on-chain Perform blind recovery with the reconstructed features. If the clear sensitive target can be recovered, a complete evidence chain of "on-chain evidence index + off-chain technology reproduction" is established, realizing closed loop verification.

[0176] According to the above steps, the publishing stage realizes zero-embedded and traceless publishing, and the output of the repaired carrier image Both the use value and the sensitive information are completely isolated. The evidence storage stage adopts the "ciphertext off-chain storage + hash on-chain solidification" strategy, which not only guarantees the privacy and security of the recovery parameters, but also realizes the non-tamperable ownership binding. In the recovery and evidence collection stage, the authorized person or the evidence collector can accurately reproduce the feature extraction context by using the decrypted parameters (especially the repair algorithm parameters) through the "environment reconstruction" mechanism, so that in complex infringement scenarios such as regular authorized recovery and original image leakage, the sensitive target can be stably recovered and closed loop verified.

[0177] The zero-watermark and blockchain-based remote sensing image sensitive target secure distribution method of the embodiment solves the problems of the existing sensitive target hiding scheme that "the usability of the published image and the authorized recoverability of the sensitive target are difficult to balance", and the dispersion of the coordinates and parameters required for recovery, which are difficult to reproduce and audit. The embodiment takes "zero watermark not embedded" as the core, generates a zero watermark combined with robust feature extraction and a reproducible coordinate system, and optionally binds and stores the key hash, CID and Coord through distributed storage and blockchain, so that the quality of the published image is maintained while authorized recovery and verifiable traceability are realized.

[0178] The zero-watermark and blockchain-based remote sensing image sensitive target secure distribution method of the embodiment has the following beneficial effects.

[0179] 1. Not embedded, low distortion: zero watermark Not embedded in the published image , avoiding secondary distortion. The published image is consistent or approximately consistent with the repaired carrier.

[0180] 2. Authorized recoverable: authorized parties can recover the sensitive target image from the robust features extracted from the to-be-verified image .

[0181] 3. Stronger robustness: through "standardization + central reference Coord + robust features (such as DWT + SVD)", the adaptability to common processing such as compression and noise is enhanced.

[0182] 4、Auditability: key credentials and coordinates are fixed by hash, CID and blockchain timestamp, which is convenient for third-party verification of consistency and traceability.

[0183] 5、Engineering landing is more concise: decoupling of image generation and zero-watermark registration. The verification end only needs to reproduce the Coord and feature flow to restore, reducing the complexity of cross-system coupling.

[0184] It can be understood that the sensitive image distribution device can be a portable notebook computer, a desktop computer, a server, a smart phone or a tablet computer, and the like electronic device with computing performance.

[0185] Obviously, the embodiments described in the foregoing are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0186] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed apparatus and method can also be implemented by other manners. The apparatus and method embodiments described above are only illustrative, for example, the flowcharts and block diagrams in the drawings show the possible implementation architecture, function and operation of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowchart or block diagram can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementation manners, the functions noted in the blocks can also occur in different order from that noted in the drawings. For example, two continuous blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for executing the specified function or action, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0187] In addition, each functional module in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0188] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts of the prior art that make contributions or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and various media that can store program codes. It should be noted that in this document, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the processes, methods, articles or devices that include a series of elements not only include those elements, but also include other elements that are not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0189] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application are also intended to include the plural forms unless the context clearly indicates otherwise.

[0190] It should be understood that the term "and / or" used herein is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this document generally represents an "or" relationship between the front and rear associated objects.

[0191] Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if it is determined" or "if (a stated condition or event) is detected" can be interpreted as "when it is determined" or "in response to determining" or "when (a stated condition or event) is detected" or "in response to detecting (a stated condition or event)".

[0192] The "first / second" mentioned in the embodiments only distinguishes similar objects, and does not represent a specific order for the objects. Understandably, the "first / second" can be interchanged in a specific order or sequence as appropriate. It should be understood that the objects distinguished by "first / second" can be interchanged as appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0193] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Various modifications and changes can be made to the present application by those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for secure distribution of sensitive targets in remote sensing images based on zero-watermarking and blockchain, characterized in that, Comprise: S1, extract sensitive target image from original remote sensing image according to mask, and repair sensitive area using image inpainting algorithm to generate publishing image; S2, normalize publishing image to uniform size and determine effective area, and extract robust feature map therefrom; S3, exclusive or sensitive target image and robust feature map after scrambling respectively to generate zero watermark image; S4, package zero watermark image, effective area coordinates and image inpainting algorithm parameters into record file and encrypt, and then upload to distributed storage together with publishing image and zero watermark image to obtain index; S5, split the key of record file, and write the index and key split information into blockchain; S6, when sensitive target needs to be recovered, authorized user obtains data and key according to blockchain record, reproduces feature extraction process from to-be-verified image according to the effective area coordinates and algorithm parameters, and restores sensitive target image by inverse operation combined with the zero watermark image; S3 specifically comprises: Generate scrambling sequence based on chaotic mapping parameters; Chaotic mapping parameters include mapping type, initial value, control parameter, iteration number or sequence length; The scrambled sequence is used to scramble the sensitive target image and the robust feature map respectively to obtain an encrypted image and an encrypted feature map respectively Performing exclusive-or operation to generate zero-watermark image : ； In the formula represents a scrambling encryption function based on a chaotic mapping; represents a bitwise XOR operation; It also includes infringement evidence steps for leaked images: Obtain suspected infringing image, calculate its hash value and retrieve storage record on the blockchain, obtain zero watermark image and record file; Use the effective area coordinates in the record file to mask and shield the sensitive area of the suspected infringing image; Use the repair algorithm and parameters in the record file to perform repair processing on the shielded area to reconstruct the verification environment; Extract features on the reconstructed image and combine the on-chain zero watermark image to perform recovery; if a clear sensitive target can be recovered, it is determined as infringement.

2. The method of claim 1, wherein, S1 specifically comprises: Acquiring remote sensing images Masking sensitive areas ; If the mask is empty, directly use the original remote sensing image as the publishing image; If there is a sensitive area, the sensitive area bounding box is determined by using a mask, and the sensitive target image is cropped If there are multiple sensitive areas, each area is cropped respectively Filling original remote sensing images with an image inpainting algorithm of sensitive areas to make them continuous with the background, obtaining a published image ; wherein the image inpainting algorithm is selected from partial differential equation-based inpainting, sample block matching-based inpainting or deep generative model-based inpainting.

3. The method of claim 1, wherein, S2 specifically comprises: publishing an image to a standard size of the image ; to a width of ; to a height of ; Reading sensitive target image of the size ; is width; is height; Computing satisfying and , a scale factor where is the base size of the singular value decomposition sub-block; In The rectangular range is calculated as the effective area coordinate Coord based on the center point, and a boundary processing rule is set; the boundary processing rule includes truncation and / or mirror filling and / or zero filling; According to the effective region coordinate Coord, the effective region is cropped from .​ 4. The method of claim 3, wherein, S2 further comprises: effective region performing two-level discrete wavelet transform to extract a low-frequency subband and dividing the low-frequency subband into subblocks Perform singular value decomposition on each sub-block, extract the largest singular value and take the mean of all largest singular values as the threshold Generate a binary feature matrix ; ； wherein and are the row index and column index of the sub-block, respectively; To Perform bit expansion to generate robust feature maps of the same size as the sensitive target image .

5. The method of claim 1, wherein, The scrambling operation adopts Henon chaotic mapping.

6. The method of claim 1, wherein, S4 specifically comprises: Package zero watermark image, effective area coordinates and image inpainting algorithm parameters into record file; Generate a random symmetric key Key to encrypt the record file; Upload publishing image, zero watermark image and encrypted record file to distributed storage to obtain index; The record file contains the following fields: Standardization rule field, used to record the rule of unifying the image to the same size; Coordinate field, recording effective area coordinates Coord, used to crop the effective area from the standardized image; Repair metadata field, recording the image repair algorithm type, key parameters and random seed used when generating the publishing image, used for environment reconstruction when taking evidence; Feature extraction parameter field, recording the algorithm identifier and binary strategy of feature extraction; Scrambling parameter field, recording the scrambling algorithm type and parameters.

7. The method of claim 1, wherein S5 Comprise: The Shamir secret sharing algorithm with a threshold of is used to split the key Key into secret shares; is the minimum number of shares needed to reconstruct the key. Will Each secret share is distributed to a hybrid storage layer, which includes private cloud nodes, notary nodes, and IPFS nodes. 8.The method of claim 1 to 7, wherein, S6 specifically comprises: Download the encrypted record file from the distributed storage system, collect enough key shares to reconstruct the key, decrypt the record file using the key, and parse to obtain effective area coordinates, scrambling parameters and feature extraction strategy; The effective region coordinates obtained by analysis are used to normalize and cut the to-be-verified image, extract the effective region, and generate a test feature map according to the feature extraction strategy obtained by analysis ; The test feature map is encrypted by using the obtained scrambling parameters , and the obtained zero-watermark image is combined to obtain the encrypted sensitive target : ； In the formula represents a scrambling encryption function based on a chaotic mapping; represents a bitwise XOR operation; The scrambling parameters and the key are obtained by analysis An inverse scrambling operation is performed to restore the sensitive target image .

Images