Digital archive safe circulation tracking method and system based on dynamic watermark

Abstract

The invention discloses a digital archive safe circulation tracking method and system based on a dynamic watermark, and relates to the technical field of digital archives, and the method comprises the steps: collecting the physical layout features and semantic features of an archive at the same time, generating a comprehensive archive feature map, and dynamically adjusting the watermark intensity in combination with a rank screening threshold; a visual fidelity and watermark extraction degree double-index evaluation system is introduced, watermark position evaluation is carried out, and archive carrier sequence substandard and standard labels are obtained; and by monitoring the watermark survival rate, the visual fidelity and the circulation error rate, a correction strategy selection rule is constructed and comprises position iteration correction or sequence correction, so that evaluation and dynamic calibration of the watermark embedding position are realized, and dynamic selection of a correction strategy is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital archives, and specifically relates to a method and system for secure transfer tracking of digital archives based on dynamic watermarking. Background Art

[0002] Digital archives are increasingly widely used in modern information management, but their security and protection during the transfer process have become problems that need to be solved urgently. Traditional methods of file storage and management are difficult to cope with the challenges brought by the digital environment, such as security risks such as data leakage and tampering. In addition, during the transfer process, files are easily affected by various factors, resulting in damage to the integrity of the data, and even loss or error of information. Traditional file protection measures, such as encryption technology and digital signatures, although they can provide a certain degree of security, lack effective management of the dynamic changes during the transmission and use of files, and cannot monitor and adjust protection strategies in real time to cope with new threats.

[0003] Based on this, a method and system for secure transfer tracking of digital archives based on dynamic watermarking are proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for secure transfer tracking of digital archives based on dynamic watermarking.

[0005] A method for secure transfer tracking of digital archives based on dynamic watermarking includes: collecting physical layout features and semantic features of digital archive data to generate archive feature maps, constructing each archive carrier sequence based on the rank screening threshold and watermark strength corresponding to each archive feature map, and performing watermark embedding operation on it to obtain an initial watermark embedding result; Based on each archive carrier sequence and its corresponding initial watermark embedding result, through data analysis, the corresponding visual fidelity and watermark extraction degree are obtained to evaluate the watermark position of each initial watermark embedding result, and archive carrier sequence non-compliance and compliance labels are obtained; Collect the non-compliance labels of each initial watermark embedding result, form a set of non-compliant archive carrier sequences, combine their corresponding visual fidelity and watermark extraction degree, construct a multi-objective optimization function and output the optimal watermark position to obtain a secondary watermark embedding result; Monitor the transfer situation of each archive carrier sequence, obtain the watermark survival rate, visual fidelity, and transfer error rate corresponding to each secondary watermark embedding result, and construct a correction strategy selection rule, and execute the corresponding secondary correction strategy according to the correction strategy selection rule; the secondary correction strategy includes iterative correction of the watermark embedding position, correction of each archive carrier sequence, and no correction.

[0006] As a further solution of the present invention: collecting the physical layout features and semantic features of digital archive data and generating an archive feature map, where the physical layout features are text block position coordinates, font sizes, and paragraph spacings, and the semantic features are text keywords, text topics, and text semantic relationships; The generating of the archive feature map is as follows: based on a pre-trained convolutional neural network (CNN) model, the physical layout features and semantic features are weighted and fused to obtain the archive feature map denoted as , where \(R\) represents the set of real numbers of the feature data in the archive feature map, \(n\) is the number of archives, \(g\) and \(w\) are the height and width of the feature map, and \(d\) is the number of channels of the feature map.

[0007] As a further solution of the present invention: constructing an archive carrier sequence based on the rank screening threshold and watermark strength corresponding to each archive feature map, specifically: For each archive feature map , where \(i\) is the identifier of each archive feature map and 1, 2,..., \(n\); \(n\) is the number of archives; combine into the total number of spatial positions \(N = , to obtain the two-dimensional matrix , where \(N\) is the total number of spatial positions; For each perform singular value decomposition to obtain ; in the formula, and represent matrices of order \(N\) and \(d\) respectively, and the superscript \(T\) represents the conjugate transpose of the matrix, is the matrix, the main diagonal is composed of singular values , \(P\) is the theoretical maximum rank, and the remaining elements are all 0, and satisfy , Each column vector of is an eigenvector of the matrix , and they are mutually orthogonal, and are called the left singular vectors of the matrix = { , ,..., } Each column vector of is an eigenvector of the matrix , and they are mutually orthogonal, and are called the right singular vectors of the matrix Based on the above decomposition process, all the information of the archive feature map is obtained, where \(P\) is the theoretical maximum rank and \(j\) is the index variable of \(P\); Define the minimum value in all the information of the feature map as the rank screening threshold , and take the first The cumulative energy proportion of singular values is compared with a preset threshold ; When , retain the sub-regions corresponding to the first singular values as the feature sub-region, i.e., the watermark embedding region; When , retain all singular values, and do not process them as they have no analytical value; Based on the preset watermark strength and introducing a rank screening threshold to obtain a dynamically adjusted strength coefficient , according to the formula ; where P is the theoretical maximum rank; Convert the feature sub-regions corresponding to each file feature map into a one-dimensional carrier sequence that can embed the watermark, denoted as the file carrier sequence , and combine the dynamically adjusted strength coefficients corresponding to each feature sub-region , and perform sorting and splicing to obtain the global file carrier sequence ; where represents the number of file carrier sequences; Perform watermark embedding operations on each file carrier sequence to obtain the initial watermark embedding result.

[0008] As a further solution of the present invention: Collect each file carrier sequence and its corresponding initial watermark embedding result, and obtain the corresponding visual fidelity and watermark extraction degree through data analysis. Specifically: Each file carrier sequence is the original digitalized file without watermark embedding operation; The initial watermark embedding result is the digitalized file of the watermarked carrier after the watermark embedding operation; Based on each file carrier sequence and each initial watermark embedding result Calculate the visual fidelity based on the structural similarity formula; Perform watermark extraction operations on each initial watermark embedding result to obtain the extracted watermark sequence and the original watermark sequence corresponding to each initial watermark embedding result, and calculate the watermark extraction degree , according to the following formula: ; where, is the total number of digitalized files, is the length of the watermark sequence of the digitalized file q is the index of the element in the watermark sequence, is the qth element in the watermark sequence of the digitalized file, is the q-th element in the watermark sequence extracted from the digital file, which is obtained based on the mean value formula of the mean of the original watermark sequences of the i-th digital file, and is the mean of the watermark sequence of the watermark extracted from the i-th digital file.

[0009] As a further solution of the present invention: the evaluation of the watermark position for each initial watermark embedding result specifically includes: based on each file carrier sequence and each initial watermark embedding result corresponding visual fidelity and watermark extraction degree , and a watermark position evaluation model is constructed based on the preset visual fidelity and the compliance intervals of the watermark extraction degree, specifically as shown in the following formula: ; in the formula, , respectively refer to the compliance intervals of the preset visual fidelity and the watermark extraction degree, 1 and -1 are evaluation identifiers, indicating that the watermark position of the corresponding initial watermark embedding result is normal or the watermark position state is abnormal, respectively.

[0010] As a further solution of the present invention: generating the non-compliance and compliance labels for the file carrier sequence includes: According to the value of the evaluation identifier -1, the calculated visual fidelity and the watermark extraction degree are compared with their corresponding compliance intervals of the preset visual fidelity and the watermark extraction degree , for comparison and judgment; When the visual fidelity or the watermark extraction degree does not belong to its corresponding compliance interval , , an abnormal label of non-compliance of the initial watermark embedding result is generated; When the visual fidelity and the watermark extraction degree belong to their corresponding compliance intervals , , a compliance label of the initial watermark embedding result is generated.

[0011] As a further solution of the present invention: constructing a multi-objective optimization function and outputting the optimal watermark position to obtain the secondary watermark embedding result specifically: traversing to obtain each non-compliance label of the initial watermark embedding result and each file carrier sequence corresponding to each non-compliance label of the initial watermark embedding result, forming a non-compliance file carrier sequence set; Based on the visual fidelity and watermark extraction degree indexes of each non-compliance set of the file carrier sequence, a visual fidelity sub-function is constructed and the watermark extraction degree sub-function and introduce a preset weight coefficient to obtain a multi-objective optimization function ; Combined with the constraint conditions: the watermark embedding position belongs to the preset archive carrier boundary D, denoted as x ∈ D, and the watermark strength is less than or equal to the preset maximum watermark strength, denoted as , Based on the above multi-objective optimization function and constraint conditions, use the Pareto front solution algorithm to obtain the optimal watermark position as: Adopt the Pareto front solution algorithm, and for each individual (candidate solution) in the population Encode it as the watermark embedding position x and the watermark strength , adopt real number encoding, and calculate its objective sub-function value and ; Compare the dominance relationships among all individuals: When individual is superior to in all objectives, and is superior to in at least one objective, then dominates ; Divide the population into multiple front layers, and calculate the crowding degree. Calculate the crowding degree for the solutions within the same front layer. According to the crowding degree, generate a new population through the three basic operations of selection, crossover, and mutation of the genetic algorithm; According to the preset iteration termination condition, that is, the preset maximum number of iterations, retain all non-dominated solutions to form the Pareto optimal solution; And take the one in the Pareto optimal solution as the optimal watermark position ; Based on the above optimal watermark position Perform watermark embedding operation on the corresponding archive carrier sequence to obtain the secondary watermark embedding result.

[0012] As a further solution of the present invention: construct a correction strategy selection rule. Specifically: define the watermark survival rate, visual fidelity, and transfer error rate as input variables, and divide them into different fuzzy sets respectively; Define the correction strategy selection rule as the output variable and divide it into a fuzzy set; Formulate fuzzy rules to describe the influence of the watermark survival rate, visual fidelity, and transfer error rate on the correction strategy selection rule; Perform fuzzy reasoning according to the fuzzy rules to determine the result of the correction strategy selection rule; The correction strategy selection rule includes iterative correction of the watermark embedding position, correction of the archive carrier sequence, and no correction.

[0013] As a further solution of the present invention: the watermark embedding position is iteratively corrected, that is, the position for watermark embedding is iteratively corrected, and the corresponding file carrier sequence is returned to step two for re-evaluation of the optimal position; Each file carrier sequence is corrected, specifically: S1: Based on each evaluation index and the initial construction parameters corresponding to each file carrier sequence, an association model is established; the evaluation indexes are visual fidelity , watermark survival rate , transfer error rate , and the initial construction parameters corresponding to each file carrier sequence are that the initial construction parameters include a rank screening threshold , watermark strength , and the connection density of feature map nodes ; where r represents the number of carrier sequences to be corrected; S2: Construct an association model between the evaluation index and the construction parameter ; S3: Calculate the difference between each evaluation index, visual fidelity , watermark survival rate , transfer error rate and its corresponding preset threshold , , to obtain its corresponding deviation value , , , and calculate the correction coefficient matrix according to each deviation value by using the linear interpolation method , where Q represents the proportional coefficient of each parameter to be adjusted; S4: Update the initial construction parameters corresponding to each file carrier sequence according to the correction coefficient corresponding to each construction parameter to obtain the updated construction parameters , , ; And reconstruct each file carrier sequence corresponding to the updated construction parameters to generate a corrected carrier sequence.

[0014] The digital file security transfer tracking system with dynamic watermark includes a file initial watermark embedding module, a watermark position initial evaluation module, a secondary watermark embedding module, and a transfer monitoring and correction strategy construction module; The file initial watermark embedding module is used to collect the physical layout features and semantic features of digital file data and generate a file feature map, construct each file carrier sequence based on the rank screening threshold and watermark strength corresponding to each file feature map, and perform a watermark embedding operation on it to obtain an initial watermark embedding result; The initial watermark position evaluation module is used to evaluate the watermark position of each initial watermark embedding result based on each file carrier sequence and its corresponding initial watermark embedding result, and obtain the unqualified and qualified labels of the file carrier sequence by analyzing the corresponding visual fidelity and watermark extraction degree through data analysis; The secondary watermark embedding module is used to collect the unqualified labels of each initial watermark embedding result, form a set of unqualified file carrier sequences, combine their corresponding visual fidelity and watermark extraction degree, construct a multi-objective optimization function, and output the optimal watermark position to obtain the secondary watermark embedding result; The circulation monitoring and correction strategy execution module is used to monitor the circulation of each file carrier sequence, obtain the watermark survival rate, visual fidelity, and circulation error rate of its corresponding secondary watermark embedding results, construct a correction strategy selection rule, and execute the corresponding secondary correction strategy according to the correction strategy selection rule.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) In the present invention, by simultaneously collecting the physical layout features and semantic features of the file, generating a comprehensive file feature map, and dynamically adjusting the watermark strength in combination with the rank screening threshold; introducing a dual-index evaluation system of visual fidelity and watermark extraction degree, and combining label classification with a multi-objective optimization function, the dynamic calibration of the watermark embedding position is realized; (2) In the present invention, by monitoring the watermark survival rate, visual fidelity, and circulation error rate, constructing a correction strategy selection rule including position iterative correction or sequence correction; and combining the correlation analysis of the watermark survival rate and the circulation error rate; not only realizes the dynamic maintenance of the entire life cycle of the watermark, but also realizes the dynamic selection of the correction strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic diagram of the method framework structure of the present invention; Figure 2 is a schematic diagram of the system framework structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0018] Embodiment 1: Please refer to Figure 1 , this application provides a digital file secure circulation tracking method based on dynamic watermarking, including; Step 1: Collect the physical layout features and semantic features of the digital archive data and generate an archive feature map. Based on the rank screening threshold and watermark strength corresponding to each archive feature map, construct each archive carrier sequence, and perform a watermark embedding operation on it to obtain an initial watermark embedding result; Specifically, the physical layout features are the visual structure and spatial distribution attributes presented after the digitization of the archive, including the position coordinates of text blocks, font sizes, and paragraph spacings; the semantic features refer to the context information carried by the archive content, including: text keywords, text themes, and text semantic relationships; The principle of generating the archive feature map is as follows: Based on a pre-trained convolutional neural network (CNN) model, the physical layout features and semantic features are weighted and fused to obtain the archive feature map denoted as , where R represents the set of real numbers of the feature data in the archive feature map, n is the number of archives, g and w are the height and width of the feature map, and d is the number of channels of the feature map; the weights are obtained through training and learning to ensure that the fused feature map can better express the comprehensive features of the archive; the archive feature map can simultaneously contain the physical layout information and semantic information of the archive; perform normalization processing on the generated archive feature map to ensure that the feature values are within a reasonable range.

[0019] It should be noted that obtaining the archive feature map based on the convolutional neural network (CNN) model as described above is a prior art and will not be elaborated too much in this embodiment; Constructing the archive carrier sequence based on the rank screening threshold and watermark strength corresponding to each archive feature map includes: For each archive feature map , where i is the identifier of each archive feature map and 1, 2,..., n; n is the number of archives; combine into the total number of spatial positions N = , to obtain the two-dimensional matrix , where N is the total number of spatial positions; For each , perform singular value decomposition to obtain ; in the formula, and represent matrices of order N and d respectively, the superscript T represents the conjugate transpose of the matrix, is the matrix, the main diagonal is composed of singular values , P is the theoretical maximum rank, and the remaining elements are all 0, and satisfy , Each column vector of is the eigenvector of the matrix , and they are mutually orthogonal, called the left singular vectors of the matrix . Similarly, = { , ,..., Each column vector of} is a matrix The eigenvectors of are mutually orthogonal, called matrices The right singular vectors of ; Based on the above decomposition process, the total information of the archive feature map is obtained , where P is the theoretical maximum rank and j is the index variable of P; The minimum value of all information in the feature map is defined as the rank screening threshold , and the front The cumulative energy ratio of the singular values ​​is compared with the preset threshold ; when , keep the previous The sub-regions corresponding to the singular values As the characteristic sub-region, that is, the watermark embedding region; when , retain all singular values, and do not process those that have no analytical value; It should be noted that in this example, the case where the rank filtering threshold is greater than the threshold usually represents a part of the feature map with higher information content and importance, while the case where the rank filtering threshold is less than the threshold may increase the calculation complexity and time cost. By ignoring these minor parts, the calculation process can be simplified.

[0020] Based on preset watermark strength And introduce the rank screening threshold Get the dynamic adjustment strength coefficient , according to the formula ; Where P is the theoretical maximum rank; The characteristic sub-regions corresponding to each archive feature map are converted into a one-dimensional carrier sequence that can be embedded with a watermark and recorded as the archive carrier sequence , and combined with the dynamic adjustment intensity coefficient corresponding to each characteristic sub-region , sort and splice to get the global archive vector sequence ;in Indicates the number of archive carrier sequences; Perform watermark embedding operation on each archive carrier sequence to obtain an initial watermark embedding result; specifically, the watermark sequence corresponding to each archive carrier sequence is based on the preset , and use watermark embedding technology to embed the watermark sequence corresponding to each archive carrier sequence Embedded into the corresponding sub-region of each archive carrier sequence Get the initial watermark embedding result; It should be noted that the generation of the above-mentioned preset watermark sequence is a prior art such as formation through spread spectrum technology, and the specific formation process is not described in detail.

[0021] Step 2: Collect the sequence of each archival carrier and its corresponding initial watermark embedding result. Through data analysis, obtain the corresponding visual fidelity and watermark extraction degree for each initial watermark embedding result, and conduct a watermark position evaluation to obtain the non-compliance and compliance labels of the archival carrier sequence; The sequences of each archival carrier are the original digitalized archives without watermark embedding operation; The initial watermark embedding results are the digitalized archives of the watermark-containing carriers after the watermark embedding operation; Based on the sequences of each archival carrier and the initial watermark embedding results of each calculate the structural similarity index, defined as the visual fidelity , according to the formula ; where a and b are the archival image blocks of the sequences of each archival carrier and the initial watermark embedding results respectively , and are the means of a and b respectively, and are the standard deviations of a and b respectively, is the covariance of a and b, and are the stability coefficients; It should be noted that the preset of the above stability coefficient can be based on the formula: and , where and are preset small constants, and L is the preset pixel dynamic range of the archival image block; Conduct a watermark extraction operation on each initial watermark embedding result to obtain the extracted watermark sequence and the original watermark sequence corresponding to each initial watermark embedding result, and calculate the watermark extraction degree , according to the following formula: ; where is the total number of digitalized archives, is the length of the watermark sequence of the digitalized archives q is the index of the element in the watermark sequence, is the q-th element in the watermark sequence of the digitalized archives, is the q-th element in the extracted watermark sequence of the digitalized archives, is obtained according to the mean formula as the mean of the original watermark sequence of the i-th digitalized archive, The mean of the watermark sequence for extracting the watermark from the i-th digital file

[0022] It should be noted that the above original watermark sequence is the watermark information initially embedded in the file, usually generated by a watermark generation algorithm and embedded in the file through an embedding method. The extracted watermark sequence is the watermark information extracted from the file; Based on the above sequence of each file carrier and the initial watermark embedding results of each corresponding visual fidelity and watermark extraction degree , a watermark position evaluation model is constructed based on the preset visual fidelity and the compliance intervals of the watermark extraction degree, specifically as shown in the following formula: ; In the formula, , respectively refer to the compliance intervals of the preset visual fidelity and the watermark extraction degree, and their specific values are set by researchers according to experience, or can be dynamically set according to the mean of historical data plus or minus a certain multiple of the standard deviation; Among them, the evaluation flag contains values of 1 or -1, indicating that the watermark position of the corresponding initial watermark embedding result is normal or the watermark position status is abnormal; Generate a prompt for the normal watermark position of the initial watermark embedding result and the normal initial watermark embedding result according to the value of 1 in the evaluation flag, and generate a prompt for the abnormal watermark position of the initial watermark embedding result and the abnormal initial watermark embedding result according to the value of -1 in the status flag; Automatically implement abnormal traceability verification for the abnormal watermark position of the initial watermark embedding result and the abnormal initial watermark embedding result generated according to the value of -1 in the status flag according to the value of -1 in the status flag; Among them, according to the value of -1 in the status flag, the calculated visual fidelity and watermark extraction degree are compared with their corresponding compliance intervals of the preset visual fidelity and the watermark extraction degree , for comparison and judgment; If the visual fidelity or the watermark extraction degree does not belong to its corresponding compliance interval , , an abnormal label indicating that the initial watermark embedding result does not meet the standard is generated; If the visual fidelity and the watermark extraction degree belong to its corresponding compliance interval , , a label indicating that the initial watermark embedding result meets the standard is generated; Step 3: Collect the labels with unqualified initial watermark embedding results, form a set of sequences of unqualified file carrier, combine their corresponding visual fidelity and watermark extraction degree, construct a multi-objective optimization function, and output the optimal watermark position to obtain the secondary watermark embedding result; And based on traversing to obtain the labels with unqualified initial watermark embedding results and the sequences of each file carrier corresponding to the labels with unqualified initial watermark embedding results, form a set of sequences of unqualified file carrier; Based on the visual fidelity and watermark extraction degree indexes of each set of unqualified sequences of file carrier, construct a multi-objective optimization function, specifically: Construct the visual fidelity sub-function And the watermark extraction degree sub-function , according to the formula , where is the sequence of file carrier at the candidate watermark position, is the original sequence of file carrier; The watermark extraction degree sub-function , where is the watermark extracted from the image after watermark embedding, is the original watermark sequence; Based on the above functions, construct a multi-objective optimization function. The constructed multi-objective optimization function is: ; In the formula, and respectively represent the weight coefficients corresponding to each sub-function, and their specific values are dynamically determined by researchers according to actual needs; And record that the watermark embedding position does not exceed the preset boundary D of the file carrier as x ∈ D, and the watermark strength does not exceed the preset maximum watermark strength as , as a constraint condition; Based on the above multi-objective optimization function and constraint conditions, use the Pareto front solution algorithm to obtain the optimal watermark position, specifically including: Adopt the Pareto front solution algorithm, randomly generate a certain number B of initial populations, and encode each individual (candidate solution) in the population as the watermark embedding position x and the watermark strength , using real number coding, randomly generate the position x during initialization to form the initial population E0, ensure that the gene values of each chromosome are within the allowed range and meet the basic constraint conditions; For each individual in the population, calculate its objective sub-function values and ; Compare the dominance relationships among all individuals: If the individual is not inferior to in all objectives, and is strictly better in at least one objective, then Dominate ; Divide the population into multiple front layers: Layer 1: All non-dominated solutions Layer 2: Solutions dominated by Layer 1 but not by other layers, and so on; Crowding degree calculation: Calculate the crowding degree for solutions within the same front layer to measure the distribution density of solutions in the objective space and avoid selecting overly concentrated solutions; Generate a new population through the three basic operations of selection, crossover, and mutation of the genetic algorithm; According to the preset iteration termination condition, i.e., the preset maximum number of iterations, if reached, terminate the algorithm; when the termination condition is met, retain all non-dominated solutions (the first-layer front) to form the Pareto front; And select from the Pareto front the solution that maximizes the multi-objective optimization function as the optimal solution, i.e., the optimal watermark position ; Based on the above optimal watermark position perform watermark embedding operation on the corresponding archive carrier sequence to obtain the secondary watermark embedding result; It should be noted that the solution algorithm based on the Pareto front provides a scientific and efficient solution for the selection of watermark embedding positions through multi-objective collaborative optimization, global search, and constraint handling.

[0023] Step 4: Collect the watermark survival rate, visual fidelity, and transfer error rate of the corresponding secondary watermark embedding results based on the transfer situation of each archive carrier sequence, and construct a correction strategy selection rule, and execute the corresponding secondary correction strategy based on the correction strategy selection rule; the secondary correction strategies include iterative correction of watermark embedding positions, correction of archive carrier sequences, and no correction; Collect the visual fidelity, watermark survival rate, and transfer error rate of the corresponding secondary watermark embedding results for each archive carrier sequence during the transfer process; the watermark survival rate is the proportion of watermarks that can still be correctly extracted after transfer, and its acquisition logic is: calculate the ratio of the number of correctly extracted watermarks to the total number of watermarks; the transfer error rate is the error rate caused by watermark embedding during the transfer of archives, and its acquisition logic is: calculate the ratio of the number of archives with errors to the total number of archives; Based on the visual fidelity of the corresponding secondary watermark embedding results for each archive carrier sequence during the transfer process , watermark survival rate , transfer error rate , construct a fuzzy rule table For example, "High", "Low", "Medium" for visual fidelity, watermark survival rate, and transfer error rate; Formulate a set of fuzzy rules to describe the influence of different input variables on the output variable. The definition of the rules can be based on professional knowledge or obtained through data analysis and experiments. For example: Label the visual fidelity as , label the watermark survival rate as , label the transfer error rate as , label the correction strategy selection rule as Rfcs, including the iterative correction of the watermark embedding position, labeled as , record the correction of each archival carrier sequence as , label no correction as

[0024] Then it can be defined as: Rule 1: IF ( is Low) AND ( is Low)AND( is High) THEN (Rfcs is ) Rule 2: IF (U is High) AND (U is High) AND ( is Low) THEN (Rfcs is ) Rule 3: IF (U is Low) AND (U is Medium) AND ( is Low) THEN (Rfcsis ) It should be noted that the division of the fuzzy set can be adjusted according to the actual situation. For example, although this embodiment takes three fuzzy sets as an example, in fact, the visual fidelity, watermark survival rate, and transfer error rate can be divided into more than three sets to facilitate the better selection of the correction strategy.

[0025] Furthermore, for the judgment of high, medium, and low of the visual fidelity, watermark survival rate, and transfer error rate, thresholds can be set according to the actual situation for judgment. For example, when the visual fidelity value exceeds 45%, it is labeled as "High", when the watermark survival rate value is lower than 75%, it is labeled as "Low", and when the transfer error rate is between 10% - 35%, it is labeled as "Medium", etc., which will not be elaborated here.

[0026] Statistically, for the correction strategy selection rule of " ", the iterative correction of the watermark embedding position is to return the corresponding archival carrier sequence to step two for re-evaluation of the optimal position; Statistics is " " The correction strategy selection rule, and the specific correction for each file carrier sequence correction is as follows: S1: Based on each evaluation index and the initial construction parameters corresponding to each file carrier sequence, establish an association model; the evaluation indexes are visual fidelity , watermark survival rate , transfer error rate , and the initial construction parameters corresponding to each file carrier sequence are initial construction parameters including rank screening threshold , watermark strength , feature map node connection density ; where r represents the number of carrier sequences to be corrected; S2: Construct an association model between evaluation indexes and construction parameters ; S3: Calculate the difference between each evaluation index visual fidelity , watermark survival rate , transfer error rate and its corresponding preset threshold , , to obtain its corresponding deviation value , , , and calculate the correction coefficient matrix according to each deviation value , where Q represents the proportional coefficient of each parameter to be adjusted; It should be noted that the above calculation of the correction coefficient can adopt various methods, such as linear interpolation, non-linear function, etc., to form a simple linear interpolation; S4: According to the correction coefficients corresponding to each construction parameter, update the initial construction parameters corresponding to each file carrier sequence to obtain updated construction parameters , , ; And reconstruct each file carrier sequence corresponding to each updated construction parameter to generate a corrected carrier sequence; It should be noted that the specific implementation steps of the above reconstruction of each file carrier sequence corresponding to each updated construction parameter to generate a corrected carrier sequence have been described in detail in step one of this embodiment, and will not be elaborated here; In this embodiment, by simultaneously collecting the physical layout features and semantic features of the files, a comprehensive file feature map is generated, and the watermark intensity is dynamically adjusted in combination with the rank screening threshold; and a dual-index evaluation system of visual fidelity and watermark extraction degree is introduced, combined with label classification and a multi-objective optimization function; to achieve dynamic calibration of the watermark embedding position; by monitoring the watermark survival rate, visual fidelity, and transfer error rate, a correction strategy selection rule is constructed, including position iterative correction or sequence correction; and combined with the correlation analysis of the watermark survival rate and transfer error rate; not only realizes the dynamic maintenance of the watermark throughout its life cycle, but also realizes the accurate positioning of the leakage nodes.

[0027] Embodiment 2: Please refer to Figure 2 , the digital file security transfer tracking system with dynamic watermark includes an initial watermark embedding module for files, a preliminary watermark position evaluation module, a secondary watermark embedding module, a transfer monitoring and correction strategy construction module; The initial watermark embedding module for files is used to collect the physical layout features and semantic features of the digital file data and generate a file feature map, construct each file carrier sequence based on the rank screening threshold and watermark intensity corresponding to each file feature map, and perform a watermark embedding operation on it to obtain the initial watermark embedding result; The preliminary watermark position evaluation module is used to evaluate the watermark position of each initial watermark embedding result based on each file carrier sequence and its corresponding initial watermark embedding result, and obtain the visual fidelity and watermark extraction degree corresponding to it through data analysis, and obtain the unqualified and qualified labels of the file carrier sequence; The secondary watermark embedding module is used to collect the unqualified labels of each initial watermark embedding result, form a set of unqualified file carrier sequences, combine their corresponding visual fidelity and watermark extraction degree, construct a multi-objective optimization function, and output the optimal watermark position to obtain the secondary watermark embedding result; The transfer monitoring and correction strategy execution module is used to monitor the transfer situation of each file carrier sequence, obtain the watermark survival rate, visual fidelity, and transfer error rate corresponding to each secondary watermark embedding result, and construct a correction strategy selection rule, and execute the corresponding secondary correction strategy according to the correction strategy selection rule.

[0028] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0029] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for tracking the safe circulation of digital archives based on dynamic watermarks, characterized in that: include: The physical layout features and semantic features of digital archive data are collected and archive feature graphs are generated. Based on the rank screening threshold and watermark strength corresponding to each archive feature graph, each archive carrier sequence is constructed, and a watermark embedding operation is performed on it to obtain the initial watermark embedding result. Based on each archive carrier sequence and its corresponding initial watermark embedding result, the corresponding visual fidelity and watermark extraction degree are obtained through data analysis to evaluate the watermark position of each initial watermark embedding result, and obtain the archive carrier sequence substandard and standard labels; Collect the labels of each initial watermark embedding result that does not meet the standard, and form a set of substandard archive carrier sequences, combined with their corresponding visual fidelity and watermark extraction degree, construct a multi-objective optimization function and output the optimal watermark position to obtain the secondary watermark embedding result; Monitor the circulation of each archive carrier sequence, obtain the watermark survival rate, visual fidelity, and circulation error rate of each secondary watermark embedding result, and construct correction strategy selection rules, and execute the corresponding secondary correction strategy according to the correction strategy selection rules; The secondary correction strategy includes iterative correction of the watermark embedding position, correction of each archive carrier sequence and no correction required.

2. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: The method comprises collecting physical layout features and semantic features of digital archive data and generating an archive feature map: the physical layout features are text block position coordinates, font size, paragraph spacing and the semantic features are text keywords, text themes, and text semantic relationships; The generating of the archive feature map is as follows: based on the pre-trained convolutional neural network model, the physical layout features and the semantic features are weightedly integrated to obtain the archive feature map denoted as M and , where R represents the real number set of feature data in the archive feature map, n is the number of archives, g and w are the height and width of the feature map, and d is the number of channels of the feature map.

3. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: The archive carrier sequence is constructed based on the rank screening threshold and watermark strength corresponding to each archive feature graph, specifically: For each profile feature map , where i is the identifier of each archive feature map and 1, 2, ..., n; n is the number of files; Combined into the total number of spatial locations N= , and get a two-dimensional matrix , N is the total number of spatial positions; For each Perform singular value decomposition to obtain ; In the formula, and Represent N-order and d-order matrices respectively, and the superscript T represents the conjugate transpose of the matrix. for The matrix, with the main diagonal composed of Singular Values Composition, P is the theoretical maximum rank, the remaining elements are 0, and satisfy , Each column vector of the matrix The eigenvectors of are mutually orthogonal, called matrices The left singular vectors of , similarly, ={ , ,..., Each column vector of} is a matrix The eigenvectors of are mutually orthogonal, called matrices The right singular vectors of ; Based on the above decomposition process, the total information of the archive feature map is obtained , where P is the theoretical maximum rank and j is the index variable of P; The minimum value of all information in the feature map is defined as the rank screening threshold , and the front The cumulative energy ratio of the singular values ​​is compared with the preset threshold ; when , keep the previous The sub-regions corresponding to the singular values As the characteristic sub-region, that is, the watermark embedding region; when , retain all singular values, and do not process those that have no analytical value; Based on preset watermark strength And introduce the rank screening threshold Get the dynamic adjustment strength coefficient , according to the formula ; Where P is the theoretical maximum rank; The characteristic sub-regions corresponding to each archive feature map are converted into a one-dimensional carrier sequence that can be embedded with a watermark and recorded as the archive carrier sequence , and combined with the dynamic adjustment intensity coefficient corresponding to each characteristic sub-region , sort and splice to get the global archive vector sequence ;in Indicates the number of archive carrier sequences; Perform watermark embedding operation on each archive carrier sequence to obtain the initial watermark embedding result.

4. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: Collect each archive carrier sequence and its corresponding initial watermark embedding result, and obtain the corresponding visual fidelity and watermark extraction degree through data analysis. Specifically: each archive carrier sequence It is the original digitized file without watermark embedding operation; The initial watermark embedding result A digitized file containing a watermark carrier after a watermark embedding operation; Based on the sequence of each archive carrier And the initial watermark embedding results Visual fidelity is calculated based on the structural similarity formula ; For each initial watermark embedding result Perform watermark extraction operation to obtain the extracted watermark sequence and original watermark sequence corresponding to each initial watermark embedding result, and calculate the watermark extraction degree , according to the following formula: ;in, is the total number of digitized archives, for Watermark sequence length of digitized archives q is the index of the element in the watermark sequence, for The qth element in the watermark sequence of the digitized archive, for The qth element in the extracted watermark sequence of the digitized archive, is the mean of the original watermark sequence of each digitized file, The mean of the watermark sequence for extracting the watermark for the i-th digitized archive.

5. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: The watermark position evaluation of each initial watermark embedding result specifically includes: based on each archive carrier sequence And the initial watermark embedding results Corresponding visual fidelity and watermark extraction degree , and the preset visual fidelity and watermark extraction degree reaching standard interval to build the watermark position evaluation model, which is specifically shown in the following formula: ; In the formula, , They refer to the preset visual fidelity and watermark extraction standard intervals respectively. 1 and -1 are evaluation marks, which respectively indicate that the watermark position of the initial watermark embedding result is normal or the watermark position state is abnormal.

6. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 5 is characterized in that: The archive carrier sequence non-compliant and compliant labels include: The visual fidelity obtained will be calculated based on the value of evaluation flag -1 and watermark extraction degree The corresponding preset visual fidelity and watermark extraction degree range , Make comparative judgments; When visual fidelity or watermark extraction degree Does not belong to the corresponding standard range , , then an abnormal label is generated if the initial watermark embedding result does not meet the standard; When visual fidelity And watermark extraction degree Belongs to the corresponding target range , , then the initial watermark embedding result reaches the standard label.

7. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: Constructing a multi-objective optimization function and outputting the optimal watermark position to obtain the secondary watermark embedding result includes: traversing to obtain each initial watermark embedding result non-compliant label and each archive carrier sequence corresponding to each initial watermark embedding result non-compliant label to form a non-compliant archive carrier sequence set; Based on the visual fidelity and watermark extraction degree of each archive carrier sequence that does not meet the standard, a visual fidelity sub-function is constructed And watermark extraction function And introduce the preset weight coefficient to obtain the multi-objective optimization function ; Combined with the constraint conditions: the watermark embedding position does not exceed the preset archive carrier boundary D, denoted by x ∈ D, and the watermark intensity does not exceed the preset maximum watermark intensity, denoted by , Based on the above multi-objective optimization function and constraints, the optimal watermark position is obtained by using the Pareto frontier solution algorithm: Using the Pareto front solution algorithm, for each individual in the population (candidate solution) Encoded as watermark embedding position x and watermark strength , using real number coding and calculating its target sub-function value and ; Compare the dominance relationships among all individuals: When individual Outperforms on all targets ,but Dominate ; Divide the population into multiple frontier layers and perform crowding calculations. Calculate the crowding for the solutions in the same frontier layer. According to the crowding, generate a new population through the three basic operations of genetic algorithm: selection, crossover, and mutation. According to the preset iteration termination condition, i.e. the preset maximum number of iterations, all non-dominated solutions are retained to form a Pareto optimal solution; And take the Pareto optimal solution as the optimal watermark position ; Based on the above optimal watermark position Perform watermark embedding operation on the corresponding archive carrier sequence to obtain the secondary watermark embedding result.

8. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 1 is characterized in that: The construction of correction strategy selection rules is specifically as follows: watermark survival rate, visual fidelity, and flow error rate are defined as input variables, and they are divided into different fuzzy sets respectively; Define the correction strategy selection rule as the output variable and divide it into fuzzy sets; Formulate fuzzy rules to describe the impact of watermark survival rate, visual fidelity, and flow error rate on the correction strategy selection rules; Perform fuzzy reasoning based on fuzzy rules to determine the results of the revised strategy selection rules; The correction strategy selection rules include iterative correction of watermark embedding position, archive carrier sequence correction and no correction required.

9. The method for secure circulation and tracking of digital archives based on dynamic watermark according to claim 7 is characterized in that: Iterative correction of watermark embedding position is to iteratively correct the position of watermark embedding, and to return the corresponding archive carrier sequence to step 2 to re-evaluate the optimal position; The sequence of each file carrier is modified to modify the sequence of each file carrier, specifically: S1: Establish an association model based on the initial construction parameters corresponding to each evaluation indicator and each archive carrier sequence; the evaluation indicator is visual fidelity , watermark survival rate , Flow error rate The initial construction parameters corresponding to each archive carrier sequence are the initial construction parameters including the rank screening threshold , Watermark Strength , feature graph node connection density ; Where r represents the number of corresponding vector sequences that need to be corrected; S2: Construct a correlation model between evaluation indicators and construction parameters ; S3: Visual fidelity of each evaluation indicator , watermark survival rate , Flow error rate The corresponding preset threshold , , Calculate the difference to get the corresponding deviation value , , , the correction coefficient matrix is ​​calculated by linear interpolation according to each deviation value , where Q represents the proportional coefficient of each parameter to be adjusted; S4: According to the correction coefficients corresponding to the construction parameters, the initial construction parameters corresponding to the archive carrier sequences are updated to obtain the updated construction parameters. , , ; And based on each updated construction parameter, the corresponding archive carrier sequences are reconstructed to generate a revised carrier sequence.

10. A system for tracking the safe circulation of digital archives with dynamic watermarks, used to implement the method for tracking the safe circulation of digital archives with dynamic watermarks as described in any one of claims 1 to 9, characterized in that: Archive initial watermark embedding module, watermark position initial assessment module, secondary watermark embedding module, circulation monitoring and correction strategy construction module; The archive initial watermark embedding module is used to collect the physical layout features and semantic features of digital archive data and generate archive feature maps. Based on the rank screening threshold and watermark strength corresponding to each archive feature map, each archive carrier sequence is constructed, and a watermark embedding operation is performed on it to obtain the initial watermark embedding result. The watermark position initial evaluation module is used to evaluate the watermark position of each initial watermark embedding result based on each archive carrier sequence and its corresponding initial watermark embedding result. The corresponding visual fidelity and watermark extraction degree are obtained through data analysis to obtain the non-compliant and compliant labels of the archive carrier sequence. The secondary watermark embedding module is used to collect the labels of each initial watermark embedding result that does not meet the standard, and form a set of substandard archive carrier sequences, combined with their corresponding visual fidelity and watermark extraction degree, to construct a multi-objective optimization function and output the optimal watermark position to obtain the secondary watermark embedding result; The circulation monitoring and correction strategy execution module is used to monitor the circulation of each archive carrier sequence, obtain the watermark survival rate, visual fidelity, and circulation error rate of each corresponding secondary watermark embedding result, and construct correction strategy selection rules, and execute the corresponding secondary correction strategy according to the correction strategy selection rules.

Images