Method and system for embedding and extracting frequency domain-based watermarks in table data

Abstract

The invention provides a system and a method for embedding frequency domain-based watermarks in table data. In some methods, a covariance matrix is calculated for the tabular data; projecting a row vector of the table data to two standard orthogonal vectors e1 and e2 of the covariance matrix; determining a signal space for the projected row vector; finding the frequency f * with the lowest power from the signal space; and embedding a watermark in the table data according to the frequency f *.

Description

[0001] This application claims priority to the Chinese patent application filed with the U.S. Patent Office on October 20, 2023, with application number 18 / 491,517 and application name “METHODS AND SYSTEMS FOR EMBEDDING AND EXTRACTING FREQUENCY DOMAIN-BASED WATERMARKS IN TABULAR DATA”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present invention relates to digital data management, and more particularly to a method and system for embedding and extracting frequency domain based watermarks in tabular data. Background Art

[0003] Data is becoming increasingly important in today’s society, especially with the rise of machine learning in all aspects of our lives. As data becomes more important, the proper use, legal redistribution, and ownership of data will become an integral part of our society. However, digital assets are different from physical assets in that they can be easily copied, modified, and shared between parties. This raises the question of intellectual property ownership of digital assets. While it is not easy to prevent the copying, modification, and sharing of digital assets, other measures can be taken to strongly dissuade data consumers from performing any such actions.

[0004] One such method is data watermarking, which comes in many forms and has many different applications. Data watermarking is a method of embedding a mark or message into a piece of data, making it difficult to remove. There are two types of watermarking algorithms: blind and non-blind. Blind watermarking refers to watermarking algorithms that do not require the original unwatermarked data at the time of extraction, while non-blind watermarking requires the original unwatermarked data at the time of extraction. Since the original data may not be available when the watermark needs to be extracted, blind watermarking is a more attractive option.

[0005] Watermarking algorithms can also be categorized into zero-bit watermarking and multi-bit watermarking. Zero-bit watermarking is actually a binary classification problem. Watermarking algorithms process data and set a "flag" in the data, marking the flag as "watermarked". Watermark extraction algorithms simply classify whether a given piece of data has this "flag" set. Multi-bit watermarking embeds a custom message in the data. Then, the watermark extraction algorithm checks whether a given piece of data has been watermarked, and if so, recovers the embedded message.

[0006] Data comes in a variety of different formats, such as videos, images, tables, text, and 3D models, etc. Current research has not yet found a single algorithm that can handle all data formats while being resistant to attacks and not causing significant damage to the value of assets. Therefore, a different algorithm must be used for each data format.

[0007] In one current approach, watermark embedding methods utilize hash functions and a certain degree of bit manipulation. To embed a watermark, the hash method typically hashes each row of table data and divides the data into a specified number of partitions based on the hash result. Each partition has one or more bits that are embedded by flipping a bit determined by a certain attribute of the partition.

[0008] In another approach, the watermark embedding method divides the data into multiple partitions and changes the statistical properties of the data set. For example, the algorithm can change the mean or variance of the data in the selected partitions according to the watermark processing parameters. This statistical change is then detected at the time of extraction.

[0009] In another method, the original data is transformed into a time signal, and then the signal is transformed from the time domain to the frequency domain using methods such as DCT and DFT and techniques such as Fourier transform. Then, the watermark is embedded in the frequency domain by modifying the coefficients of the transformed signal. The watermark can be extracted from the watermarked signal by inverse transforming the modified signal.

[0010] Current bit manipulation methods are vulnerable to modification and deletion attacks. Additionally, most of these prior art techniques rely on tables containing primary keys that remain unchanged after an attack. This means that if an attacker decides to remove the primary key column, the watermark is virtually impossible to recover. Similarly, a simple attack to remove the watermarked bit is to add a small constant to every value in the data set. Since bit manipulation typically only changes the least significant bit (to avoid serious data corruption), adding a small constant is likely to erase the watermark bit.

[0011] Current statistical methods rely heavily on assumed properties of the data itself. This means that sometimes, even if the watermarked data is not subject to any attack, the watermark extraction algorithm may fail. For example, if the watermark embedding algorithm relies on data variance within a certain range, it may produce a pathology dataset with variance far outside this range, thus affecting the quality of the generated watermarked data.

[0012] In some current schemes, frequency domain methods are used to embed watermarks in video, audio, and image data. However, as mentioned above, these methods are only applicable to multimedia data and are not directly applicable to tabular data. For this reason, since multimedia data is a continuous space and is much more resistant to noise than discrete tabular data, the embedding algorithm can freely change this data and the degree of change is much greater than that of discrete data. Therefore, even if these methods are directly applied to tabular data, the watermarking method may severely damage the data, thus affecting the effectiveness of the watermarked data. Summary of the invention

[0013] According to a first aspect of the present invention, a computer-implemented method for embedding a frequency-domain-based watermark in tabular data is provided. The method comprises: calculating a covariance matrix for the tabular data; projecting a row vector of the tabular data onto two standard orthogonal vectors e1 and e2; determining a signal space for the projected row vector; and finding a frequency f with the lowest power from the signal space. * According to the frequency f * A watermark is embedded in the table data.

[0014] In some or all exemplary embodiments of the first aspect, determining the signal space includes using a binning algorithm.

[0015] In some or all exemplary embodiments of the first aspect, the signal space is generated as S=([x i ′,y i ]|i=1…N), where x is the projection of the tabular data on e1, y is the projection of the tabular data on e2, r is the embedded sinusoidal signal resolution, and b is the bin width determined by the binning algorithm.

[0016] In some or all exemplary embodiments of the first aspect, the method further includes: saving r, b, f*, e1 and e2 as watermark processing parameters.

[0017] In some or all exemplary embodiments of the first aspect, the finding includes using a domain-based transformation.

[0018] In some or all exemplary embodiments of the first aspect, the domain-based transform is a Lomb-Scargle transform.

[0019] In some or all exemplary embodiments, the two orthonormal vectors are normalized eigenvectors of the covariance matrix.

[0020] In some or all exemplary embodiments of the first aspect, the method further includes: determining the watermark matrix W as in, express The outer product of and e2.

[0021] In some or all exemplary embodiments of the first aspect, the watermarking data is determined as D'=D+W, where D is the table data.

[0022] In a second aspect of the present invention, a computer-implemented method for extracting a frequency-domain based watermark in tabular data is provided. The method comprises: receiving watermark parameters; calculating a covariance matrix for the tabular data; projecting a row vector of the tabular data onto two standard orthogonal vectors e1 and e2; determining a signal space for the projected row vector; determining a frequency f specified in the watermark parameters; * The power; According to the frequency f * The power is used to extract a frequency domain based watermark from the table data.

[0023] In some or all exemplary embodiments of the second aspect, the method further includes: determining a false alarm probability.

[0024] In a third aspect of the present invention, a computing system for embedding a frequency domain-based watermark in tabular data is provided. The computing system comprises: one or more processors; a memory storing machine-readable instructions, wherein the machine-readable instructions, when executed by the one or more processors, enable the computing system to perform the following operations: calculate a covariance matrix for the tabular data; project a row vector of the tabular data onto two standard orthogonal vectors e1 and e2; determine a signal space for the projected row vector; find a frequency f* with the lowest power from the signal space; and embed a watermark in the tabular data according to the frequency f*.

[0025] In some or all exemplary embodiments of the third aspect, the machine-readable instructions, when executed by the one or more processors, cause the computing system to determine the signal space including using a binning algorithm.

[0026] In some or all exemplary embodiments of the third aspect, when the machine-readable instructions are executed by the one or more processors, the computing system generates the signal space as S=([x i ′,y i ]|i=1…N), where x is the projection of the tabular data on e1, y is the projection of the tabular data on e2, r is the embedded sinusoidal signal resolution, and b is the bin width determined by the parameter adjustment algorithm.

[0027] In some or all exemplary embodiments of the third aspect, when the machine-readable instructions are executed by the one or more processors, the computing system saves r, b, f*, e1, and e2 as watermark processing parameters.

[0028] In some or all exemplary embodiments of the third aspect, the machine-readable instructions, when executed by the one or more processors, cause the computing system to find the frequency f* using a domain-based transform.

[0029] In some or all of the exemplary embodiments of the third aspect, the domain-based transform is a Lomb-Scargle transform.

[0030] In some or all exemplary embodiments of the third aspect, the two standard orthogonal vectors are normalized eigenvectors of the covariance matrix.

[0031] In some or all exemplary embodiments of the third aspect, when the machine-readable instructions are executed by the one or more processors, the computing system performs the following operations: determining the watermark matrix W as in, express The outer product of and e2.

[0032] In some or all exemplary embodiments of the third aspect, when the machine-readable instructions are executed by the one or more processors, the computing system determines the watermark data as D′=D+W, where D is the table data.

[0033] In a fourth aspect of the present invention, a computing system for extracting a frequency domain-based watermark from tabular data is provided. The computing system comprises: one or more processors; a memory storing machine-readable instructions, wherein the machine-readable instructions, when executed by the one or more processors, cause the computing system to perform the following operations: receiving watermark parameters; calculating a covariance matrix for the tabular data; projecting a row vector of the tabular data onto two standard orthogonal vectors e1 and e2; determining a signal space for the projected row vector; determining the power of a frequency f* specified in the watermark parameters; and calculating a power of a frequency f* specified in the watermark parameters according to the frequency f * The power of is used to extract a frequency domain based watermark from the tabular data.

[0034] In some or all exemplary embodiments of the fourth aspect, the machine-readable instructions, when executed by the one or more processors, cause the computing system to determine a false alarm probability.

[0035] In a fifth aspect of the present invention, a non-transitory machine-readable medium tangibly storing executable instructions for execution by one or more processors is provided, wherein, in response to the execution of the one or more processors, the executable instructions, when executed by the one or more processors, cause the one or more processors to perform the following operations: calculate a covariance matrix for the tabular data; project a row vector of the tabular data onto two standard orthogonal vectors e1 and e2 of the covariance matrix; determine a signal space for the projected row vector; find a frequency f with the lowest power from the signal space * According to the frequency f * , embedding a watermark in the table data.

[0036] In a sixth aspect of the present invention, a non-transitory machine-readable medium tangibly storing executable instructions for execution by one or more processors is provided. When the executable instructions are executed by the one or more processors, the one or more processors perform the following operations: receiving watermark parameters; calculating a covariance matrix for the tabular data; projecting a row vector of the tabular data onto two standard orthogonal vectors e1 and e2 of the covariance matrix; determining a signal space for the projected row vector; determining a frequency f specified in the watermark parameters * The power; According to the frequency f * The power of is used to extract a frequency domain based watermark from the tabular data.

[0037] Other aspects and features of the present invention will become apparent to those skilled in the art by reading the following specific implementation description of the present application in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Reference is now made, by way of example, to the accompanying drawings which show exemplary embodiments of the present application.

[0039] Figure 1 is a schematic diagram of an architecture for embedding and extracting frequency-domain based watermarks in tabular data provided by exemplary embodiments described herein.

[0040] Figure 2 The exemplary embodiments described herein provide a general method for embedding and extracting frequency-domain based watermarks in tabular data.

[0041] Figure 3 The exemplary embodiments described herein provide Figure 2 Schematic diagram of the parameter adjustment process in the method of embedding and extracting frequency-domain based watermarks in tabular data.

[0042] Figure 4 The exemplary embodiments described herein provide Figure 3Schematic diagram of the parameter tuning algorithm used in the process.

[0043] Figure 5 The exemplary embodiments described herein provide Figure 2 Schematic diagram of the watermark embedding process in the method for embedding and extracting frequency domain based watermarks in data.

[0044] Figure 6 The exemplary embodiments described herein provide Figure 5 Schematic diagram of the watermark embedding algorithm used in the process.

[0045] Figure 7 The exemplary embodiments described herein provide Figure 2 Schematic diagram of the watermark extraction process in the method for embedding and extracting frequency domain based watermarks in data.

[0046] Figure 8 The exemplary embodiments described herein provide Figure 7 Schematic diagram of the watermark extraction algorithm used in the process.

[0047] Fig. 9 is a schematic diagram of a computing system for embedding and / or extracting a frequency-domain based watermark in and / or from tabular data provided by exemplary embodiments described herein.

[0048] Like reference numerals may be used in different drawings to denote like components. Unless otherwise specifically noted, items depicted in the drawings are not necessarily drawn to scale. DETAILED DESCRIPTION

[0049] The present invention is referenced to the accompanying drawings showing embodiments. However, many different embodiments may be used, and therefore the description should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to make the present application comprehensive and complete. In the accompanying drawings and the following description, the same reference numerals are used as much as possible to refer to the same elements, and apostrophes are used in alternative embodiments to represent similar elements, operations or steps. The separate boxes or shown separations of the functional elements of the illustrated systems and devices do not necessarily require the physical separation of these functions, because the communication between these elements can be carried out by message passing, function calls, shared memory spaces, etc. without any such physical separation. Therefore, the functions do not need to be implemented in physically or logically separated platforms, but for ease of explanation, these functions are shown separately herein. Different devices may have different designs, so that although some devices implement some functions in fixed-function hardware, other devices may implement these functions in a programmable processor having a code obtained from a machine-readable medium. Finally, unless the context clearly or inherently indicates, elements in the singular form may be plural, and vice versa.

[0050] The present invention describes exemplary embodiments of methods, systems, and machine-readable media for embedding and extracting frequency-domain based watermarks in tabular data. The goal of the tabular data watermarking method proposed by the present invention is to embed a watermark in tabular data that is relatively resistant to various attacks and does not affect the value of the generated watermarked data.

[0051] Reference below Figure 1 , a proposed architecture 20 consists of three different components, namely a parameter adjustment component 24, an embedding component 28, and an extraction component 32. The parameter adjustment component 24 receives the tabular data 36 and uses LSP to generate a set of locally optimal watermark parameters 40 for watermarking. The algorithm used by the parameter adjustment component 24 finds the lowest frequency in the signal space of the data set and embeds the watermark 44 in this frequency to ensure minimal impact on the data utility. The watermark 44 is encoded with a watermark message, such as an identifier of the licensor, a license transaction identifier, etc. The embedding component 28 then reconstructs the signal space using the watermark parameters generated by the parameter adjustment component 24. A sinusoidal signal is embedded in the tabular data 36 at the frequency given by the parameter adjustment component 24 to generate watermarked tabular data 48, and the watermark 44 is mapped to this frequency and securely stored somewhere. To extract the watermark 44 and read the corresponding message, the extraction component 32 receives the watermarked data set and a key stored by the data owner, and outputs a Boolean value that determines whether a given frequency is embedded in the watermarked tabular data 48. This three-step process ensures that the watermark 44 is embedded in a manner that minimizes the impact on the utility of the tabular data 36 and can be accurately extracted when needed.

[0052] The method disclosed in the present invention is applicable to any system that is used to provide data watermarking, including but not limited to tabular data. For example, the method disclosed in the present invention will be beneficial to data exchange platforms, such as Web service data exchange, because the owner of tabular data can protect the copyright of the data by embedding watermarks in the data. Using the disclosed method proposed by the present invention, parties who need to identify the ownership after selling the data can achieve this goal. The method disclosed by the present invention can be used in digital asset trading platforms and data markets. The method can also be used in machine learning dataset websites, which provide datasets for machine learning purposes.

[0053] In addition, the method proposed by the present invention does not rely on the primary key in the table data, while maintaining strong resistance to insertion, deletion and modification attacks.

[0054] Figure 2 A general method 100 for frequency domain based watermarking of tabular data is shown. In the method 100, parameters are adjusted for the data to be watermarked (110). The watermark is then embedded in the data using the adjusted parameters (120). Once the watermark is embedded in the data, the watermark can be extracted (130).

[0055] Figure 3 The parameter adjustment process 110 is shown for finding unique parameters that can be used to embed and extract watermarks in a data set. The first step is to calculate the covariance matrix and its corresponding eigenvectors (111). The row vectors of the data matrix are then projected onto the orthonormal basis e1, e2. d (112). e1 and e2 are the x and y projection vectors of the signal, respectively. Although any set of standard orthogonal basis vectors will produce an embedding, principal component analysis (PCA) shows that the variance is minimized when the data matrix is ​​projected onto the eigenvector with the smallest eigenvalue of the covariance matrix. d Defining it as this vector further reduces the changes to the watermarked data. On the other hand, the eigenvector with the largest eigenvalue will maximize the variance of the projected data. Defining e1 as this vector ensures that the signal space points are evenly distributed, thereby improving the quality of the embedding. The projection is calculated by multiplying the data matrix by e1 and e2. The value of the embedded sinusoidal resolution r is then determined using a binning algorithm and a bin width b to scale the data (113). Part of the parameter adjustment algorithm determines the relatively optimal bin width b by the following operation: each candidate bin width b is input into the binning algorithm and the optimality of the bin width is checked as described in the parameter adjustment algorithm.

[0056] Finally, the projected signal is subjected to a Lomb-Scargle transform to find the frequency with the lowest power, denoted as f * (114). Then, save r, b, and f as the only parameters. * , e1 and e2, wherein these parameters can be used to extract the watermark (115).

[0057] Figure 4 A version of the parameter adjustment algorithm 200 is shown and implementation details are provided. First, the covariance matrix D is calculated. T D(210). Project the row vectors of the data matrix D onto the standard orthogonal basis e1, e2 (220), where e1, e2 are D T Any two normalized eigenvectors of D. Since D T D is symmetric, so e1·e2 = 0. In the method proposed by the present invention, any standard orthogonal basis is applicable; however, eigenvectors are chosen because they are convenient to use and can be determined using only the data matrix.

[0058] The projection is calculated as:

[0059]

[0060] Generates the signal space as:

[0061] S=([x i ′,y i ]|i=1…N),

[0062] in,

[0063] r and b are determined by the lowest power frequency f * The power of (240) is determined.

[0064] Next, the Lomb-Scargle transform (a generalization of the discrete Fourier transform to allow for unequal sampling intervals) is used to find the frequency with the lowest power, f * ∈[f1,f2](250). Then, save the generated parameters r, b, f * , e1, e2 (260). These parameters are unique parameters ("secret keys") that can be used to extract the watermark later.

[0065] Figure 5The process of embedding a watermark in a data set is shown 120. The first step is to extract all continuous real number columns in the tabular data and use these real number columns to form a data matrix (121). The data matrix is ​​then normalized so that the mean of each column is 0 and the variance is 1 (122). Next, the covariance matrix is ​​calculated (123) and the row vectors of the data matrix are projected onto a standard orthogonal basis (124). This generates a series of discrete values ​​where each y coordinate represents a value on the axis spanned by e2. A binning algorithm is used on the signal space to improve the attack resistance of the resulting embedding (125). The embedded signal is represented as a vector whose elements are points on the sinusoidal signal. The watermark matrix is ​​the outer product of the embedded signal and the eigenvector with the smallest eigenvalue. The watermark matrix is ​​then added to the data matrix to produce a watermarked data matrix (126).

[0066] Reference below Figure 6 , Figure 6 A version of the embedding algorithm 300 is shown and implementation level details are provided. All consecutive real columns in the CSV are extracted to form a data matrix D, where each attribute forms a column vector in D (310). The data matrix D is then normalized so that each column has a mean of 0 and a variance of 1 (320). The covariance matrix D is calculated T D (330). Project the row vectors of the data matrix D onto the standard orthogonal basis e1, e2 (340). Calculate the projection as:

[0067]

[0068] Next, the signal space is generated as S = ([x i ′,y i ]|i=1…N), where The embedded signal is represented as:

[0069]

[0070] The watermark matrix is ​​calculated as:

[0071]

[0072] Then, the watermarked data matrix is ​​determined as

[0073] D′=D+W(370).

[0074] Once a watermark has been inserted into the data, it can be extracted using parameters.

[0075] Figure 7An extraction process 130 for extracting a watermark from a watermarked data set is shown. The first step is to project the watermarked data set onto a given orthonormal basis (131). This generates a signal space. A binning algorithm is then used to group the points in the signal space so that for all points sharing the same x-coordinate, all points are mapped to one coordinate using the average of the corresponding y-coordinates (132). Next, a Lomb-Scargle transform is used to find the power and false alarm probability for a given frequency f (133). This allows the watermark to be extracted from the watermarked data set (134).

[0076] Figure 8 A version of the extraction algorithm 400 is shown, and implementation details are provided. Assuming there is a watermarked data set D′=D+W, the data set is projected onto a given standard orthogonal basis e1, e2,

[0077]

[0078] Generates the signal space as:

[0079] S=([x i ′,y i ]|i=1…N),

[0080] in,

[0081]

[0082] For all points in S that share the same x-coordinate, map them to one coordinate by taking the average of their corresponding y-coordinates.

[0083] The Lomb-Scargle transform is then used to find the power and false alarm probability for a given frequency f (430) using the method described in "Assessing statistical significance of periodogram peaks" by RV Baluev, 2008 (arXiv:0711.0330).

[0084] Fig. 9 Various physical and logical components of an exemplary computing system 200 for frequency-domain based watermarking of tabular data provided by one embodiment of the present invention are shown. Although an exemplary embodiment of the computing system 200 is shown and discussed below, other embodiments may be used to implement the examples disclosed herein, and these examples may include components different from those shown. Fig. 9 A single instance of each component in computing system 200 is shown, but there may be multiple instances of each component shown.

[0085] The computing system 200 includes one or more processors 204, such as a central processing unit, a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a dedicated logic circuit, a tensor processing unit, a neural processing unit, a dedicated artificial intelligence processing unit, or a combination thereof. The one or more processors 204 may be collectively referred to as processors 204. The computing system 200 may include a display screen 208 for outputting data and / or information in some applications, but may not include it in other applications.

[0086] The computing system 200 includes one or more memories 212 (collectively referred to as "memory 212"), which may include volatile or non-volatile memory (e.g., flash memory, random access memory (RAM) and / or read-only memory (ROM)). The non-transitory memory 212 may store machine-executable instructions for execution by the processor 204. The memory 212 stores a set of machine-executable instructions 216, which define an application process for embedding and / or extracting a frequency-domain-based watermark (as described herein) from tabular data, and may be executed by the processor 204 to perform the method steps described herein. The memory 212 may include other machine-executable instructions for execution by the processor 204, such as machine-executable instructions for implementing an operating system and other applications or functions.

[0087] Memory 212 may also store other data, information, rules, policies, and machine-executable instructions described herein.

[0088] In some examples, computing system 200 may also include one or more electronic storage units (not shown), such as a solid state drive, a hard disk drive, a magnetic disk drive, and / or an optical disk drive. In some examples, one or more data sets and / or modules may be provided by an external memory (e.g., an external drive in wired or wireless communication with computing system 200), or by a transient or non-transient computer-readable medium. Examples of non-transient computer-readable media include RAM, ROM, erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, CD-ROM, or other portable memory. Storage units and / or external memory may be used in conjunction with memory 212 to implement data storage, retrieval, and caching functions of computing system 200.

[0089] For example, the components in the computing system 200 can communicate with each other via a bus. In some embodiments, the computing system 200 is a distributed computing system and can include multiple computing devices that communicate with each other via a network, and optionally include one or more other components. In some embodiments, the various operations described herein can be performed by different computing devices in a distributed system. In some embodiments, the computing system 200 is a virtual machine provided by a cloud computing platform.

[0090] Although the components for embedding and extracting frequency-domain based watermarks in tabular data are shown as being part of the computing system 200, it should be understood that a separate computing device may be used to embed and extract frequency-domain based watermarks.

[0091] Using the method described in this paper, the watermark embedding quality does not depend on the original data, nor on some properties of the data to guarantee high-quality embedding. The method has fast embedding and extraction speeds and was able to watermark at a rate of about 67K tuples / second when measured.

[0092] In addition to being used for copyright protection and proof of ownership, tabular data watermarking can also be applied to other forms of data. Since the method proposed in the present invention only requires discrete signals, there are many other forms of data that can be modified or simplified to discrete signals. For example, audio data can be converted into discrete signals and subjected to the watermarking process described in this article. Since the modification ratio using the method is small, the amount of noise in the generated watermarked audio can be minimized. For this reason, the method proposed in the present invention is also applicable to image data. Image data can be described as tabular data, where each column of pixel values ​​corresponds to an attribute in the tabular data. Therefore, images can be watermarked using the method proposed in this article. There are many forms of data that can ultimately be simplified to discrete signals. The method proposed in this article is applicable to all such data. Many other forms of data (e.g., images, audio, video) can be simplified to digital tabular data, and therefore, the watermarking method discussed can be used for these other forms of data.

[0093] As long as the amplitude A of the sinusoidal signal is a reasonable value, this method will hardly damage the original data, with the average modification of each column value being ±2A / π. The worst-case modification is expected to be ±A.

[0094] Additionally, the embedding algorithm is resistant to attacks that attempt to delete rows / columns of table data and does not rely on an immutable primary key.

[0095] Although in the above embodiment, the row vectors of the tabular data are projected onto two normalized eigenvectors of the covariance matrix, any two standard orthogonal vectors can be used for the projection. The eigenbasis causes minimal damage to the data utility while maintaining the embedding quality. Since these basis vectors are also part of the secret key required to extract the watermark, there is a trade-off because an attacker can easily calculate the eigenbasis and obtain part of the key.

[0096] Although in the above embodiment, the Lomb-Scargle transform is used to find the frequency with the lowest power from the signal space, any other suitable domain-based transform may be used instead of the Lomb-Scargle transform, such as discrete Fourier transform, discrete time Fourier transform, discrete cosine transform, etc.

[0097] The steps (also referred to as operations) in the flowcharts and drawings described herein are for illustration only. There may be many variations of these steps and / or operations without departing from the teachings of the present invention. For example, the steps may be performed in a different order, and steps may be added, deleted, or modified as needed.

[0098] In other embodiments, other modes may employ the same methods described herein.

[0099] Overview

[0100] Through the description of the above embodiments, the present invention can be implemented only by hardware, can be implemented by software and necessary general hardware platforms, and can also be implemented by a combination of hardware and software. For those of ordinary skill in the art, encoding the software for implementing the above method belongs to the scope of the present invention. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product. The software product can be stored in a non-volatile or non-transient storage medium, which can be an optical storage medium, a flash drive, or a hard disk. The software product includes many instructions that enable a computing device (a personal computer, a server, or a network device) to execute the method provided in the embodiments of the present invention.

[0101] The present invention also discloses all values ​​and sub-ranges within the disclosed range. In addition, although the systems, devices and processes disclosed and shown herein may include specific multiple elements, the systems, devices and components may be modified to include more or fewer elements in these elements. Although several exemplary embodiments are described herein, modifications, adjustments and other implementations are also possible. For example, the elements shown in the accompanying drawings may be replaced, added or modified, and the exemplary methods described herein may also be modified by replacing, reordering or adding steps to the disclosed methods.

[0102] Features from one or more of the above embodiments may be selected to create alternative embodiments consisting of sub-combinations of features not explicitly described above. Additionally, features from one or more of the above embodiments may be selected and combined to create alternative embodiments consisting of combinations of features not explicitly described above. Features suitable for such combinations and sub-combinations will be apparent to those skilled in the art upon review of the invention as a whole.

[0103] In addition, many specific details are set forth herein to provide a thorough understanding of the exemplary embodiments described herein. However, it will be appreciated by those of ordinary skill in the art that the exemplary embodiments described herein may be implemented without providing these specific details. In addition, well-known methods, processes, and elements are not described in detail to avoid obscuring the exemplary embodiments described herein. The subject matter described herein and set forth in the claims is intended to cover and encompass all suitable technical changes.

[0104] Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the invention as defined by the appended claims.

[0105] Without departing from the subject matter of the claims, the present invention may be implemented in other specific forms. The exemplary embodiments described are merely illustrative in all respects and are not restrictive. The present invention is intended to cover and encompass all suitable technical changes. Therefore, the scope of the present invention is described by the appended claims rather than by the above description. The scope of the claims should not be limited to the embodiments set forth in the examples, but should be given the broadest interpretation consistent with the entire description.

Claims

1. A computer-implemented method for embedding a frequency-domain based watermark in tabular data, characterized in that: include: Calculating a covariance matrix for the tabular data; Projecting the row vectors of the table data onto two standard orthogonal vectors e1 and e2; determining a signal space for said projected row vectors; Find the frequency f with the lowest power from the signal space * ; According to the frequency f * A watermark is embedded in the table data.

2. The computer-implemented method of claim 1, wherein: Determining the signal space includes using a binning algorithm.

3. The computer-implemented method of claim 2, wherein: The signal space is generated as: S=([x i ′,y i ]|i=1…N), in, x is the projection of the table data on e1, y is the projection of the table data on e2, r is the embedded sinusoidal signal resolution, b is the bin width determined by the binning algorithm.

4. The computer-implemented method of claim 3, wherein: Also includes: Save r, b, f*, e1, and e2 as watermark processing parameters.

5. The computer-implemented method of claim 1, wherein: The finding includes using a domain-based transformation.

6. The computer-implemented method of claim 1, wherein: The domain-based transform is the Lomb-Scargle transform.

7. The computer-implemented method of claim 1, wherein: The two orthonormal vectors are normalized eigenvectors of the covariance matrix.

8. The computer-implemented method of claim 3, wherein: Also includes: The watermark matrix W is determined as: in, in, express The outer product of and e2.

9. The computer-implemented method of claim 8, wherein: The watermarked data is determined as: D'=D+W, Wherein, D is the table data.

10. A computer-implemented method for extracting a frequency-domain based watermark from tabular data, characterized in that: include: Receive watermark parameters; Calculating a covariance matrix for the tabular data; Projecting the row vectors of the table data onto two standard orthogonal vectors e1 and e2; determining a signal space for said projected row vectors; Determine the frequency f specified in the watermark parameters * Power; According to the frequency f * The power is used to extract a frequency domain based watermark from the table data.

11. The computer-implemented method of claim 10, wherein: Also includes: Determine the false alarm probability.

12. A computing system for embedding a frequency domain based watermark in tabular data, characterized in that: The computing system comprises: one or more processors; a memory storing machine-readable instructions, wherein the machine-readable instructions, when executed by the one or more processors, cause the computing system to: Calculating a covariance matrix for the tabular data; Projecting the row vectors of the table data onto two standard orthogonal vectors e1 and e2; determining a signal space for said projected row vectors; Find the frequency f with the lowest power from the signal space * ; According to the frequency f * , embedding a watermark in the table data.

13. The computing system according to claim 12, characterized in that: The machine readable instructions, when executed by the one or more processors, cause the computing system to determine the signal space including using a binning algorithm.

14. The computing system according to claim 13, characterized in that: The machine-readable instructions, when executed by the one or more processors, cause the computing system to generate the signal space as: S=([x i ′,y i ]|i=1…N), in, x is the projection of the table data on e1, y is the projection of the table data on e2, r is the embedded sinusoidal signal resolution, b is the bin width determined by the binning algorithm.

15. The computing system according to claim 14, characterized in that: The machine-readable instructions, when executed by the one or more processors, cause the computing system to save r, b, f*, e1, and e2 as watermark processing parameters.

16. The computing system according to claim 12, characterized in that: The machine readable instructions, when executed by the one or more processors, cause the computing system to find the frequency f* using a domain-based transform.

17. The computing system according to claim 16, characterized in that: The domain-based transform is the Lomb-Scargle transform.

18. The computing system according to claim 12, characterized in that: The two orthonormal vectors are normalized eigenvectors of the covariance matrix.

19. The computing system of claim 14, wherein: The machine-readable instructions, when executed by the one or more processors, cause the computing system to: The watermark matrix W is determined as: in, in, express The outer product of and e2.

20. The computing system of claim 19, wherein: The machine-readable instructions, when executed by the one or more processors, cause the computing system to determine the watermarked data as: D'=D+W, Wherein, D is the table data.

Images