A multimedia watermark embedding and extraction method based on deep separable convolution
The multimedia watermark is preprocessed and embedded and extracted through the depth separation convolution model, which solves the problem of insufficient robustness and concealment in the existing technology, realizes more efficient watermark embedding and extraction, and enhances the copyright protection of digital media ability.
Patent Information
- Application Number
- CN202411337361.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-25
AI Technical Summary
The existing multimedia watermark embedding extraction methods have shortcomings in terms of robustness and concealment, and it is difficult to effectively protect the copyright of digital media, especially in the face of complex media and powerful attack methods.
Depth separable convolution model is adopted to preprocess and embed watermark information, including integer transformation, quantization processing, Z-shaped scanning conversion and restoration code generation, and multimedia audio watermark embedding and extraction with depth separable convolution model, and evaluate The watermark embed effect is used to determine whether the requirements are met.
It improves the robustness and concealment of the multimedia watermark embedding and extraction process, thereby improving the efficiency of watermark embedding and extraction, and can more effectively protect the copyright of digital media.
Smart Images

Figure CN118864218B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of multimedia watermark embedding and extraction, and in particular to a multimedia watermark embedding and extraction method of depth-separable convolution. Background Art
[0002] With the rapid development of information technology, digital media such as images, audio, and video have become an indispensable part of people's daily lives. However, the easy duplication and dissemination of digital media have brought great challenges to copyright protection. Although traditional encryption and signature technologies can protect the security of digital media to a certain extent, they still have some limitations in practical applications, such as the complexity of encryption algorithms, the high computational cost of the decryption process, and the ease of being cracked.
[0003] In order to solve these problems, digital watermark technology came into being. Digital watermark technology is an information hiding technology that embeds specific information (i.e., watermark) into digital media to achieve copyright protection and integrity verification of digital media. This technology can not only ensure the security of digital media during transmission and use, but also provide effective evidence for copyright disputes.
[0004] In recent years, the rapid development of deep learning technology has provided new ideas for digital watermarking technology. As an important algorithm in convolutional neural networks, deep separable convolution reduces the number of parameters required for convolution calculation by splitting the correlation between spatial dimensions and channel (depth) dimensions, and has been proven in some studies to improve the efficiency of convolution kernel parameters. This feature makes deep separable convolution have potential application value in the field of multimedia watermark embedding and extraction.
[0005] In the research of digital watermark technology, how to embed and extract watermarks effectively is a key issue. Traditional watermark embedding methods are often based on simple algorithms, such as least significant bit replacement, spread spectrum algorithm, etc. Although these methods are simple and easy to implement, their robustness and concealment are often difficult to guarantee when facing complex digital media and powerful attack methods.
[0006] Therefore, a multimedia watermark embedding and extraction method based on deep separable convolution is urgently needed to solve the above problems. Summary of the invention
[0007] The purpose of the present invention is to provide a multimedia watermark embedding and extraction method of depth-separable convolution: to solve the technical problems of low robustness and low concealment in the multimedia watermark embedding and extraction process in the existing solutions.
[0008] The purpose of the present invention can be achieved through the following technical solutions:
[0009] A multimedia watermark embedding and extraction method based on deep separable convolution, the method comprising:
[0010] Acquire watermark information, wherein the watermark information includes a gray image;
[0011] Preprocess the watermark information: divide the gray image into several non-overlapping blocks , for each block Take integer Transform and quantize to obtain a two-dimensional coefficient matrix , converted into a one-dimensional coefficient sequence by zigzag scanning ,based on The one-dimensional coefficient sequence obtains the recovery code;
[0012] The pre-processed watermark information is input into the deep separable convolutional model to complete the multimedia audio watermark embedding;
[0013] Based on the watermark embedding effect, an evaluation is performed to obtain a watermark embedding evaluation effect index, and based on the watermark embedding evaluation effect index, it is determined whether the watermark embedding meets the watermark embedding requirement;
[0014] For multimedia audio that meets the watermark embedding requirements, watermark extraction is completed based on the recovery code.
[0015] Furthermore, for each block Take integer Transform and quantize to obtain a two-dimensional coefficient matrix The process includes:
[0016] Take integer Transform and quantize to obtain a two-dimensional coefficient matrix :
[0017]
[0018] in, Represents an integer Transformation and quantization processing.
[0019] Further, it is converted into a one-dimensional coefficient sequence by zigzag scanning. The process includes:
[0020] Zigzag scanning from a two-dimensional coefficient matrix Starting from the upper left corner, scan along a Z or zigzag path. At the beginning of the scan, move diagonally from the upper left corner to the lower right corner, then from the lower right corner to the upper right corner. After reaching the top boundary, move along another diagonal to the lower left corner, then from the lower left corner to the upper left corner, and so on. In each step, if there are coefficients to be filled, fill the current coefficient into the corresponding position; if there are no more coefficients, fill zeros in the current position and convert it into a one-dimensional coefficient sequence through zigzag scanning. ,in, Represents the position of the coefficients after zigzag scanning.
[0021] Further, based on The process of recovering the one-dimensional coefficient sequence to obtain the recovery code includes the following steps:
[0022] right The one-dimensional coefficient sequence is transformed to obtain the recovery code :
[0023]
[0024] in, and is a positive integer, , Respectively represent Maximum and minimum values of the one-dimensional coefficient sequence.
[0025] Furthermore, the preprocessed watermark information is input into the deep separable convolutional model to complete the multimedia audio watermark embedding, which includes the following processes:
[0026] Redirect the carrier image from Color space conversion Color space, where the carrier image is a multimedia audio image used to carry watermark information. The conversion formula is as follows:
[0027]
[0028] in, Represents the red light intensity value, Indicates the green light intensity value, Indicates the blue light intensity value, represents the brightness component, , represents the chrominance component;
[0029] Calculate the chromaticity saliency of any point in the carrier image , the calculation formula is as follows:
[0030]
[0031] in, , They are , The normalized mapping of the components, represents the significance parameter, which is taken as 0.25;
[0032] based on The operator calculates the edge saliency of the carrier image :
[0033]
[0034] in, Indicates the canny operator on the pixel The result of the calculation is, represents the threshold value, which is 2;
[0035] Based on chromatic saliency and marginal saliency Get the pixel value of the carrier image , the pixel Used to enhance the robustness of the carrier image:
[0036]
[0037] in, is the number of pixels, Pixel The original pixel value of
[0038] The watermark information to be embedded is recorded as a vector ,in, Indicates the number of dimensions and obtains the weights of the convolutional layer in the depthwise separable convolutional model ,make , and Represent the size of the convolution kernel of any convolution layer, the number of channels of the feature map of the previous layer, and the number of convolution kernels of this layer respectively. The weight of the convolution layer as a carrier can be expressed as ;
[0039] Will In the matrix Upward projection:
[0040] ;in, Represents the sequence number of any convolutional layer; represents a projection expression, represents the i-th convolutional layer;
[0041] In order to make the arrangement order of all the projected values modulated consistent with the watermark vector to be embedded The cross entropy function is used to define the watermark embedding regularization term. :
[0042]
[0043] in, is the modulation function;
[0044] The regular term Add to the loss function to get the loss function for training :
[0045] ;
[0046] in, Represents the original loss function of the model without watermark embedding, It is a balance parameter used to find the balance between completing the original task goal and achieving watermark embedding;
[0047] Based on the loss function The enhanced carrier image is trained with a depthwise separable convolutional model to complete watermark embedding.
[0048] Further, based on the watermark embedding effect, an evaluation is performed to obtain a watermark embedding evaluation effect index, and based on the watermark embedding evaluation effect index, it is determined whether the watermark embedding meets the watermark embedding requirement, which specifically includes the following process:
[0049] Calculate the peak signal-to-noise ratio of the watermark and structural similarity :
[0050] ;
[0051] ;
[0052] in, Representative Points The pixel values in the original video frame, Representative Points The pixel value corresponding to the video frame after the watermark is embedded, and Indicates the size of the video frame;
[0053] Calculate the bit error rate of the watermark :
[0054] ;
[0055] in, Indicates the number of bits that were incorrectly extracted. Indicates the total number of watermark embedded bits;
[0056] Peak signal-to-noise ratio , structural similarity and bit error rate Substitute the watermark embedding evaluation effect index into the correlation formula to obtain the watermark embedding evaluation effect index , the association formula is as follows:
[0057] ;
[0058] Among them, the weight factor , , is a constant coefficient, which is set by the user;
[0059] Determine the watermark embedding evaluation effect index Whether it exceeds the preset threshold, if so, it meets the watermark embedding requirements, if not, it does not meet the watermark embedding requirements, where the watermark embedding evaluation effect index The lower it is, the better the corresponding embedding effect is.
[0060] Furthermore, the watermark extraction based on the recovery code for multimedia audio that meets the watermark embedding requirements includes:
[0061] Extracting watermark sequence based on recovery code :
[0062] when and , is 1; otherwise, is 0.
[0063] Compared with the existing solutions, the present invention achieves the following beneficial effects:
[0064] The present invention obtains watermark information, wherein the watermark information includes a gray image; pre-processes the watermark information: divides the gray image into a plurality of non-overlapping blocks , for each block Take integer Transform and quantize to obtain a two-dimensional coefficient matrix , converted into a one-dimensional coefficient sequence by zigzag scanning ,based on The one-dimensional coefficient sequence is used to obtain the restoration code; the preprocessed watermark information is input into the deep separable convolution model to complete the multimedia audio watermark embedding; the watermark embedding effect is evaluated to obtain the watermark embedding evaluation effect index, and the watermark embedding evaluation effect index is used to determine whether the watermark embedding meets the watermark embedding requirements; the multimedia audio that meets the watermark embedding requirements is extracted based on the restoration code. This can improve the robustness and concealment of the multimedia watermark embedding extraction process, thereby improving the efficiency of watermark embedding extraction. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0066] Figure 1 It is a flow chart of a multimedia watermark embedding and extraction method of a depth-separable convolution according to an embodiment of the present invention;
[0067] Figure 2 It is a flowchart of another multimedia watermark embedding and extraction method of depth-separable convolution according to an embodiment of the present invention. DETAILED DESCRIPTION
[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0069] In addition, the described features, structures or characteristics may be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. may be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.
[0070] This embodiment provides a multimedia watermark embedding and extraction method based on deep separable convolution. Figure 1 is a flowchart of a multimedia watermark embedding and extraction method of a deep separable convolution according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:
[0071] Step S101: obtaining watermark information, wherein the watermark information includes a gray image;
[0072] Step S102: pre-process the watermark information: divide the gray image into several non-overlapping blocks , for each block Integer DCT transform and quantization are used to obtain a two-dimensional coefficient matrix , converted into a one-dimensional coefficient sequence by zigzag scanning ,based on The one-dimensional coefficient sequence obtains the recovery code;
[0073] Step S103: input the pre-processed watermark information into the depth-separable convolutional model to complete multimedia audio watermark embedding;
[0074] Step S104: Evaluate the watermark embedding effect to obtain a watermark embedding evaluation effect index, and determine whether the watermark embedding meets the watermark embedding requirement based on the watermark embedding evaluation effect index;
[0075] Step S105: extracting the watermark based on the recovery code for the multimedia audio that meets the watermark embedding requirements.
[0076] In summary, the present invention obtains watermark information, wherein the watermark information includes a gray image; pre-processes the watermark information: divides the gray image into a plurality of non-overlapping blocks , for each block Take integer Transform and quantize to obtain a two-dimensional coefficient matrix , converted into a one-dimensional coefficient sequence by zigzag scanning ,based on The one-dimensional coefficient sequence is used to obtain the restoration code; the preprocessed watermark information is input into the deep separable convolution model to complete the multimedia audio watermark embedding; the watermark embedding effect is evaluated to obtain the watermark embedding evaluation effect index, and the watermark embedding evaluation effect index is used to determine whether the watermark embedding meets the watermark embedding requirements; the multimedia audio that meets the watermark embedding requirements is extracted based on the restoration code. This can improve the robustness and concealment of the multimedia watermark embedding extraction process, thereby improving the efficiency of watermark embedding extraction.
[0077] In some embodiments, for each block Take integer Transform and quantize to obtain a two-dimensional coefficient matrix The process includes:
[0078] Take integer Transform and quantize to obtain a two-dimensional coefficient matrix :
[0079]
[0080] in, Represents an integer Transformation and quantization processing.
[0081] It is worth noting that integer ( , Discrete Cosine Transform (DCT) is a widely used technique in signal processing and image processing, especially in the field of image and audio compression. Transform and Embedding hidden information (such as text, images, audio, etc.) in the coefficients can achieve data hiding in images. This method has good robustness, and the embedded hidden information has little impact on the quality of the original image. The transformation is A special form of the transform that uses integer arithmetic to avoid the complexity and precision issues of floating-point arithmetic. In video coding standards such as Transformation (such as ).
[0082] In some embodiments, the conversion to a one-dimensional coefficient sequence is performed by zigzag scanning. The process includes:
[0083] Zigzag scanning from a two-dimensional coefficient matrix Starting from the upper left corner, scan along a Z or zigzag path. At the beginning of the scan, move diagonally from the upper left corner to the lower right corner, then from the lower right corner to the upper right corner. After reaching the top boundary, move along another diagonal to the lower left corner, then from the lower left corner to the upper left corner, and so on. In each step, if there are coefficients to be filled, fill the current coefficient into the corresponding position; if there are no more coefficients, fill zeros in the current position and convert it into a one-dimensional coefficient sequence through zigzag scanning. ,in, Represents the position of the coefficients after zigzag scanning.
[0084] It is worth mentioning that the key to zigzag scanning is to manage the direction of traversal and handle boundary conditions, such as when the scan reaches the upper or right boundary of the matrix, the direction needs to be changed correctly. In image compression, zigzag scanning can aggregate continuous non-zero data, thereby improving compression efficiency.
[0085] In some embodiments, inputting the pre-processed watermark information into the depth-separable convolutional model to complete multimedia audio watermark embedding includes the following process:
[0086] Redirect the carrier image from Color space conversion Color space, where the carrier image is a multimedia audio image used to carry watermark information. The conversion formula is as follows:
[0087]
[0088] in, Represents the red light intensity value, Indicates the green light intensity value, Indicates the blue light intensity value, represents the brightness component, , represents the chrominance component;
[0089] Calculate the chromaticity saliency of any point in the carrier image , the calculation formula is as follows:
[0090] ;
[0091] in, , They are , The normalized mapping of the components, represents the significance parameter, which is taken as 0.25;
[0092] based on The operator calculates the edge saliency of the carrier image :
[0093] ;
[0094] in, Indicates the canny operator on the pixel The result of the calculation is, represents the threshold value, which is 2;
[0095] Based on chromatic saliency and marginal saliency Get the pixel value of the carrier image , the pixel Used to enhance the robustness of the carrier image:
[0096]
[0097] in, is the number of pixels, Pixel The original pixel value of
[0098] The watermark information to be embedded is recorded as a vector ,in, Indicates the number of dimensions and obtains the weights of the convolutional layer in the depthwise separable convolutional model ,make , and Represent the size of the convolution kernel of any convolution layer, the number of channels of the feature map of the previous layer, and the number of convolution kernels of this layer respectively. The weight of the convolution layer as a carrier can be expressed as ;
[0099] Will In the matrix Upward projection:
[0100] ;in, Represents the sequence number of any convolutional layer;
[0101] In order to make the arrangement order of all the projected values modulated consistent with the watermark vector to be embedded The cross entropy function is used to define the watermark embedding regularization term. :
[0102] ;
[0103] in, is the modulation function;
[0104] The regular term Add to the loss function to get the loss function for training :
[0105] ;
[0106] in, Represents the original loss function of the model without watermark embedding, It is a balance parameter used to find the balance between completing the original task goal and achieving watermark embedding;
[0107] Based on the loss function The enhanced carrier image is trained with a depthwise separable convolutional model to complete watermark embedding.
[0108] In some embodiments, Figure 2 is a flowchart of another multimedia watermark embedding and extraction method of depth-separable convolution according to an embodiment of the present invention. Figure 2 As shown, based on the watermark embedding effect, an evaluation is performed to obtain a watermark embedding evaluation effect index, and based on the watermark embedding evaluation effect index, it is determined whether the watermark embedding meets the watermark embedding requirements, specifically including the following steps:
[0109] Step S201: Calculate the peak signal-to-noise ratio of the watermark and structural similarity :
[0110] ;
[0111] ;
[0112] in, Representative Points The pixel values in the original video frame, Representative Points The pixel value corresponding to the video frame after the watermark is embedded, and Indicates the size of the video frame;
[0113] Step S202: Calculate the bit error rate of the watermark :
[0114] ;
[0115] in, Indicates the number of bits that were incorrectly extracted. Indicates the total number of watermark embedded bits;
[0116] Step S203: Peak signal-to-noise ratio , structural similarity and bit error rate Substitute the watermark embedding evaluation effect index into the correlation formula to obtain the watermark embedding evaluation effect index , the association formula is as follows:
[0117] ;
[0118] Among them, the weight factor , , is a constant coefficient, which is set by the user;
[0119] Step S204: Determine the watermark embedding evaluation effect index Whether it exceeds a preset threshold, which is set by the user and stored in the watermark embedding evaluation system. If so, proceed to step S205; if not, proceed to step S206;
[0120] Step S205: if the watermark embedding requirements are not met, the watermark embedding operation is performed again;
[0121] Step S206: If the watermark embedding requirements are met, the watermark extraction operation can be started.
[0122] In some embodiments, extracting a watermark based on a recovery code from multimedia audio that meets the watermark embedding requirements includes:
[0123] Extracting watermark sequence based on recovery code :
[0124] when and , is 1; otherwise, is 0.
[0125] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.
[0126] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0127] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0128] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0130] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A multimedia watermark embedding and extraction method based on deep separable convolution, characterized in that: Methods include: Acquire watermark information, wherein the watermark information includes a gray image; Preprocess the watermark information: divide the gray image into several non-overlapping blocks p(α, β), take integer DCT transform and quantization processing for each block p(α, β) to obtain a two-dimensional coefficient matrix p(u, v), convert it into a one-dimensional coefficient sequence p(k) through zigzag scanning, and obtain the restoration code based on the one-dimensional coefficient sequence p(k); The preprocessed watermark information is input into the deep separable convolutional model to complete the multimedia audio watermark embedding: The carrier image is converted from the RGB color space to the YCbCr color space, where the carrier image is a multimedia audio image used to carry watermark information. The conversion formula is as follows: Y = 0.29R + 0.587G + 0.114B; Cb = 0.564 (BY); Cr=0.713(RY); Among them, R represents the red light intensity value, G represents the green light intensity value, B represents the blue light intensity value, Y represents the brightness component, and Cb and Cr represent the chromaticity components; Calculate the chromaticity saliency S of any point in the carrier image c (x), the calculation formula is as follows: Among them, f b (x), f c (x) are the normalized maps of Cb and Cr components, respectively, δ represents the significance parameter, which is set to 0.25; Calculate the edge saliency S of the carrier image based on the canny operator E (x): Among them, canny(x) represents the result of the canny operator calculating the pixel x. represents the threshold value, which is 2; Based on chromaticity saliency S c (x) and edge saliency S E (x) Get the pixel value I of the carrier image in , the pixel I in Used to enhance the robustness of the carrier image: Among them, q is the number of pixels, I co is the original pixel value of pixel x; The watermark information to be embedded is recorded as vector b∈{0,1} l , where l represents the number of dimensions. We obtain the weight w of the convolutional layer in the depthwise separable convolutional model. Let (s, s), d, and H represent the size of the convolution kernel of any convolutional layer, the number of channels of the feature map of the previous layer, and the number of convolution kernels of this layer, respectively. The weight of the convolutional layer as a carrier can be expressed as w∈T s×s×d ; Let w∈T s×s×d Projection on matrix X: ρ(x)=∑ i XW i ; where i represents the number of any convolutional layer; ρ(x) represents the projection expression, w i represents the i-th convolutional layer; In order to make the arrangement order of all the projected values modulated the same as the sequence of the watermark vector b to be embedded, the cross entropy function is used to define the watermark embedding regularization term E R : Among them, z j is the modulation function; The regularization term E R Add to the loss function to get the loss function E used for training ss (w): E ss (in)=E0(in)+γE R( In); Among them, E0(w) represents the original loss function of the model without watermark embedding, and γ is a balance parameter used to find the balance point between completing the original task goal and achieving watermark embedding; Based on the loss function E ss (w) and the depth-wise separable convolutional model are used to perform watermark embedding training on the enhanced carrier image to complete watermark embedding; Based on the watermark embedding effect, an evaluation is performed to obtain a watermark embedding evaluation effect index, and based on the watermark embedding evaluation effect index, it is determined whether the watermark embedding meets the watermark embedding requirement; For multimedia audio that meets the watermark embedding requirements, watermark extraction is completed based on the recovery code.
2. According to claim 1, a multimedia watermark embedding and extraction method of depth-separable convolution is characterized in that: Taking integer DCT transform and quantization processing for each block p(α, β) to obtain a two-dimensional coefficient matrix p(u, v) includes the following process: Integer DCT transform and quantization are used to obtain a two-dimensional coefficient matrix p(u, v): p(u,v)=IntQDCT{p(α,β)}; Here, IntQDCT{·} represents integer DCT transform and quantization processing.
3. According to claim 1, a multimedia watermark embedding and extraction method of depth-separable convolution is characterized in that: The conversion to a one-dimensional coefficient sequence p(k) by zigzag scanning includes the following process: The zigzag scan starts from the upper left corner of the two-dimensional coefficient matrix p(u, v) and scans along a zigzag path. At the beginning of the scan, it moves diagonally from the upper left corner to the lower right corner, and then from the lower right corner to the upper right corner. After reaching the top boundary, it moves along another diagonal to the lower left corner, and then from the lower left corner to the upper left corner, and so on. In each step, if there are coefficients to be filled, the current coefficient is filled in the corresponding position; if there are no more coefficients, zero is filled in the current position and converted into a one-dimensional coefficient sequence p(k) through zigzag scanning, where k∈[0,15] represents the position of the coefficient after the zigzag scan.
4. According to claim 1, a multimedia watermark embedding and extraction method of depth-separable convolution is characterized in that: The recovery code based on the one-dimensional coefficient sequence of p(k) includes the following process: Transform the one-dimensional coefficient sequence of p(k) to obtain the recovery code m(k): Wherein, k∈[0, 15] is a positive integer, and max and min represent the maximum and minimum values of the one-dimensional coefficient sequence of p(k), respectively.
5. According to claim 1, a multimedia watermark embedding and extraction method of depth-separable convolution is characterized in that: Based on the watermark embedding effect, an evaluation is performed to obtain a watermark embedding evaluation effect index, and based on the watermark embedding evaluation effect index, it is determined whether the watermark embedding meets the watermark embedding requirements. Specifically, the process includes the following steps: Calculate the peak signal-to-noise ratio PSNR and structural similarity MSE of the watermark: Among them, U m,n Represents the pixel value of point (m, n) in the original video frame, U'' m,n The pixel value corresponding to the point (m, n) in the video frame after the watermark is embedded, and M and N represent the size of the video frame; Calculate the bit error rate BER of the watermark: Among them, WBT represents the number of bits extracted incorrectly, and ZBT represents the total number of watermark embedded bits; Substitute the peak signal-to-noise ratio PSNR, structural similarity MSE and bit error rate BER into the watermark embedding evaluation effect index association formula to obtain the watermark embedding evaluation effect index LMS. The association formula is as follows: Among them, the weight factors θ, π, and C1 are constant coefficients, which are set by the user; Determine whether the watermark embedding evaluation effect index LMS exceeds the preset threshold. If not, the watermark embedding requirements are met. If so, the watermark embedding requirements are not met. The lower the watermark embedding evaluation effect index LMS, the better the corresponding embedding effect.
6. The multimedia watermark embedding and extraction method of depth-separable convolution according to claim 1 is characterized in that: The watermark extraction based on the recovery code for multimedia audio that meets the watermark embedding requirements includes: Extracting watermark sequence b based on recovery code j : When m(k)>5 and ∑ i XW i >0.5, b j is 1; otherwise, b j is 0.
