CN114529441A - Image frequency domain digital watermarking method, system, device and medium - Google Patents

Abstract

The invention discloses a method, a system, a device and a medium for image frequency domain digital watermarking, wherein the method comprises the following steps: designing an embedding and extracting model of the image frequency domain digital watermark; inputting a carrier picture and a watermark into a watermark embedding and extracting self-encoder to obtain a first picture with the watermark, randomly adding the first picture into noise attack, and inputting the first picture into a decoder of the watermark embedding and extracting self-encoder to obtain a rough extracted watermark; randomly selecting the first picture to cut or not attack to obtain a second picture and a standard confidence map, inputting the second picture into an attention scorer, and outputting an attention confidence map; and (4) multiplying the roughly extracted watermark and the attention confidence map pixel by pixel on a two-dimensional plane for modification to obtain a final watermark. The invention analyzes the picture of the watermark to be extracted through the attention scoring device, reduces the influence of the watermark-free part on the real part with the watermark, obtains better extraction accuracy rate, and can be widely applied to the technical field of information security.

Description

A kind of image frequency domain digital watermarking method, system, device and medium

technical field

The present invention relates to the technical field of information security, and in particular, to a method, system, device and medium for digital watermarking in the frequency domain of images.

Background technique

In the Internet era, digital watermarking technology is widely used in daily life and commercial activities. It has functions such as protecting the copyright of digital media and preventing tampering. It is an important part of the field of information security. Among the many possible carriers for embedding additional information, digital images are the most common. The widespread dissemination of digital images in a variety of fields, from everyday communications on social networks to medicine, the military and space, has facilitated this process.

Digital watermarking technology needs to be divided into two main processes of embedding and extraction. Embedding refers to embedding the watermark into digital media without significantly changing the original perceptual effect of its content, while the extraction process is to extract the original embedded content from the embedded digital media. One of the key technical difficulties of digital watermarking technology is how to extract the original watermark as much as possible after the digital media embedded with the watermark has undergone various digital signal processing processes, that is, the robustness of the digital watermark.

Usually, according to the position of the watermark embedding, it can be roughly divided into frequency domain embedding and spatial domain embedding. Specifically, the object modified by the spatial domain watermarking algorithm is the pixel value of the point on the image, that is, the watermark is placed on the premise of imperceptibility. Embedding into the image, typically based on the least significant bit algorithm. One of the advantages of spatial domain watermarking algorithm is that it is easy to implement, but it has poor robustness. The specific method of frequency domain digital watermarking technology is to first perform mathematical transformation on the image, and then embed the watermark in the transformation domain. Common mathematical transformations include discrete Fourier transform, discrete wavelet transform, and discrete cosine transform, all of which are in the field of digital watermarking. Gradually it has been widely used.

With the development of computer vision, neural networks shine in many fields such as object detection and face recognition. Considering that the digital watermarking technology of images is also a special branch of digital image processing, neural network can also be introduced into digital watermarking. At present, on the basis of traditional spatial domain algorithm and transform domain algorithm, researchers gradually combine digital watermarking technology with neural network and other technologies, which greatly improves the robustness of the algorithm and has its unique advantages in some aspects. , which greatly promotes the development of digital watermarking technology. Ahmadi et al. of Isfahan University of Science and Technology proposed a method for watermark embedding in the frequency domain using a convolutional autoencoder, which has attracted the attention of many scholars.

For image digital watermarking, among various digital signal processing methods, cropping is an operation that seriously destroys watermarking. The cropping of the image means the loss of information. The more cropping, the more information is lost. How to extract the complete watermark from the remaining small part of the information becomes a difficult problem. If the watermark can still be accurately extracted from the image after cropping, an exquisite algorithm needs to be designed to embed the watermark into each local unit of the image as much as possible, and extract the watermark well.

SUMMARY OF THE INVENTION

In order to solve one of the technical problems existing in the prior art at least to a certain extent, the purpose of the present invention is to provide an image frequency domain digital watermarking method, system, device and medium.

The technical scheme adopted in the present invention is:

An image frequency domain digital watermarking method, comprising the following steps:

Design the embedding and extraction model of image frequency domain digital watermark; the embedding and extraction model includes the watermark embedding extraction from the encoder and the attention scorer;

After the carrier picture and the watermark are input into the watermark embedded and extracted from the encoder, the first picture with the watermark is obtained, and the first picture is randomly added to the noise attack, and then input into the decoder of the watermark embedded and extracted from the encoder, and the rough picture is obtained. extract watermark;

The first picture is randomly selected to be cropped to attack or not to attack, to obtain a second picture and a standard confidence level map, input the second picture to an attention scorer, and output an attention confidence level map;

The coarse extraction watermark obtained by the self-encoder extracted by the watermark embedding and the attention confidence map obtained by the attention scorer are multiplied pixel by pixel on the two-dimensional plane for correction to obtain the final watermark.

Further, the method for digital watermarking in the frequency domain of an image further comprises the step of training the watermark embedding and extracting from the encoder:

In order to minimize the error between the carrier picture and the watermarked picture, as well as the error between the input watermark and the extracted watermark, the training watermark embedding is extracted from the encoder to extract the watermark embedding;

Described a kind of image frequency domain digital watermarking method also comprises the step of training described attention scorer:

Taking minimizing the error between the output of the attention scorer and the standard confidence map as the training direction, the training attention scorer calculates the confidence that the input image has a watermark at each position.

Further, the watermark embedding is extracted from the encoder and the attention scorer is composed of a convolutional neural network;

Before inputting the picture into the watermark embedding and extracting from the encoder and the attention scorer, it includes the steps of preprocessing the picture:

If the image is a three-channel color image, first transform the color image into the YCbCr space, keep the CbCr channel unchanged, and only take the Y channel as a grayscale image for embedding;

If the image is a single-channel grayscale image, there is no need to perform color space transformation; the grayscale image is divided into non-overlapping 8×8 image blocks, and after each image block is individually discrete cosine transform, all image blocks are reshaped together is a tensor whose side length is 1/8 of the side length of the carrier image, and the number of channels is 64, and each channel represents a coefficient of discrete cosine transform;

In addition, the watermark input watermark embedding is extracted from the encoder, and the following preprocessing is performed on the watermark:

Arrange the watermarks into tensors with equal side lengths and a total of 4 channels, keep the number of channels unchanged, and repeat on the two-dimensional plane until the side length is the same as the side length of the image reconstruction tensor.

Further, in the watermark embedding extraction from the encoder, after the preprocessed image tensor and the watermark tensor are spliced, they are input to the six-layer convolution kernel with a size of 1×1, a stride of 1×1, and no padding. The neural network formed by the convolution layer and the nonlinear activation layer in series is used for encoding, and the output of the encoder is inverse discrete cosine transform, and the obtained residual image is added to the carrier image to obtain the final watermarked gray image;

If the image before preprocessing is a three-channel color image, use the grayscale image output by the encoder as the Y channel, and convert it back to the RGB color space with the CbCr channel retained during preprocessing; then select the preset digital image processing method to add Noise attack, and then after four layers of convolution kernel size are 1 × 1, step size is 1 × 1, the unfilled convolution layer and the nonlinear activation layer are connected in series to form a neural network to decode the watermark tensor. , the repeated information bits are averaged to obtain the rough extraction watermark, in order to minimize the error between the rough extraction watermark and the input watermark and the error between the watermarked picture output by the encoder and the input picture as the training direction, the training watermark embedding is extracted from the encoder Embedding and extracting the watermark.

Further, in the attention scorer, the cropping attack is to randomly generate a rectangle smaller than the image size, the image content inside the rectangle remains unchanged, and the part outside the rectangle can be modified to any other value;

The corresponding confidence map has a value of 0 at the clipping position and 1 at the reserved position. If no attack is performed, the entire confidence map is 1; then the kernel size of the confidence map is 8 × 8, Two-dimensional mean pooling with a step size of 8 × 8 to obtain a standard confidence map;

The processed image is subjected to discrete cosine transform and then input to a four-layer convolution kernel with a size of 3 × 3, a stride of 1 × 1, and a convolutional neural network formed in series with a unit zero-padding convolutional layer and a non-linear activation layer. , get the attention confidence map, in the direction of minimizing the error between the attention confidence map and the standard confidence map, training the attention scorer can learn the confidence of the input image with watermark at each position.

Further, in the training of the watermark embedding extracted from the encoder, the error between the input watermark and the extracted watermark adopts the mean square error, and the error between the pictures also adopts the mean square error, and the loss function used for training is the picture error and sum of watermark errors;

In the training of the attention scorer, the error between the standard confidence map and the attention confidence map adopts the mean square error, and the mean square error is used as the loss function of training.

Further, the rough extraction watermark obtained from the encoder extracted by the watermark embedding and the attention confidence map obtained by the attention scorer are multiplied pixel by pixel on the two-dimensional plane for correction to obtain the final watermark, including :

Roughly extract the watermark tensor, the value range of all values is (0, 1), and linearly transform the value range of the watermark tensor to (-1, 1);

The attention confidence map and the watermark tensor are multiplied element by element on the two-dimensional plane to obtain the final watermark tensor;

According to the final watermark tensor, the repeated watermark bits are averaged and rounded to obtain the final watermark.

Another technical scheme adopted by the present invention is:

An image frequency domain digital watermarking system, comprising:

The model design module is used to design the embedding and extraction model of the image frequency domain digital watermark; wherein, the embedding and extraction model includes the watermark embedding and extraction from the encoder and the attention scorer;

The watermark extraction module is used to extract the carrier picture and the watermark input watermark from the encoder to obtain the first picture with the watermark, add the first picture randomly to the noise attack, and then input it into the watermark embedded and extracted from the encoder. In the decoder, the rough extraction watermark is obtained;

Confidence acquisition module, used to randomly select the first picture to crop or not to attack, obtain the second picture and the standard confidence map, input the second picture to the attention scorer, and output the attention confidence picture;

The watermark correction module is used to multiply the coarse extraction watermark obtained by the watermark embedding from the encoder and the attention confidence map obtained by the attention scorer on a two-dimensional plane by pixel-by-pixel multiplication to obtain the final watermark.

Another technical scheme adopted by the present invention is:

An image frequency domain digital watermarking device, comprising:

at least one processor;

at least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

Another technical scheme adopted by the present invention is:

A computer-readable storage medium in which a processor-executable program is stored, the processor-executable program, when executed by the processor, is used to perform the method as described above.

The beneficial effects of the present invention are: the present invention analyzes the pictures to be watermarked through the attention scorer, gives a lower score to the tampered part, reduces the final majority voting or takes the mean value, these non-watermarked parts have no effect on the real watermark. The influence of the watermark part can get better extraction accuracy.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following descriptions are given to the accompanying drawings of the embodiments of the present invention or the related technical solutions in the prior art. It should be understood that the drawings in the following introduction are only In order to facilitate and clearly express some embodiments of the technical solutions of the present invention, for those skilled in the art, other drawings can also be obtained from these drawings without creative work.

1 is a structural diagram of an encoder whose watermark embedding is extracted from an encoder in an embodiment of the present invention;

2 is a structural diagram of a decoder whose watermark embedding is extracted from an encoder in an embodiment of the present invention;

3 is a structural diagram of an attention scorer in an embodiment of the present invention;

Fig. 4 is the flow chart of the preprocessing of the picture in the embodiment of the present invention;

5 is a schematic flowchart of preprocessing of watermark information in an embodiment of the present invention;

6 is a flow chart of an image frequency domain digital watermarking method based on an attention mechanism in an embodiment of the present invention;

FIG. 7 is a flowchart of attention scorer training in an embodiment of the present invention.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary, only used to explain the present invention, and should not be construed as a limitation of the present invention. The numbers of the steps in the following embodiments are only set for the convenience of description, and the sequence between the steps is not limited in any way, and the execution sequence of each step in the embodiments can be adapted according to the understanding of those skilled in the art Sexual adjustment.

In the description of the present invention, it should be understood that the azimuth description, such as the azimuth or position relationship indicated by up, down, front, rear, left, right, etc., is based on the azimuth or position relationship shown in the drawings, only In order to facilitate the description of the present invention and simplify the description, it is not indicated or implied that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be construed as limiting the present invention.

In the description of the present invention, the meaning of several is one or more, the meaning of multiple is two or more, greater than, less than, exceeding, etc. are understood as not including this number, above, below, within, etc. are understood as including this number. If it is described that the first and the second are only for the purpose of distinguishing technical features, it cannot be understood as indicating or implying relative importance, or indicating the number of the indicated technical features or the order of the indicated technical features. relation.

In the description of the present invention, unless otherwise clearly defined, words such as setting, installation, connection should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.

As shown in FIG. 6 , this embodiment provides an image frequency domain digital watermarking method based on an attention mechanism, including the following steps:

S1. The design of the self-encoder model and attention scorer for watermark embedding and extraction. The specific network structure is set as follows: As shown in Figure 1, the self-encoder model of this embodiment includes an encoder and a decoder, both of which are convolutional neural networks. The network, in which the encoder has a total of six layers, as shown in Figure 1, the first five layers are 64 64 × 1 × 1, stride 1 × 1, unfilled convolution kernel and ELU activation function are formed in series, The last layer is a 64×1×1, stride 1×1, unfilled convolution kernel and ELU activation function in series; the decoder has a total of four layers, as shown in Figure 2, the first three layers are 64 convolution kernels of 64×1×1, stride 1×1, and no padding are formed in series with the ELU activation function, and the last layer is 4 64×1×1, stride 1×1, no padding. The convolution kernel and the sigmoid activation function are formed in series.

As shown in Figure 3, the attention scorer has a total of four layers. The first three layers are 64 convolution kernels of 64 × 3 × 3, stride of 1 × 1, 1 unit zero-padding and ELU activation function in series. , the last layer is a 64×3×3 convolution kernel with a stride of 1×1, a unit zero-padding and a sigmoid activation function in series.

The ELU activation function is shown in the following formula:

Wherein x is the input value of the activation function ELU, ELU(x) is the output value of the activation function, and α is a constant, which is 1 in this embodiment;

The sigmoid activation function is shown in the following formula:

where x is the input value of the activation function sigmoid, and sigmoid(x) is the output value of the activation function.

S2. The preprocessing and postprocessing of pictures and watermark information before and after entering each network. The specific operations are as follows:

As shown in Figure 4, the preprocessing of the image can be divided into three steps: color space conversion, slicing, space and frequency domain conversion. If the input image is a black and white single-channel image, the color space conversion is skipped. In this embodiment, a color picture of 3×128×128 is used as the input picture. For color pictures, first change it from RGB color space to YCbCr color space. Since the change of CbCr channel is easy to cause the color of the picture to change, the CbCr channel is kept unchanged, and only the information of the Y channel is embedded. Then, the Y channel is divided into non-overlapping image blocks of size 8×8, and each image block is transformed from the spatial domain to the frequency domain. In this embodiment, 8×8 DCT (discrete cosine transform) is used to obtain the size is a 64×16×16 image tensor, where 64 is the number of channels, and each channel represents a DCT coefficient. For the encoder in the autoencoder, the output of the network also needs to be post-processed. In this embodiment, the encoder outputs a tensor with a size of 64×16×16, first performs IDCT (inverse discrete cosine transform) and then reshapes to obtain a 1×128×128 Y channel residual map, and then combines the residual map with the input The Y channel map at the encoder is added, and then inversely transformed with the CbCr channel back to the RGB color space.

As shown in Figure 5, the preprocessing of watermark information is divided into two steps: reshaping and repetition. In this embodiment, the size of the input watermark information is 64 bits, and each bit is represented by 0 and 1 respectively. It is first reshaped into a 4×4×4 tensor, and then repeated on the two-dimensional plane to the same size as the image tensor. , in this embodiment, the width and height dimensions on the two-dimensional plane are repeated 4 times, and finally a 4×16×16 watermark tensor is obtained.

S3, watermark embedding and training of the extracted autoencoder model, the specific operations are as follows:

As shown in Figure 6, the carrier picture I to be embedded in the watermark and the watermark information W to be embedded are put into the encoder of the self-encoder in step S1 after the preprocessing of step S2 to obtain the watermarked picture I', Calculate the reconstruction error between I and I', and the calculation formula is as follows:

Among them, c, h and w represent the number of image channels, height and width, respectively, and I and I' represent the carrier image and the image after embedding the watermark, respectively.

Then in each iteration process, select a common digital image processing method to add noise to I'. In this embodiment, in Gaussian noise (mean value 0, standard deviation is 0.06), Gaussian blur (radius is 2, standard deviation is 0.1), salt and pepper noise (noise intensity is 0.1), and one of JPEG compression (quality factor is 50) is randomly selected to obtain the image I _n after adding noise.

Then input In to the decoder in the self-encoder, and obtain the roughly extracted watermark tensor W _o with the same size as the watermark tensor in step S2, wherein the value range of each element is ( ₀ , 1), this implementation The tensor size in the example is 4×16×16. After obtaining the rough extracted watermark tensor, repeat step S2, and average the elements representing the repeated watermark bits to obtain a 4×4×4 rough extracted watermark tensor, and then perform the reverse operation of step S2 reshaping The 64-bit rough extraction watermark W' is obtained, and the reconstruction error between W and W' is calculated. The calculation formula is as follows:

Wherein, W and W' are the input watermark and the extracted watermark, respectively, w _i and w' _i are the ith bit of W and W' respectively, and n is the number of bits of the watermark;

Finally, the reconstruction error of the picture and the reconstruction error of the watermark are added to obtain the loss function Loss of the auto-encoder:

Loss=MSE(I,I')+aMSE(W,W')

α is a constant, which is 1 in this embodiment, and is used to balance image quality and robustness.

S4, the training of the attention scorer, the specific steps are as follows: as shown in Figure 7, after using the encoder trained in step S3 to embed the watermark on the picture, in this embodiment, in each iteration, randomly select from three methods Choose one to process the watermarked image to obtain I _c , the three methods are respectively no operation, fill with random color after cropping, and fill the corresponding position pixel of the carrier image after cropping. The cropping operation is to randomly generate a rectangle whose side length is 0.2 to 0.8 times the side length of the original image. In this embodiment, the side length ranges from 24 to 104, and the values of the pixels in the rectangle remain unchanged, and the values outside the rectangle are replaced by other . At the same time, a confidence map of the same size as the picture is generated, the value in the rectangle is 1, and the rest are 0. Without cropping, the confidence map is all 1s.

Then, perform two-dimensional mean pooling with a kernel size of 8×8 and a step size of 8 on the confidence map to obtain a standard confidence map C with the same plane size as the watermark tensor in step S2, which is 16×16 in this embodiment. . Then input I _c into the attention scorer, output the attention confidence map C', and calculate the reconstruction error between C and C' as the loss function. The calculation formula is as follows:

Among them, h and w represent the height and width of the confidence map, respectively.

S5. Use the model trained in step S4 to modify the decoder of step S3, as shown in FIG. 6 . In step S3, the trained decoder outputs a rough extraction watermark W _o , if the watermarked picture input by the decoder is subjected to a cropping attack, then W _o is on a two-dimensional plane, and the decoded information corresponding to the cropped part should be What has no reference value, only the decoding result corresponding to the remaining part of the pixels is correct. To this end, first linearly transform the value range of the W _o element to (-1, 1), and the conversion formula is as follows:

Then use the attention confidence map C' obtained in step S4 and W _o to perform element-by-element multiplication on the two-dimensional plane to obtain the corrected watermark tensor W _c . At this time, the value of the position corresponding to the clipped part on C' is close to 0, so the value of the corresponding position in W _c after the correction is also close to 0, reducing the interference to the subsequent averaging process. Then, similar to step S3, the elements of the repeated watermark bits of W _c are averaged and then rounded up, the watermark bits corresponding to elements greater than 0 are 1, and the watermark bits corresponding to less than 0 are 0.

To sum up, compared with the prior art, the embodiments of the present invention have the following beneficial effects:

(1) In the embodiment of the present invention, by reshaping the watermark information into four planes and then splicing it with the picture tensor, each 8×8 image block of the carrier image has four times the information embedded in it, which is better to use embedded in the frequency domain of the image.

(2) According to the embodiment of the present invention, the watermark information is preprocessed before the watermark self-encoder is input, that is, the watermark information is repeated more times in the plane space, so that the watermarked image can be extracted in a smaller area. Complete watermark information; at the same time, due to more repetitions on the plane, it can have better error correction ability in voting or averaging after rough watermark extraction, and improve the robustness of the watermarking method.

(3) An attention scorer is added in the embodiment of the present invention to analyze the picture to be watermarked, obtain the score of each 8×8 block of the picture, give a lower score to the tampered part, reduce the final majority vote or take In the averaging process, the influence of these non-watermarked parts on the real watermarked parts can get better extraction accuracy.

The present invention also provides an image frequency domain digital watermarking system, comprising:

The model design module is used to design the embedding and extraction model of the image frequency domain digital watermark; wherein, the embedding and extraction model includes the watermark embedding and extraction from the encoder and the attention scorer;

The watermark extraction module is used to extract the carrier picture and the watermark input watermark from the encoder to obtain the first picture with the watermark, add the first picture randomly to the noise attack, and then input it into the watermark embedded and extracted from the encoder. In the decoder, the rough extraction watermark is obtained;

Confidence acquisition module, used to randomly select the first picture to crop or not to attack, obtain the second picture and the standard confidence map, input the second picture to the attention scorer, and output the attention confidence picture;

The watermark correction module is used to multiply the coarse extraction watermark obtained by the watermark embedding from the encoder and the attention confidence map obtained by the attention scorer on a two-dimensional plane by pixel-by-pixel multiplication to obtain the final watermark.

An image frequency domain digital watermarking system in this embodiment can execute an image frequency domain digital watermarking method provided by the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effect.

This embodiment also provides an image frequency domain digital watermarking device, including:

at least one processor;

at least one memory for storing at least one program;

When the at least one program is executed by the at least one processor, the at least one processor implements the method shown in 6 .

An image frequency domain digital watermarking apparatus in this embodiment can execute the image frequency domain digital watermarking method provided by the method embodiment of the present invention, can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effect.

Embodiments of the present application further disclose a computer program product or computer program, where the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the method shown in FIG. 6 .

This embodiment also provides a storage medium, which stores an instruction or program for executing an image frequency domain digital watermarking method provided by the method embodiment of the present invention. When the instruction or program is executed, the method embodiment can be executed. Any combination of implementation steps has corresponding functions and beneficial effects of the method.

In some alternative implementations, the functions/operations noted in the block diagrams may occur out of the order noted in the operational diagrams. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality/operations involved. Furthermore, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of the various operations are altered and in which sub-operations described as part of larger operations are performed independently.

Furthermore, while the invention is described in the context of functional modules, it is to be understood that, unless stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or or software modules, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary to understand the present invention. Rather, given the attributes, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the modules will be within the routine skill of the engineer. Accordingly, those skilled in the art, using ordinary skill, can implement the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the invention, which is to be determined by the appended claims along with their full scope of equivalents.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

The logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, may be embodied in any computer-readable medium, For use with, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions from an instruction execution system, apparatus, or apparatus) or equipment. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with an instruction execution system, apparatus, or apparatus.

More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), Read Only Memory (ROM), Erasable Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Compact Disc Read Only Memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, followed by editing, interpretation, or other suitable medium as necessary process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following techniques known in the art: Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, Programmable Gate Arrays (PGA), Field Programmable Gate Arrays (FPGA), etc.

In the above description of the present specification, reference to the description of the terms "one embodiment/example", "another embodiment/example" or "certain embodiments/examples" etc. means the description in conjunction with the embodiment or example. Particular features, structures, materials, or characteristics are included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned embodiments, and those skilled in the art can also make various equivalent deformations or replacements on the premise of not violating the spirit of the present invention. Equivalent modifications or substitutions are included within the scope defined by the claims of the present application.

Claims (10)

1. A method for digital watermarking in image frequency domain is characterized by comprising the following steps:

designing an embedding and extracting model of the image frequency domain digital watermark; wherein the embedding and extracting model comprises a watermark embedding extracting from an encoder and an attention scorer;

inputting a carrier picture and a watermark into a watermark embedding and extracting self-encoder to obtain a first picture with the watermark, randomly adding the first picture into a noise attack, and then inputting the first picture into a decoder of the watermark embedding and extracting self-encoder to obtain a rough extracted watermark;

randomly selecting the first picture to cut or not to attack to obtain a second picture and a standard confidence map, inputting the second picture into an attention scorer, and outputting an attention confidence map;

and (3) carrying out pixel-by-pixel multiplication on a two-dimensional plane for correcting the coarsely extracted watermark obtained by embedding the watermark into the self-encoder and the attention confidence map obtained by the attention scorer to obtain the final watermark.

2. The image frequency-domain digital watermarking method of claim 1, further comprising the step of training the watermark embedding extraction self-encoder to:

training watermark embedding and extracting from an encoder to perform watermark embedding and extracting in a direction of minimizing an error between a carrier picture and a picture with a watermark and an error between an input watermark and an extracted watermark;

the image frequency domain digital watermarking method further comprises the step of training the attention scorer:

and taking the error between the output of the minimized attention scorer and the standard confidence map as a training direction, and calculating the confidence degree of the input picture with the watermark at each position by the training attention scorer.

3. The image frequency domain digital watermarking method according to claim 1, wherein the watermark embedding extraction encoder and the attention scorer are both composed of a convolutional neural network;

before inputting the picture into said watermark embedding extraction from the encoder and said attention scorer, a step of pre-processing the picture is included:

if the picture is a three-channel color picture, firstly converting the color space of the color picture into a YCbCr space, keeping a CbCr channel unchanged, and only taking a Y channel as a gray scale image for embedding;

if the picture is a single-channel grey-scale image, color space transformation is not needed; dividing the gray-scale image into non-overlapping image blocks with the size of 8 multiplied by 8, after each image block is subjected to discrete cosine transform independently, all the image blocks are reshaped into 1/8 with the side length being the side length of the carrier image, the number of channels is 64, and each channel represents a coefficient of the discrete cosine transform;

in addition, the watermark input watermark is embedded and extracted from the encoder, and the watermark is preprocessed as follows:

arranging the watermarks into tensors with equal side lengths and 4 channels in total, keeping the number of the channels unchanged, and repeating the steps on a two-dimensional plane until the side lengths are the same as the side lengths of the image reconstruction tensors.

4. The image frequency domain digital watermarking method according to claim 3, wherein after the preprocessed image tensor and watermark tensor are spliced in the watermark embedding extraction encoder, the input six layers of convolution kernels are all 1 x 1 in size and all 1 x 1 in step length, a neural network formed by the unfilled convolution layer and the nonlinear active layer in series is encoded, the output of the encoder is subjected to inverse discrete cosine transform, and the obtained residual image is added to the carrier image to obtain a final watermarked gray image;

if the picture before preprocessing is a three-channel color picture, the gray-scale image output by the encoder is used as a Y channel, and the Y channel and the CbCr channel reserved during preprocessing are converted back to the RGB color space again; then selecting a preset digital image processing means to add noise attack, decoding through a neural network formed by connecting four layers of convolution kernels with the size of 1 multiplied by 1 and the step length of 1 multiplied by 1 in a series manner to obtain a roughly extracted watermark tensor, averaging repeated information bits to obtain a roughly extracted watermark, and embedding and extracting the watermark from an encoder by taking the error between the roughly extracted watermark and an input watermark and the error between a watermarked picture output by the encoder and the input picture as a training direction.

5. The method for digital watermarking in image frequency domain according to claim 3, wherein in the attention classifier, the cropping attack is to randomly generate a rectangle smaller than the picture size, the image content in the rectangle is not changed, and the part outside the rectangle can be modified into any other value;

the value of the corresponding confidence map at the cut-out position is 0, the value at the reserved position is 1, and if the attack is not carried out, the whole confidence map is 1; then, performing two-dimensional mean pooling on the confidence coefficient graph with the kernel size of 8 multiplied by 8 and the step length of 8 multiplied by 8 to obtain a standard confidence coefficient graph;

after discrete cosine transform is carried out on the processed picture, the processed picture is input into a convolutional neural network formed by connecting four layers of convolutional layers with convolutional kernel size of 3 multiplied by 3 and step length of 1 multiplied by 1 and unit zero filling and a nonlinear activation layer in series, an attention confidence coefficient graph is obtained, the error of the minimum attention confidence coefficient graph and the standard confidence coefficient graph is taken as the direction, and the attention scorer can be trained to learn the confidence coefficient of the input picture with watermarks at each position.

6. The image frequency domain digital watermarking method according to claim 2, wherein in training the watermark embedding extraction self-encoder, errors of the input watermark and the extracted watermark adopt mean square errors, errors between pictures also adopt mean square errors, and a loss function for training is the sum of the picture errors and the watermark errors;

in training the attention scorer, the error between the standard confidence map and the attention confidence map adopts the mean square error, and the mean square error is used as a loss function of training.

7. The method for digital watermarking the image in the frequency domain according to claim 1, wherein the step of modifying the coarsely extracted watermark obtained by embedding the watermark into the encoder and the attention confidence map obtained by the attention scorer by multiplying the coarsely extracted watermark and the attention confidence map pixel by pixel on a two-dimensional plane to obtain the final watermark comprises:

roughly extracting a watermark tensor, wherein the value range of all values is (0, 1), and linearly converting the value range of the watermark tensor to (-1, 1);

performing element-by-element multiplication on a two-dimensional plane by adopting the attention confidence coefficient diagram and the watermark tensor to obtain a final watermark tensor;

and according to the final watermark tensor, averaging the repeated watermark bits and then rounding to obtain the final watermark.

8. An image frequency domain digital watermarking system, comprising:

the model design module is used for designing an embedding and extracting model of the image frequency domain digital watermark; wherein the embedding and extracting model comprises a watermark embedding extracting from an encoder and an attention scorer;

the watermark extraction module is used for embedding and extracting the carrier picture and the watermark into the encoder to obtain a first picture with the watermark, randomly adding the first picture into noise attack, and then inputting the first picture into a decoder of the watermark embedding and extracting encoder to obtain a rough extracted watermark;

the confidence coefficient acquisition module is used for randomly selecting the first picture to cut or not attack to obtain a second picture and a standard confidence coefficient map, inputting the second picture into the attention scorer and outputting the attention confidence coefficient map;

and the watermark correction module is used for performing pixel-by-pixel multiplication on the roughly extracted watermark obtained by embedding and extracting the watermark into the encoder and the attention confidence coefficient image obtained by the attention scorer for correction to obtain the final watermark.

9. An image frequency domain digital watermarking apparatus, comprising:

at least one processor;

at least one memory for storing at least one program;

when executed by the at least one processor, cause the at least one processor to implement the method of any one of claims 1-7.

10. A computer-readable storage medium, in which a program executable by a processor is stored, wherein the program executable by the processor is adapted to perform the method according to any one of claims 1 to 7 when executed by the processor.

Images