Reversible information hiding method and system based on neural network and global pixel segmentation

Abstract

The invention aims to provide a reversible information hiding method and system based on a neural network and global pixel segmentation. According to the method, a two-layer information embedding framework is adopted; firstly, an original image is divided into a first pixel set and a second pixel set according to a checkerboard pattern, and the first pixel set and the second pixel set are alternately used as carrier pixel sets in two-layer information embedding; in each layer of information embedding operation, a predictor based on a double-branch neural network is adopted, a reference pixel set is used for predicting a carrier pixel set, and error estimation of a corresponding predicted value is given; according to the method, carrier pixel sequences are sequenced according to predicted pixel values, then pixels with prediction errors larger than a preset error threshold value are removed according to estimated prediction error limits, and the selected final carrier pixel sequences are obtained and are divided into sub-pixel sequence sets with the same length for embedding secret information. According to the method, the secret information can be well embedded into the carrier image, information hiding and reversible recovery are achieved, and information transmission safety is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of information security technology and relates to a reversible information hiding method. The present invention uses a digital image as a cover for covert communication, hides secret information in a stego image, and the receiver can extract the secret information from the stego image while losslessly restoring the original carrier image. Background Art

[0002] Reversible information hiding is an important content security technology that can embed secret information into digital multimedia carriers while ensuring that the digital multimedia carriers can be restored without loss. One of the most commonly used multimedia carriers is digital images. The uniqueness of reversible information hiding is that it can protect the embedded data while ensuring the integrity of the original media, which is crucial in scenarios where data authenticity and integrity need to be guaranteed. Reversible information hiding based on digital images has been widely used in image authentication, medical image protection, and military communications.

[0003] With the rapid development of neural networks, how to apply neural networks to reversible information hiding is an important technical research topic. Most of the initial digital image information hiding methods based on neural networks can only ensure that the embedded information can be correctly extracted, but do not have the property of reversibility, that is, the original carrier image cannot be restored losslessly after the information is extracted. Such methods cannot be applied to applications such as image authentication that require the original image to be preserved. An effective technical route to achieve reversible information hiding based on neural networks is to apply neural networks to the reversible information hiding framework based on prediction error modulation. This model mainly consists of two parts: one is a pixel predictor based on a neural network, and the other is an information embedding and extraction method based on prediction error modulation. Neural networks are mainly used to construct pixel predictors, and the accuracy of pixel prediction is improved through the deep learning ability of neural networks, thereby improving the performance of the method. The reversibility of the method is guaranteed by the reversibility of the information embedding method based on prediction error modulation. This reversible information hiding method based on neural network predictors and prediction error modulation uses deep learning to improve the steganographic performance while maintaining the reversible property of the method.

[0004] It can be seen that how to implement a reversible information hiding method based on a neural network predictor and prediction error modulation is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art and to provide a method and system for reversible information hiding based on a dual-branch prediction neural network and global pixel segmentation, so as to realize reversible information hiding based on a neural network.

[0006] The technical solution adopted by the present invention to solve the technical problem comprises the following steps:

[0007] In a first aspect, the present invention provides a reversible information hiding method based on a neural network and global pixel segmentation, comprising:

[0008] S1, dividing the original carrier image I into two types of pixels, forming a first pixel set and a second pixel set respectively; wherein among the eight neighboring pixels of each pixel, the pixels on the four sides are all pixels of different types, and the pixels on the four diagonals are all pixels of the same type; firstly, the first pixel set is used as a reference pixel set, and the second pixel set is used as a carrier pixel set;

[0009] S2, using a prediction neural network with dual output branches to predict the pixel value and error limit of each carrier pixel in the carrier pixel set according to the reference pixel set;

[0010] S3, according to the predicted pixel values ​​of the carrier pixels, all the carrier pixels in the carrier pixel set are sorted in ascending order to obtain a first carrier pixel sequence; then, according to a preset error threshold value, carrier pixels whose predicted error limit is less than the preset error threshold value are selected from the first carrier pixel sequence to form a second carrier pixel sequence, and the second carrier pixel sequence is sequentially divided into a series of sub-pixel sequences of equal length;

[0011] S4, performing a secret information embedding operation based on pixel value ordering (PVO) on each sub-pixel sequence, and after all the secret information currently to be embedded is embedded, replacing all sub-pixel sequences into the carrier pixel set to obtain the carrier pixel set after the information is embedded;

[0012] S5. Use the second pixel set after embedding the information as the reference pixel set and the first pixel set as the carrier pixel set, and repeat S2 to S4 to obtain the first pixel set after embedding the information; finally, map the pixel values ​​in the first pixel set after embedding the information and the second pixel set after embedding the information to each pixel position of the carrier image to form a carrier image with secret information.

[0013] As a preferred embodiment of the first aspect, the specific implementation method of S2 is as follows:

[0014] A pixel value matrix of the same size as the original carrier image I is constructed as the network input. The values ​​of all carrier pixels in the matrix maintain the original pixel values ​​in the original carrier image I, and the values ​​of all reference pixels are set to zero. Then, a trained prediction neural network with two branches is used to predict the pixel value of each carrier pixel in the carrier pixel set through one output branch based on the constructed network input, and the prediction error limit of each carrier pixel in the carrier pixel set is estimated through another output branch.

[0015] As a preferred embodiment of the first aspect, the prediction neural network is composed of a feature extraction and fusion module, a pixel prediction module and a prediction error limit evaluation module;

[0016] The feature extraction and fusion module includes three parallel convolutional layers with different kernel sizes. The network input is spliced ​​after extracting features in parallel through the three convolutional layers. The spliced ​​features are sequentially passed through multiple residual dense network blocks (RDB) after the convolution operation. The features extracted by each residual dense network block are spliced ​​and then convoluted and added element by element with the spliced ​​features. Finally, the fusion result is output through the MDTA network.

[0017] The pixel prediction module and the prediction error limit evaluation module are respectively used as two output branches, each of which takes the fusion result as input and is composed of two convolutional layers connected with an activation function in the middle.

[0018] As a preferred embodiment of the first aspect, the prediction neural network needs to be trained on a data set before being used for actual reasoning.

[0019] As a preferred embodiment of the first aspect, the specific implementation method of S3 is as follows:

[0020] The carrier pixel set is converted into a carrier pixel sequence, and a predicted pixel value sequence and a predicted error limit sequence corresponding to the carrier pixel sequence are obtained at the same time; the predicted pixel value sequence is arranged in ascending order according to the pixel value size to obtain a new predicted pixel value sequence; the carrier pixel sequence is re-sorted according to the pixel index order in the new predicted pixel value sequence to obtain a first carrier pixel sequence; then, according to a preset error limit threshold value, carrier pixels whose prediction error limit is greater than or equal to the error limit threshold value are eliminated in the first carrier pixel sequence to obtain a second carrier pixel sequence; finally, the second carrier pixel sequence is divided into a series of sub-pixel sequences according to a preset segment length.

[0021] In a second aspect, the present invention provides a reversible information recovery method based on a neural network and global pixel segmentation. For a carrier image with secret information obtained by the reversible information hiding method described in any one of the first aspects above, the method first uses the second pixel set as a reference pixel set and the first pixel set as a carrier pixel set, and performs the inverse operation of the secret information embedding operation described in S2 to S4 to extract the embedded secret information and restore all the first pixels; then uses the first pixel set as a reference pixel and the second pixel as a carrier pixel, and performs the inverse operation of the secret information embedding operation described in S2 to S4 to extract the embedded secret information and restore all the second pixels.

[0022] In a third aspect, the present invention provides a reversible information hiding system based on neural network and global pixel segmentation, comprising:

[0023] An information hiding module, used to obtain a carrier image with secret information according to the reversible information hiding method based on neural network and global pixel segmentation as described in any one of the first aspects above;

[0024] An information recovery module is used to extract secret information from a carrier image with secret information and restore the carrier image according to the reversible information recovery method based on neural network and global pixel segmentation as described in any one of the second aspects above.

[0025] In a fourth aspect, the present invention provides a computer program product, comprising a computer program / instructions, characterized in that when the computer program / instructions are executed by a processor, it can implement the reversible information hiding method based on neural network and global pixel segmentation as described in any one of the first aspects above, or implement the reversible information recovery method based on neural network and global pixel segmentation as described in the second aspect above.

[0026] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it can implement the reversible information hiding method based on neural network and global pixel segmentation as described in any one of the first aspects above, or implement the reversible information recovery method based on neural network and global pixel segmentation as described in the second aspect above.

[0027] In a sixth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0028] The memory is used to store computer programs;

[0029] The processor is used to implement the reversible information hiding method based on neural network and global pixel segmentation as described in any one of the first aspects above, or to implement the reversible information recovery method based on neural network and global pixel segmentation as described in the second aspect above, when executing the computer program.

[0030] Compared with the prior art, the advantages and positive effects of the present invention are:

[0031] The present invention provides a reversible information hiding method based on neural network and global pixel segmentation. The method adopts a two-layer information embedding framework, divides the original image into two pixel sets according to a checkerboard pattern, and sequentially serves as a reference pixel set and a carrier pixel set. In each layer of information embedding operation, a predictor based on a dual-branch neural network is used to predict the carrier pixel set with the reference pixel set and give an error estimate of the corresponding predicted value; the carrier pixel sequence is sorted according to the size of the predicted pixel value, and then the pixels with a prediction error greater than a preset error threshold are eliminated according to the estimated prediction error limit to obtain the selected final carrier pixel sequence; finally, the selected carrier pixel sequence is divided into a sub-pixel sequence set of equal length. In each sub-pixel sequence, a secret information embedding operation based on pixel value ordering (PVO) is used to embed secret information. The dual-branch predictor based on the neural network in the present invention can provide higher pixel value prediction accuracy and perform error estimation on the predicted value; at the same time, the global pixel segmentation based on pixel ordering and pixel selection of the present invention can improve the embedding efficiency. The reversible information hiding method based on a neural network predictor and prediction error modulation of the present invention improves the steganographic performance by utilizing deep learning while maintaining the reversible property of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of the steps of a reversible information hiding method based on neural network and global pixel segmentation;

[0033] Figure 2 A schematic diagram of the module composition of a reversible information hiding system based on neural network and global pixel segmentation;

[0034] Figure 3 It is a schematic diagram of the structure of computer electronic equipment;

[0035] Figure 4 It is a checkerboard pattern for dividing the carrier image in the embodiment of the present invention.

[0036] Figure 5 This is a dual-output prediction neural network diagram in an embodiment of the present invention.

[0037] Figure 6 This is a flow chart based on double-layer embedding in an embodiment of the present invention.

[0038] Figure 7 The following is a schematic diagram of the effect of the example. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0040] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0041] The present invention provides a reversible information hiding method based on neural network and global pixel segmentation. The method adopts a two-layer information embedding framework: first, the original image is divided into two parts according to a checkerboard pattern: a first pixel set and a second pixel set. In the first layer of information embedding, the first pixel is used as a reference pixel set and the second pixel set is used as a carrier pixel set; in the second layer of information embedding, the second pixel set after information embedding is used as a reference pixel set and the first pixel is used as a carrier pixel set. In each layer of information embedding operation, a predictor based on a dual-branch neural network is used to predict the carrier pixel set with the reference pixel set and give an error estimate of the corresponding prediction value; the carrier pixel sequence is sorted according to the size of the predicted pixel value, and then the pixels with a prediction error greater than a preset error threshold value are eliminated according to the estimated prediction error limit to obtain the selected final carrier pixel sequence; finally, the selected carrier pixel sequence is divided into a sub-pixel sequence set of equal length. In each sub-pixel sequence, a secret information embedding operation based on pixel value ordering (PVO) is used to embed the secret information.

[0042] The specific implementation of the reversible information hiding method based on neural network and global pixel segmentation of the present invention is described below.

[0043] like Figure 1 As shown, in a preferred embodiment of the present invention, a reversible information hiding method based on neural network and global pixel segmentation is provided, which includes steps S1 to S5.

[0044] S1. Divide the original carrier image I into two types of pixels, forming a first pixel set and a second pixel set respectively; among the eight neighboring pixels of each pixel, the pixels on the four sides are all pixels of different types, and the pixels on the four diagonals are all pixels of the same type; first use the first pixel set as the reference pixel set and the second pixel set as the carrier pixel set. In the above process of dividing the first pixel set and the second pixel set, it is similar to considering the carrier image as a black and white chessboard, the first pixel set is a white pixel set, and the second pixel set is a black pixel set.

[0045] S2. Use a prediction neural network with dual output branches to predict the pixel value and error limit of each carrier pixel in the carrier pixel set based on the reference pixel set.

[0046] In an embodiment of the present invention, the specific implementation method of the above step S2 is as follows:

[0047] A pixel value matrix of the same size as the original carrier image I is constructed as the network input. The values ​​of all carrier pixels in the matrix maintain the original pixel values ​​in the original carrier image I, and the values ​​of all reference pixels are set to zero. Then, a trained prediction neural network with two branches is used to predict the pixel value of each carrier pixel in the carrier pixel set through one output branch based on the constructed network input, and the prediction error limit of each carrier pixel in the carrier pixel set is estimated through another output branch.

[0048] In an embodiment of the present invention, the above-mentioned prediction neural network is composed of a feature extraction and fusion module, a pixel prediction module and a prediction error limit assessment module. The feature extraction and fusion module includes three parallel convolutional layers with different kernel sizes. The network input is spliced ​​after extracting features in parallel through three convolutional layers. The spliced ​​features are sequentially passed through multiple residual dense network blocks (RDB) after convolution operation (the convolution operation is composed of 1×1 convolutional layer, ReLU activation function, and 3×3 convolutional layer cascade). The features extracted by each residual dense network block are spliced ​​and then passed through a convolution operation (the convolution operation is composed of 1×1 convolutional layer and 3×3 convolutional layer cascade) and then element-wise summed with the spliced ​​features. Finally, the fusion result is output through the MDTA network. The above-mentioned pixel prediction module and the prediction error limit assessment module are respectively used as two output branches, each of which takes the above-mentioned fusion result as input and is composed of a ReLU activation function connected in the middle of two convolutional layers.

[0049] It should be noted that the above prediction neural network needs to be trained on the data set before being used for actual reasoning. The training process of the neural network belongs to the prior art and will not be described in detail.

[0050] S3. According to the predicted pixel values ​​of the carrier pixels, all the carrier pixels in the carrier pixel set are sorted in ascending order to obtain a first carrier pixel sequence; then, according to a preset error threshold value, carrier pixels whose predicted error limit is less than the preset error threshold value are selected from the first carrier pixel sequence to form a second carrier pixel sequence, and the second carrier pixel sequence is divided into a series of sub-pixel sequences of equal length in sequence.

[0051] In an embodiment of the present invention, the specific implementation method of the above step S3 is as follows:

[0052] The carrier pixel set is converted into a carrier pixel sequence, and a predicted pixel value sequence and a predicted error limit sequence corresponding to the carrier pixel sequence are obtained at the same time; the predicted pixel value sequence is arranged in ascending order according to the pixel value size to obtain a new predicted pixel value sequence; the carrier pixel sequence is re-sorted according to the pixel index order in the new predicted pixel value sequence to obtain a first carrier pixel sequence; then, according to a preset error limit threshold value, carrier pixels whose prediction error limit is greater than or equal to the error limit threshold value are eliminated in the first carrier pixel sequence to obtain a second carrier pixel sequence; finally, the second carrier pixel sequence is divided into a series of sub-pixel sequences according to a preset segment length.

[0053] S4. Perform a secret information embedding operation based on pixel value ordering (PVO) on each sub-pixel sequence. After all the secret information that needs to be embedded is embedded, replace all sub-pixel sequences into the carrier pixel set to obtain the carrier pixel set after the information is embedded.

[0054] It should be noted that the secret information embedding operation is based on pixel value ordering (PVO). The secret information embedding operation based on PVO belongs to the information hiding technology in the prior art. It can rearrange the pixel values ​​through sorting to provide space for the embedding of secret information. The secret information is embedded in the sorting result of the image pixel value to achieve the secret transmission of information. The core of the present invention is to form a carrier sequence (i.e. the aforementioned sub-pixel sequence) for the secret information embedding operation, but the embedding of secret information into the carrier sequence based on PVO can be achieved through the prior art. Theoretically, any secret information embedding algorithm based on PVO can be used in the present invention. In the embodiment of the present invention, it is preferred to use an information embedding method based on Multi-pass IPVO to embed secret information in each sub-pixel sequence.

[0055] S5. Use the second pixel set after embedding the information as the reference pixel set and the first pixel set as the carrier pixel set, and repeat S2 to S4 to obtain the first pixel set after embedding the information; finally, map the pixel values ​​in the first pixel set after embedding the information and the second pixel set after embedding the information to each pixel position of the carrier image to form a carrier image with secret information.

[0056] It should be noted that, in the first layer of information embedding, when the first pixel set is used as the reference pixel set and the second pixel set is used as the carrier pixel set, S2 to S4 are executed for the first time, and secret information can be embedded in the second pixel set; and in the second layer of information embedding described in the above step S5, the second pixel set after the information is embedded is used as the reference pixel set and the first pixel set is used as the carrier pixel set, and S2 to S4 are repeated. At this time, since the reference pixel set and the carrier pixel set are swapped, secret information can be embedded in the first pixel set. However, for the two layers of information embedding, the operations of embedding secret information are all S2 to S4, and the operations are the same, and the only difference is that the carrier pixel set is different.

[0057] In addition, in an embodiment of the present invention, a reversible information recovery method based on a neural network and global pixel segmentation is provided, which can extract secret information and restore pixels for the carrier image with secret information obtained by the above reversible information hiding method. The process of secret information extraction and pixel restoration is actually the inverse operation of the above steps S1 to S5. Since information embedding is performed in two layers, secret information extraction and pixel restoration also need to be performed in two layers. First, the first layer of secret information extraction and pixel restoration is performed, that is, the second pixel set is used as the reference pixel set, the first pixel set is used as the carrier pixel set, and the inverse operation of the secret information embedding operation shown in S2 to S4 is performed in the above step S5 to extract the embedded secret information and restore all the first pixels; then the second layer of secret information extraction and pixel restoration is performed, that is, the first pixel set is used as the reference pixel set, the second pixel is used as the carrier pixel, and the inverse operation of the secret information embedding operation shown in S2 to S4 is performed for the first time to extract the embedded secret information and restore all the second pixels.

[0058] Similarly, based on the same inventive concept, Figure 2 As shown, the present invention also provides a reversible information hiding system based on neural network and global pixel segmentation, which includes:

[0059] An information hiding module, used to obtain a carrier image with secret information according to the above-mentioned reversible information hiding method based on neural network and global pixel segmentation;

[0060] The information recovery module is used to extract secret information from a carrier image with secret information and restore the carrier image according to the above-mentioned reversible information recovery method based on neural network and global pixel segmentation.

[0061] It should be noted that the above-mentioned information hiding module and information recovery module can be respectively set on different computer devices.

[0062] Similarly, based on the same inventive concept, Figure 3 As shown, the present invention also provides a computer electronic device provided in the above embodiment, which includes a memory and a processor;

[0063] The memory is used to store computer programs;

[0064] The processor is used to implement the reversible information hiding method based on neural network and global pixel segmentation as described above, or to implement the reversible information recovery method based on neural network and global pixel segmentation as described above when executing the computer program.

[0065] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0066] Therefore, based on the same inventive concept, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it can implement the reversible information hiding method based on neural network and global pixel segmentation as described above, or implement the reversible information recovery method based on neural network and global pixel segmentation as described above.

[0067] Therefore, based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the reversible information hiding method based on neural network and global pixel segmentation as described above, or implement the reversible information recovery method based on neural network and global pixel segmentation as described above.

[0068] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0069] It can be understood that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0070] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0071] In order to better understand the specific implementation process of the reversible information hiding method based on neural network and global pixel segmentation and the subsequent reversible information recovery method shown in S1 to S5 of the present invention, the specific steps and technical effects of the present invention are detailed below through an example.

[0072] Example

[0073] In this embodiment, the reversible information hiding method based on neural network and global pixel segmentation is shown in steps 1 to 5, and the subsequent reversible information recovery method is shown in step 6.

[0074] Step 1: Image segmentation. The carrier image I is divided as follows Figure 4 The checkerboard pattern shown is divided into two parts: the white pixel set I W and black pixel set I B The specific division method can be formulated as follows:

[0075] I W ={I(x,y)|mod(x+y,2)=0},

[0076] I B ={I(x,y)|mod(x+y,2)=1},

[0077] Where (x, y) represents the coordinates of the pixel in image I.

[0078] Step 2: Pixel prediction of the first layer. W As the reference pixel I R , I B As carrier pixel I C . Using reference pixel set I R Construct the input image, use the prediction neural network to predict the carrier pixel set IC and estimate the prediction error limit. The detailed prediction steps are as follows:

[0079] Step 2-1: Construct a prediction neural network with two output branches, denoted as net(). The structure of the dual-branch prediction neural network used in this embodiment is as follows: Figure 5 As shown in (a) in the figure, the prediction neural network consists of a feature extraction and fusion module, a pixel prediction module, and a prediction error limit assessment module. The feature extraction and fusion module contains three parallel convolutional layers with different kernel sizes (the convolution kernel sizes are 7×7, 5×5, and 3×3, respectively). The network input is spliced ​​after extracting features through three convolutional layers in parallel. The spliced ​​features are convolutionally operated (the convolution operation consists of a 1×1 convolutional layer, a ReLU activation function, and a 3×3 convolutional layer cascade) and then sequentially passed through multiple residual dense network blocks (RDB). The features extracted by each residual dense network block are spliced ​​and convolutionally operated (the convolution operation consists of a 1×1 convolutional layer and a 3×3 convolutional layer cascade) and then element-wise summed with the spliced ​​features. Finally, the fusion result is output through the MDTA network. The above-mentioned pixel prediction module and prediction error limit assessment module are respectively used as two output branches, each of which takes the above-mentioned fusion result as input and is composed of two convolutional layers connected with a ReLU activation function.

[0080] It should be noted that the MDTA network belongs to the existing technology, and its structure is as follows Figure 5 As shown in (b) in .

[0081] Step 2-2: I R Based on, construct the input I of the prediction network in First, use I to I in Initialize and then keep I RThe corresponding pixel is then set to zero for the remaining half of the pixel values. This process can be formulated as follows:

[0082] I in =[I R ,0].

[0083] Step 2-3: I in As the input of the prediction neural network, the trained prediction network net() calculates the carrier pixel set I C The predicted value of each pixel in and the absolute error limit of the corresponding predicted value. The output predicted pixel value set is recorded as The corresponding prediction error set is I ε , then the prediction network process can be formulated as follows:

[0084]

[0085] Step 3: According to the predicted value and the estimated prediction error limit, the I C Split into a sequence of equal-length sub-pixel sequences, the specific steps are as follows:

[0086] Step 3-1: Set the carrier pixel set I C Convert to sequence (I c (1),I c (2),…,I c (L)), and its corresponding predicted value sequence is And the forecast error limit sequence is Where L represents I C The total number of pixels in .

[0087] Step 3-2: Forecast value sequence Arrange in ascending order The mapping σ means that the σ(i)th pixel in the original sequence is the i-th pixel in the new sequence.

[0088] Step 3-3: According to the permutation mapping σ, the carrier pixel sequence (I c (1),I c (2),…,I c (L)) are sorted to obtain (I c (σ(1)),I c (σ(2))…,I c (σ(L))).

[0089] Step 3-4: According to the preset error limit threshold t ε , in the sorted carrier pixel sequence (I c (σ(1)),I c (σ(2)),…,Ic (σ(L))) removes the prediction error greater than t ε pixels, and obtain a new sequence (I c (σ(i1)),I c (σ(i2)),…,I c (σ(i T ), recorded as Where index i j Indicates that the prediction error is less than or equal to t ε The position of the pixel, T represents the total number of pixels retained.

[0090] Step 3-5: Sequence Split into a sequence of carrier pixel subsequences of length t, that is, in represents the k-th pixel subsequence, 1≤k≤L t .

[0091] Step 4: Use the information embedding method based on Multi-pass IPVO to embed the secret information s1 into the sub-pixel sequence in sequence until the allocated secret information load is fully embedded. Assume that the sub-sequence where p i yes The [(k-1)*t+i]th pixel of The information embedding process is as follows:

[0092] Step 4-1: Substitute the sequence (p1, p2, …, p t ) are sorted in ascending order to obtain (p τ(1) ,p τ(2) ,…,p τ(t) ), where the mapping τ means that the τ(i)th pixel in the original sequence is the i-th pixel in the new sequence. Calculate p τ(t) and (p τ(1) ,p τ(2) ,…,p τ(t-1) ), the specific formula is as follows:

[0093] d L (i) = p x -p y ,

[0094] Among them, 1≤i≤t-1, x=min{τ(t),τ(i)}, y=max{τ(t),τ(i)}.

[0095] Step 4-2: For the sequence (p τ(1) ,p τ(2) ,…,pτ(t-1) ) are embedded as follows:

[0096]

[0097] Where b is the first bit of the information not embedded in the current s, 1≤i≤t-1. The new carrier pixel sequence obtained after this embedding operation is recorded as in

[0098] Step 4-3: Sequence Sort in ascending order again to get a new sequence The meaning of the mapping ω is similar to that in step 3. Calculation with sequence The difference of all pixels, the specific formula is as follows:

[0099]

[0100] Among them, x=min{ω(1),ω(i)}, y=max{ω(1),ω(i)}, 2≤i≤t.

[0101] Step 4-4: Sequence The pixels in are embedded as follows:

[0102]

[0103] Where b is the first bit of the information not embedded in the current s, 2≤i≤t. The pixel sequence after this embedding operation is recorded as in This sequence is for The final result after information embedding is recorded as

[0104] Step 5: Execute the second layer of information embedding. The black pixel set after embedding information is used as the reference pixel, and the white pixel is used as the carrier pixel. Refer to step 2 to use the dual-branch prediction neural network to predict the carrier pixel value and estimate the prediction error, and then perform steps 3 and 4 in sequence. The two-layer embedding process used in this method is as follows Figure 6 shown.

[0105] Step 6: Information extraction and image restoration. The receiver extracts and restores the carrier pixel set according to the inverse operation of the information embedding operation, and performs two layers of information extraction and carrier pixel restoration. In the first layer of information extraction and carrier pixel restoration, the black pixel set is used as the reference pixel for pixel prediction, and the white pixel set is segmented, and then information extraction and carrier pixel restoration are performed; in the second layer of information extraction and carrier pixel restoration, the restored white pixel set is used as the reference pixel for pixel prediction, and the black pixel set is segmented, and then information extraction and carrier pixel restoration are performed. The specific operations of information extraction and carrier pixel restoration for the embedding operation in step 4 are as follows:

[0106] Step 6-1: Sort by calculate with sequence The difference of all pixels is as follows:

[0107]

[0108] Among them, x=min{ω′(1),ω′(i)}, y=max{ω′(1),ω′(i)}, 2≤i≤t.

[0109] Step 6-2: From the sequence Any pixel in The operation of extracting secret information b is as follows:

[0110]

[0111] and will Perform the following recovery operations:

[0112]

[0113] The result after this recovery operation is in

[0114] Step 6-3: Sequence Sort by calculate with sequence The difference of all pixels in is as follows:

[0115]

[0116] Among them, x=min{τ′(t),τ′(i)}, y=max{τ′(t),τ′(i)}, 1≤i≤t-2.

[0117] Step 6-4: From Any pixel in The operation of extracting secret information b is as follows:

[0118]

[0119] And the pixels Perform the following recovery operations:

[0120]

[0121] The final recovered carrier pixel subsequence is (p1, p2, ..., p t ),in

[0122] Figure 7 The experimental results of this embodiment on the test image are given, where (a) is the input of the prediction neural network, (b) is the output of the pixel prediction module, (c) is the statistical distribution of the output of the prediction error limit evaluation module, and (d) is the steganographic image after embedding information (load: 5000 bits, t ε =1). After experimental verification, the above-mentioned reversible information hiding method based on neural network and global pixel segmentation in this embodiment can better embed secret information with high embedding efficiency because the dual-branch predictor based on neural network can provide higher pixel value prediction accuracy and perform error estimation on the prediction value. And through step 6, the image can be further restored and secret information can be extracted from it. The reversible information hiding method based on neural network predictor and prediction error modulation of the present invention improves the steganographic performance by using deep learning while maintaining the reversible property of the method.

[0123] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A reversible information hiding method based on neural network and global pixel segmentation, characterized in that: include: S1, dividing the original carrier image I into two types of pixels, forming a first pixel set and a second pixel set respectively; Among the eight neighboring pixels of each pixel, the pixels on the four sides are all pixels of different types, and the pixels on the four diagonals are all pixels of the same type; firstly, the first pixel set is used as the reference pixel set, and the second pixel set is used as the carrier pixel set; S2, using a prediction neural network with dual output branches to predict the pixel value and error limit of each carrier pixel in the carrier pixel set according to the reference pixel set; S3, according to the predicted pixel values ​​of the carrier pixels, all the carrier pixels in the carrier pixel set are sorted in ascending order to obtain a first carrier pixel sequence; then, according to a preset error threshold value, carrier pixels whose predicted error limit is less than the preset error threshold value are selected from the first carrier pixel sequence to form a second carrier pixel sequence, and the second carrier pixel sequence is sequentially divided into a series of sub-pixel sequences of equal length; S4, performing a secret information embedding operation based on pixel value ordering (PVO) on each sub-pixel sequence, and after all the secret information currently to be embedded is embedded, replacing all sub-pixel sequences into the carrier pixel set to obtain the carrier pixel set after the information is embedded; S5. Use the second pixel set after embedding the information as the reference pixel set and the first pixel set as the carrier pixel set, and repeat S2 to S4 to obtain the first pixel set after embedding the information; finally, map the pixel values ​​in the first pixel set after embedding the information and the second pixel set after embedding the information to each pixel position of the carrier image to form a carrier image with secret information.

2. The reversible information hiding method based on neural network and global pixel segmentation according to claim 1 is characterized in that: The specific implementation method of S2 is as follows: A pixel value matrix of the same size as the original carrier image I is constructed as the network input. The values ​​of all carrier pixels in the matrix maintain the original pixel values ​​in the original carrier image I, and the values ​​of all reference pixels are set to zero. Then, a trained prediction neural network with two branches is used to predict the pixel value of each carrier pixel in the carrier pixel set through one output branch based on the constructed network input, and the prediction error limit of each carrier pixel in the carrier pixel set is estimated through another output branch.

3. The reversible information hiding method based on neural network and global pixel segmentation according to claim 2 is characterized in that: The prediction neural network consists of a feature extraction and fusion module, a pixel prediction module and a prediction error limit evaluation module; The feature extraction and fusion module includes three parallel convolutional layers with different kernel sizes. The network input is spliced ​​after extracting features in parallel through the three convolutional layers. The spliced ​​features are sequentially passed through multiple residual dense network blocks (RDB) after the convolution operation. The features extracted by each residual dense network block are spliced ​​and then convoluted and added element by element with the spliced ​​features. Finally, the fusion result is output through the MDTA network. The pixel prediction module and the prediction error limit evaluation module are respectively used as two output branches, each of which takes the fusion result as input and is composed of two convolutional layers connected with an activation function in the middle.

4. The reversible information hiding method based on neural network and global pixel segmentation according to claim 1, characterized in that: The prediction neural network needs to be trained on a dataset before being used for actual reasoning.

5. The reversible information hiding method based on neural network and global pixel segmentation according to claim 1, characterized in that: The specific implementation method of S3 is as follows: The carrier pixel set is converted into a carrier pixel sequence, and a predicted pixel value sequence and a predicted error limit sequence corresponding to the carrier pixel sequence are obtained at the same time; the predicted pixel value sequence is arranged in ascending order according to the pixel value size to obtain a new predicted pixel value sequence; the carrier pixel sequence is re-sorted according to the pixel index order in the new predicted pixel value sequence to obtain a first carrier pixel sequence; then, according to a preset error limit threshold value, carrier pixels whose prediction error limit is greater than or equal to the error limit threshold value are eliminated in the first carrier pixel sequence to obtain a second carrier pixel sequence; finally, the second carrier pixel sequence is divided into a series of sub-pixel sequences according to a preset segment length.

6. A reversible information recovery method based on neural network and global pixel segmentation, characterized in that: For a carrier image with secret information obtained according to the reversible information hiding method described in any one of claims 1 to 5, first, the second pixel set is used as a reference pixel set, and the first pixel set is used as a carrier pixel set, and the inverse operation of the secret information embedding operation described in S2 to S4 is performed to extract the embedded secret information and restore all the first pixels; then, the first pixel set is used as a reference pixel, and the second pixel is used as a carrier pixel, and the inverse operation of the secret information embedding operation described in S2 to S4 is performed to extract the embedded secret information and restore all the second pixels.

7. A reversible information hiding system based on neural network and global pixel segmentation, characterized in that: include: An information hiding module, used to obtain a carrier image with secret information according to the reversible information hiding method based on neural network and global pixel segmentation according to any one of claims 1 to 5; An information recovery module is used to extract secret information from a carrier image with secret information and restore the carrier image according to the reversible information recovery method based on neural network and global pixel segmentation as described in claim 6.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it can implement the reversible information hiding method based on a neural network and global pixel segmentation as described in any one of claims 1 to 5, or implement the reversible information recovery method based on a neural network and global pixel segmentation as described in claim 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, it can implement the reversible information hiding method based on neural network and global pixel segmentation as described in any one of claims 1 to 5, or implement the reversible information recovery method based on neural network and global pixel segmentation as described in claim 6.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the reversible information hiding method based on neural network and global pixel segmentation as described in any one of claims 1 to 5, or the reversible information recovery method based on neural network and global pixel segmentation as described in claim 6 when executing the computer program.

Images