A reversible information hiding method based on intelligent predictors and additive secret shared ciphertext domain.
By using an intelligent predictor and additive secret sharing of the ciphertext domain, the method utilizes residual networks to predict the real pixels of the carrier image and performs encrypted compression, thus overcoming the shortcomings of traditional reversible information hiding methods in terms of accuracy and robustness, and achieving reversible information hiding with high security and privacy protection.
Patent Information
- Application Number
- CN202311513182.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-14
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-11-14
AI Technical Summary
Traditional reversible information hiding methods have low accuracy and robustness in embedding and extracting hidden information, and cannot meet the requirements of high security and privacy protection.
A method based on intelligent predictors and additive secret sharing ciphertext domains is adopted. The real pixels of the original carrier image are predicted by the residual network. Combined with the additive secret sharing algorithm and the adaptive joint coding method, the carrier image is encrypted and compressed to generate a shared encrypted carrier image. The space freed up by the secret information is obtained by the adaptive joint coding method.
It improves the accuracy and robustness of embedding and extracting hidden information, meeting the needs of high security and privacy protection.
Smart Images

Figure CN117560454B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of multimedia information security, and in particular relates to a reversible information hiding method based on intelligent predictors and additive secret shared ciphertext domains. Background Technology
[0002] Reversible information hiding is a technique for embedding information into digital media. Traditional information hiding methods typically employ an irreversible approach, fusing hidden information with the carrier media, rendering the original content of the carrier media unrecoverable. However, for certain applications, such as digital watermarking, copyright protection, and secure communication, reversible information hiding is an important technological requirement. Reversible information hiding technology aims to embed hidden information into media without losing its original content, and to fully recover the original content of the media. This technology requires the combination of various algorithms and methods, such as data compression, error diffusion, and optimal coding, to ensure the security and recoverability of the hidden information.
[0003] However, in some scenarios, the requirements for the security and privacy protection of hidden information are higher, but traditional reversible information hiding methods have low accuracy and robustness in embedding and extracting hidden information. Summary of the Invention
[0004] This invention provides a method for reversible information hiding based on intelligent predictors and additive secret shared ciphertext domains to overcome the aforementioned technical problems.
[0005] To achieve the above objectives, the technical solution of the present invention is as follows:
[0006] A reversible information hiding method based on intelligent predictors and additive secret shared ciphertext domains includes the following steps:
[0007] S1: Preprocess the original carrier image to obtain a preprocessed image that retains some of the original carrier image information;
[0008] S2: Establish a residual network for pixel prediction to obtain a predicted image containing the real pixels of the original carrier image based on the preprocessed image that retains part of the original carrier image information;
[0009] S3: Using an additive secret sharing algorithm, the original carrier image and the predicted image containing the real pixels of the original carrier image are encrypted to obtain a shared encrypted carrier image;
[0010] S4: Based on the adaptive joint coding method and the shared encrypted carrier image, the shared encrypted carrier image is compressed to free up space for secret information, so as to obtain the carrier image obtained by encoding and compression;
[0011] S5: Based on the carrier image obtained by the encoding and compression, acquire the encrypted image and send it to the secret receiver to complete the hiding of reversible information.
[0012] Beneficial Effects: This invention provides a reversible information hiding method based on an intelligent predictor and an additive secret shared ciphertext domain. Using a residual network for pixel prediction, the method predicts the preprocessed original carrier image and employs an additive secret sharing algorithm to obtain a shared encrypted carrier image sent to the party hiding the secret information. The party hiding the secret information then compresses the shared encrypted carrier image using an adaptive joint coding method to free up space for the secret information, thereby obtaining the encoded and compressed carrier image and ultimately the ciphertext image, completing the hiding of reversible information. The residual network for pixel prediction improves the accuracy and robustness of embedding and extracting hidden information. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0014] Figure 1 This is a flowchart of the overall method of the present invention.
[0015] Figure 2 This is an example diagram of image splitting according to an embodiment of the present invention.
[0016] Figure 3 This is a structural diagram of the intelligent predictor based on residual networks designed in this invention.
[0017] Figure 4 This is an example diagram illustrating the prediction process and prediction results of an embodiment of the present invention.
[0018] Figure 5 This is an example diagram of the carrier image generated according to an embodiment of the present invention.
[0019] Figure 6 This is an example diagram of the original image bit-plane splitting and random combination in an embodiment of the present invention.
[0020] Figure 7 This is an example diagram of generating a shared image from an original image according to an embodiment of the present invention.
[0021] Figure 8 This is an example diagram of the bit-plane splitting and random combination of the image bit-plane carrying the true value of the fork set in an embodiment of the present invention.
[0022] Figure 9This is an example diagram of generating a shared image by retaining the true value carrier image of the fork set in an embodiment of the present invention.
[0023] Figure 10 This is an example diagram of compression in an embodiment of the present invention.
[0024] Figure 11a This is the original image of the carrier in an embodiment of the present invention.
[0025] Figure 11b This is a compressed example diagram of an embodiment of the present invention.
[0026] Figure 11c This is an example diagram illustrating the embedding of secret information in an embodiment of the present invention.
[0027] Figure 11d This is the image restored according to an embodiment of the present invention.
[0028] Figure 12 This is a flowchart of the overall method in the embodiments of the present invention. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] This embodiment provides a reversible information hiding method based on an intelligent predictor and additive secret shared ciphertext domain, such as... Figure 1 and Figure 12 It includes the following steps:
[0031] S1: Preprocess the original carrier image of H×W to obtain a preprocessed image that retains some of the original carrier image information;
[0032] Preferably, in step S1, the method for preprocessing the original image is as follows:
[0033] S11: Divide the original carrier image into a non-overlapping 4x4 matrix of image blocks, and number each image block in the matrix according to the raster scan order, namely p1, p2, p3, p4, p5, p6, p7, p8, p9, p 10 ,p 11 ,p 12 ,p 13 ,p 14 ,p 15 ,p 16 ;
[0034] S12: Rearrange the numbers p1, p3, p9, p... 11 The segmented image blocks are used as a fork set; the blocks numbered p2, p4, p 10 ,p 12 The segmented image blocks are used as a set of circles;
[0035] The numbers are p5, p7, p 13 ,p 15 The segmented image blocks are used as a triangular set;
[0036] The numbers p6, p8, p 14 ,p 16 The segmented image blocks are treated as a set of squares;
[0037] S13: Set the pixel values of all block image blocks in the original carrier image except for the fork set to 0 to form the first preprocessed image;
[0038] Set the pixel values of all image blocks in the original carrier image except for the circle set to 0 to form the second preprocessed image;
[0039] Set the pixel values of all image blocks in the original carrier image except for the triangular set to 0 to form the third preprocessed image;
[0040] The pixel values of all image blocks in the original carrier image, except for the square set, are set to 0 to form the fourth preprocessed image.
[0041] Specifically, in this embodiment, the original carrier image C is then divided into a non-overlapping 4x4 matrix of image blocks, and each image block in the matrix is numbered. The image blocks are then divided into four sets according to their numbers: a cross set C1, a circle set C2, a triangle set C3, and a square set C4. The cross set C1 contains image blocks numbered p1, p3, p9, p1, p2, p3, p4, p5, p6, p7, p8, p9, p1 ... 11 The pixel values at other locations are set to 0, forming the first preprocessed image; the triangular set C2 contains image blocks numbered p2, p4, p 10 ,p 12 The pixel values at other locations are set to 0, forming the second preprocessed image; the triangular set C3 contains image blocks numbered p5, p7, p8, p9, p1, p1, p2, p3, p4, p5, p6, p7, p8, p9, p1, p1, p1, p1, p1, p1, p1, p2, p1 ... 13 ,p 15 The pixel values at other locations are set to 0, forming the third preprocessed image; the square set C4 contains image blocks numbered p6, p8, p9, p10, p10, p11, p20, p30, p40, p50, p60, p70, p8, p90, p10, p11, p12, p13, p14, p15, p16, p17, p18, p19, p10, p 14 ,p 16 The pixel values at other locations are set to 0, forming the fourth preprocessed image.
[0042] S2: Establish a residual network for pixel prediction to obtain a predicted image containing the real pixels of the original carrier image based on the preprocessed image that retains part of the original carrier image information;
[0043] Preferably, in S2, the residual network for pixel prediction includes an input layer, a feature extraction layer, a prediction layer and an output layer, an attention residual block and a residual block;
[0044] The input layer is used to input the preprocessed image, which retains some of the original carrier image information, into the residual network used for pixel prediction;
[0045] The input end of the feature extraction layer is connected to the output end of the input layer, and is used to extract the feature representation of the real pixel part of the image after preprocessing while retaining some of the original carrier image information;
[0046] Specifically, the feature extraction layer includes multiple convolutional layers, batch normalization layers, LeakyReLU activation function, and attention residual blocks;
[0047] Multiple convolutional layers are connected end-to-end in sequence. The first convolutional layer is connected to the output of the input layer, and the last convolutional layer is connected to a batch normalization layer. The batch normalization layer is activated by the LeakyReLU activation function to obtain local features of the real pixel part of the preprocessed image.
[0048] The attention residual block is connected to the LeakyReLU activation function to obtain the feature representation of the real pixel part of the preprocessed image, which retains some of the original carrier image information, based on the local features of the real pixel part of the preprocessed image.
[0049] Specifically, the feature extraction layer is used to extract feature representations from the preprocessed image. The feature extraction layer is deployed after the input layer, and consists of 18 convolutional layers connected in sequence. Each convolutional layer includes multiple convolutional kernels, with a kernel size of 3×3. The first convolutional layer in the feature extraction layer has a 1-channel input, a 4-channel output, a stride of 1, and padding of 1. Next, a batch normalization layer and a LeakyReLU activation function are connected. Then, an attention residual block is connected. Next, the local features of the preprocessed image are fed into a residual block, where each residual block includes... Finally, the final feature representation is obtained through an attention residual block.
[0050] Specifically, the feature representation of the true pixel portion of the image, after preprocessing and retaining some of the original carrier image information, is input to the prediction layer. This includes feature information about the known image portion in the image pixel prediction task. Specifically, these features may include the following: Local and global contextual information: The feature extraction layer extracts local and global contextual information from the known three-quarters of the image pixels, which can be used to help predict the missing quarter pixel. This information includes the values, texture, and structure of surrounding pixels, as well as the overall background and context of the image. Structural features: The feature extraction layer may extract information about the image structure, such as edges, lines, and object outlines. These structural features help predict the shape and layout of the missing pixel. Texture features: The feature extraction layer can include texture features of the image, which describe the texture patterns between local pixels, helping to better estimate the texture of the missing pixel. Other image features: Depending on the network design, the feature extraction layer may also extract other task-related image features, such as contrast. In summary, the goal of the feature extraction layer is to extract various information and features relevant to the image pixel prediction task from the known three-quarters of the image pixels. These features will be used to predict the missing quarter pixel to reconstruct the complete image as accurately as possible. This design allows the network to use known information to fill in missing pixels in order to generate an estimate of the missing parts.
[0051] The input of the prediction layer is connected to the output of the feature extraction layer, and is used to obtain a predicted image containing the real pixels of the original carrier image based on the feature representation of the real pixel part of the image after preprocessing and retaining some of the original carrier image information.
[0052] Specifically, the prediction layer includes seven residual blocks. Assuming the number of input channels for the first residual block is x, the number of output channels for the first residual block is 2x. The number of input channels for the second, third, fourth, fifth, sixth, and seventh residual blocks are 2x, 3x, 4x, 3x, 2x, and x, respectively. The number of output channels for the second, third, fourth, fifth, sixth, and seventh residual blocks are 3x, 4x, 3x, 2x, x, and x, respectively.
[0053] The input terminal of the output layer is connected to the output terminal of the prediction layer, and is used to output a predicted image containing the real pixels of the original carrier image.
[0054] Specifically, a prediction layer is connected after the feature extraction layer. The prediction layer contains seven residual blocks. The first residual block has 8 input channels and 16 output channels. The second residual block has 16 input channels and 32 output channels. The third residual block has 32 input channels and 64 output channels. The fourth residual block has 64 input channels and 32 output channels. The fifth residual block has 32 input channels and 16 output channels. The sixth residual block has 16 input channels and 8 output channels. The seventh residual block has 8 input channels and 1 output channel.
[0055] Specifically, the residual network for pixel prediction in this embodiment includes an input layer, a feature extraction layer, a prediction layer, and an output layer. It also includes multiple attention residual blocks and residual groups. Each residual group comprises multiple residual blocks, and each residual block includes two or three convolutional layers and a skip connection. In a residual block containing two convolutional layers, the first convolutional layer is used to further extract features, and the second convolutional layer is used to map the features to the spatial dimension of the target output. Padding is 2, stride is 1, and kernel size is 3×3. A residual block containing three convolutional layers, compared to a residual block containing two layers, only adds a pointwise convolutional layer with a kernel size of 1×1 on the skip connections to adjust the input feature dimension. Padding is 0, stride is 1, and kernel size is 3×3. Each attention residual block consists of a residual block containing two convolutional layers and a channel attention machine module. The channel attention machine module first performs channel-based global max pooling and global average pooling to obtain two H×W×1 feature maps. Then, a 7×7 convolution operation is applied to reduce the dimension to one channel. The resulting features are then used to generate spatial attention features through a sigmoid activation function. Finally, the spatial attention features are multiplied element-wise with the input of the attention residual block to obtain the final generated features. Each residual block consists of a residual block containing three convolutional layers and a residual block containing two convolutional layers connected together.
[0056] Preferably, in step S2, the method for obtaining the predicted image containing the actual pixels of the original carrier image is as follows:
[0057] S21: Combine the second preprocessed image, the third preprocessed image, and the fourth preprocessed image to form the first input image;
[0058] The first preprocessed image, the third preprocessed image, and the fourth preprocessed image are combined to form the second input image;
[0059] The first preprocessed image, the second preprocessed image, and the fourth preprocessed image are combined to form the third input image;
[0060] The first preprocessed image, the second preprocessed image, and the third preprocessed image are combined to form the fourth input image;
[0061] S22: Input the first input image, the second input image, the third input image, and the fourth input image into the residual network for pixel prediction to obtain the first predicted image, the second predicted image, the third predicted image, and the fourth predicted image, respectively; and combine the first predicted image, the second predicted image, the third predicted image, and the fourth predicted image to obtain the predicted image of the original carrier image; wherein, the method of combining the first predicted image, the second predicted image, the third predicted image, and the fourth predicted image is a conventional technique and will not be described in detail here.
[0062] S23: Replace the image block at the position corresponding to the block image block in the fork set in the predicted image of the original carrier image with the block image block in the fork set to obtain the first predicted image containing the real pixels of the original carrier image.
[0063] The image block at the position corresponding to the block image block in the circle set in the predicted image of the original carrier image is replaced with the block image block in the circle set to obtain a second predicted image containing the real pixels of the original carrier image.
[0064] The image block at the position corresponding to the block image block in the triangle set in the predicted image of the original carrier image is replaced with the block image block in the triangle set to obtain a third predicted image containing the real pixels of the original carrier image.
[0065] The image block at the position corresponding to the block image block in the square set in the predicted image of the original carrier image is replaced with the block image block in the square set to obtain a fourth predicted image containing the real pixels of the original carrier image.
[0066] Specifically, the set of three sets—C1 (fork), C2 (circle), C3 (triangle), and C4 (square)—are combined to form an input image. This input image is then fed into a trained residual network for pixel prediction to obtain the first, second, third, and fourth predicted images. The input image combining C1, C2, and C3 is used as input to obtain the prediction result C4′; the input image combining C2, C3, and C4 is used as input to obtain the prediction result C1′; the input image combining C1, C3, and C4 is used as input to obtain the prediction result C2′; and the input image combining C1, C2, and C3 is used as input to obtain the prediction image C′ of the original carrier image. Finally, C1′, C2′, C3′, and C4′ are merged to obtain the predicted image C′ of the original carrier image. Finally, the cross set C1, circle set C2, triangle set C3, and square set C4 from the original carrier image are replaced with their corresponding positions in the predicted image C′, resulting in a carrier image that retains the true values of the cross set C1, circle set C2, triangle set C3, and square set C4. This is the first predicted image C′ containing the true pixels of the original carrier image. fork The second predicted image C contains the actual pixels of the original carrier image. c ′ ircle The third is the predicted image C, which contains the actual pixels of the original carrier image. t ′ riangle The fourth is the predicted image C, which contains the actual pixels of the original carrier image. s ′ quare .
[0067] S3: Using an additive secret sharing algorithm, the original carrier image and the predicted image containing the real pixels of the original carrier image are encrypted to obtain a shared encrypted carrier image;
[0068] Preferably, in step S3, the method for obtaining the shared encrypted carrier image is as follows:
[0069] S31: Decompose all pixels in the original carrier image and the predicted image containing the real pixels of the original carrier image into 8 bit planes as follows:
[0070]
[0071] In the formula: This represents the k-th bit plane of the block image in the i-th row and j-th column of the block image matrix, where k = 7, 6, 5, 4, 3, 2, 1, 0; p nn (i,j) represents the nth pixel of the i-th row and j-th column of the segmented image block in the segmented image block matrix; mod represents the modulo operation; This indicates the integer rounding operation;
[0072] S32: Randomly combine the 8 bit planes to obtain 4 sets of bit plane combinations as follows:
[0073]
[0074]
[0075]
[0076]
[0077]
[0078] In the formula: Let p represent the first, second, third, and fourth bit-plane combinations of the nth pixel of the i-th row and j-th column of the segmented image block in the segmented image block matrix, respectively; a, q, and t all represent the bit-plane numbers, i.e., a = 0, 1, 2, 3, 4, 5, 6, 7, q = 0, 1, 2, 3, 4, 5, 6, 7, t = 0, 1, 2, 3, 4, 5, 6, 7 and a ≤ q ≤ t; nn (i,j) represents the nth pixel of the block image block located in the i-th row and j-th column of the block image block matrix; wherein, the sum of the combinations of the four sets of bit planes is the nth pixel of the block image block located in the i-th row and j-th column of the block image block matrix, that is, the bit planes are randomly combined according to formulas (2)-(5), and need to satisfy formula (6).
[0079] S33: Randomly split the 4 sets of dimensional plane combinations into maxm addends. as follows:
[0080] In the formula: maxm represents the total number of addends, and m represents the number of the addends; This represents the m-th addend in the r-th bit plane, where m = 1, 2, ..., maxm; add is a positive integer.
[0081] S34: Based on the bit plane combination and the addend, obtain the shared encrypted carrier image as follows, to obtain the shared encrypted carrier image generated from the original carrier image and the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image;
[0082]
[0083] In the formula: share n (i,j) represents the shared encrypted carrier image based on additive secrets; n represents the number of the shared encrypted carrier image;
[0084] Specifically, according to formula (8), the combination is performed in each bit plane. Randomly select an addend without repetition. Combine to generate n shared encrypted carrier images n n = 1, 2, 3... It is important to note that the same bit plane combination will not appear in different shared encrypted carrier images (share). The same addend within. The shared encrypted carrier image generated from the original carrier image is denoted as ori_share. n The shared encrypted carrier image generated by combining the first predicted image containing the actual pixels of the original carrier image, the second predicted image containing the actual pixels of the original carrier image, the third predicted image containing the actual pixels of the original carrier image, and the fourth predicted image containing the actual pixels of the original carrier image is denoted as fork_share. n ,circle_share n triangle_share n square_share n Among them, fork_share n This represents the first shared encrypted carrier image; circle_share n This represents the second shared encrypted carrier image; triangle_share n Indicates a third-party shared encrypted carrier image; square_share n This represents the fourth shared encrypted carrier image.
[0085] S4: Based on the adaptive joint coding method and the shared encrypted carrier image, the shared encrypted carrier image is compressed to free up space for secret information; specifically, compression coding is performed based on the prediction error to free up embedding space; and the resulting encoded and compressed carrier image, i.e., the shared image share_C n The encoded and compressed carrier image is then sent to the data hiding party.
[0086] S41: The shared encrypted carrier image generated from the original carrier image and the shared encrypted carrier image generated from the predicted image containing the actual pixels of the original carrier image are each divided into non-overlapping 2x2 blocks. The pixels in the blocks of the shared encrypted carrier image generated from the original carrier image are designated R, sp1, sp2, sp3 according to the raster scan order; the pixels in the blocks of the shared encrypted carrier image generated from the predicted image containing the actual pixels of the original carrier image are designated R, spp1, spp2, spp3 according to the raster scan order.
[0087] S42: Obtain the maximum error value between the pixels in a block of the shared encrypted carrier image generated from the predicted image containing the true pixels of the original carrier image and the pixels in a block of the shared encrypted carrier image generated from the original carrier image.
[0088] maxe=max(|spp1-sp1|,|spp2-sp2|,|spp3-sp3|) (9)
[0089] In the formula: maxe represents the maximum error value of a pixel; max(·) is the function for finding the maximum value.
[0090] Specifically, in this embodiment, the shared encrypted carrier image fork_share of the predicted image containing the real pixels of the original carrier image is used. n ,circle_share n triangle_share n square_share n The shared encrypted carrier image ori_share, which is the original carrier image and its corresponding source image. n The comparison is performed block by block. The maximum error value within a block is calculated using formula (9). R is the reference pixel value, which is not processed here to ensure the complete reversibility of the image.
[0091] S43: Obtain the highest difference bit of the block between the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image and the shared encrypted carrier image generated from the original carrier image;
[0092]
[0093] In the formula: maxd represents the highest different bit of the block; dif(·) means comparing whether two elements are the same. If they are different, the result is the current value of k plus one. If k is traversed from 7 to 0 and both elements are the same, then 0 is returned. This represents the k-th bit plane of pixel value spp1; This represents the k-th bit plane of pixel value sp1; This represents the k-th bit plane of pixel value spp2; This represents the k-th bit plane of pixel value sp2; This represents the k-th bit plane of pixel value spp3; This represents the k-th bit plane of pixel value sp3;
[0094] Specifically, using method S31, the pixels in the block of the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image are decomposed into 8 bit planes, that is, sp1, sp2, sp3 and spp1, spp2, spp3 are decomposed into bit planes respectively. The highest distinct bit in the block is calculated using formula (10). Where k = 7, 6, 5, 4, 3, 2, 1, 0.
[0095] S44: Based on the highest different bit and the maximum error value of the pixel of the block, obtain the compressed position map of the embeddable and non-embedded blocks;
[0096] Preferably, the method for obtaining the compressed image is as follows:
[0097] When satisfied
[0098] maxe≤7&&(3+3×maxd)>12 (11)
[0099] At that time, execute S441;
[0100] If satisfied
[0101] maxe≤7&&(3+3×maxd)≤12||maxd≤7&&maxe>7 (12)
[0102] When executing, S442 is executed; where && represents logical AND;
[0103] If satisfied
[0104] maxe>7&&maxd>7 (13)
[0105] At that time, execute S443;
[0106] S441: Obtain the first compressed embeddable block, as follows:
[0107] Each of spp1-sp1, spp2-sp2, and spp3-sp3 is represented using 4 binary bits: the first bit (0) represents a positive number, and 1 represents a negative number; the second, third, and fourth bits of spp1-sp1 are represented in binary using the conversion from |spp1-sp1|; the second, third, and fourth bits of spp2-sp2 are represented in binary using the conversion from |spp2-sp2|; and the second, third, and fourth bits of spp3-sp3 are represented in binary using the conversion from |spp3-sp3|.
[0108] Therefore, the four binary bits of the obtained spp1-sp1 are replaced in the high 4 bits of sp1; the four binary bits of the obtained |spp2-sp2| are replaced in the high 4 bits of sp2; and the four binary bits of the obtained |spp3-sp3| are replaced in the high 4 bits of sp3; thus, the first compressed embeddable block corresponding to the block is obtained.
[0109] Specifically, a pixel value can be decomposed into 8 bit planes, or 8 bits of binary data. The high 4 bits, starting from the most significant bit (leftmost bit), are 4 bits. "High bit plane" is a common term. The low 4 bits are 4 bits starting from the right. For example... Figure 10 .
[0110] S442: Obtain the second compressed embeddable block, as follows:
[0111] Represent maxd using 3 binary bits and replace the high three bits of sp1, sp2, and sp3;
[0112] Replace the maxd bit plane starting from the fourth bit of sp1 with the lower maxd bit plane of sp1;
[0113] Replace the maxd bit plane starting from the fourth bit of sp2 with the lower maxd bit plane of sp2;
[0114] Replace the maxd bit plane starting from the fourth bit of sp3 with the lower maxd bit plane of sp3;
[0115] That is, the second compressed embeddable block corresponding to the block is obtained;
[0116] Specifically, maxd is represented by 3 binary bits, and the high three bits of sp1, sp2, and sp3 are replaced. The low maxd bit planes of sp1, sp2, and sp3 are also recorded. The 3-bit representation of maxd and the low maxd bit planes of sp1, sp2, and sp3 are sequentially replaced in the high bits of sp1, sp2, and sp3 to obtain the compressed embeddable block corresponding to the block.
[0117] S443: Generate a Non-Modifiable Block Location Map (NLM) to obtain non-modifiable blocks, as follows:
[0118] When the block satisfies formula (13), it is marked as 1; otherwise, it is marked as 0. This means that the non-intercalable block location map can be obtained, and thus the non-intercalable block can be obtained.
[0119] Specifically, assuming the size of the carrier image is H*W, a matrix of size H / 2*W / 2 is used to record 1 or 0. For example, if the first block satisfies formula (13), then 1 is recorded in the first row and first column of the matrix.
[0120] Specifically, the second compressed embeddable blocks are arranged at the beginning of the image, followed by the first compressed embeddable blocks, and the non-embedded blocks are arranged at the end; thus, the compressed image is obtained.
[0121] S45: Generate a compression method location map (CLM) based on the first compressed insertable block and the second compressed insertable block;
[0122] Specifically, if the current block is compressed using formula (11), it is recorded as 1; if the current block is compressed using formula (12), it is recorded as 0; otherwise, no record is made. In this embodiment, the shared encrypted carrier image generated from the original carrier image and the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image are respectively divided into non-overlapping 2-row, 2-column blocks. Therefore, there are a total of 4 blocks in the figure. The first and second blocks are compressed using formula (11) to obtain the first compressed embeddable block. The third and fourth blocks are compressed using formula (12) to obtain the second compressed embeddable block. The position map is used to record what compression method is used for each block. The encoding of the compressed image is "1100". The method for generating the compression method position map is existing technology in the field and will not be described in detail here.
[0123] S46: Based on the arithmetic coding compression method, the non-interlocking location map (NLM) and the compression method location map (CLM) are compressed into cNLM and cCLM; where cNLM represents the non-interlocking location map (NLM) of arithmetic coding compression; and cCLM represents the compression method location map (CLM) of arithmetic coding compression.
[0124] Specifically, the lengths of cNLM and cCLM are represented by log2(512×512 / 4) bits respectively.
[0125] S47: Based on the compressed embeddable and non-embedded block position map, arrange the second compressed embeddable block, the first compressed embeddable block, and the non-embedded block in sequence to obtain the compressed image; then obtain the number of the compressed image and the space freed up for secret information;
[0126] Specifically, the first two bits of the compressed image encoding represent the set of retained image blocks: 00 represents the set of retained crosses C1, 01 represents the set of retained circles C2, 10 represents the set of retained triangles C3, and 11 represents the set of retained squares C4. The last two bits represent the number n of shared encrypted carrier images generated within the set of image blocks.
[0127] S48: Embed the compressed image's number, length_cNLM, cNLM, length_cCLM, cCLM, and the 32-bit coordinates of the last embeddable block into the space freed up by the secret information, performing position replacement to obtain the encoded and compressed carrier image, i.e., the shared image share_C. n The encoded and compressed carrier image is then sent to the data hiding party.
[0128] S5: Based on the carrier image obtained by the encoding compression, embed secret information; obtain the secret image and send it to the secret receiver to complete the hiding of reversible information.
[0129] S51: Use chaotic mapping to encrypt the secret information to be embedded, and obtain the encrypted secret information;
[0130] S52: The data hiding party receives the encoded and compressed carrier image (i.e., the shared image share_C). n Extract the coordinates of length_cNLM, cNLM, length_cCLM, cCLM and the last embeddable block, and decompress them;
[0131] S53: Based on the Non-Insertive Location Map (NLM) and the Compression Method Location Map (CLM), the length of the secret information and the encrypted secret information are sequentially embedded into the carrier image obtained by the encoding and compression, and the encrypted image is obtained and sent to the secret receiver.
[0132] Specifically, the length of the secret information and the encrypted secret information are embedded into the encoded and compressed carrier image share_C using bit substitution. n In the process, a dense image is obtained and sent to the receiver.
[0133] Preferably, step S5 is followed by:
[0134] S6: Extract secret information from a cryptic image;
[0135] Specifically, the length of the secret information and the encrypted secret information are extracted from the received encrypted image, and then chaotic decryption is performed to obtain the original secret information.
[0136] Specifically, in this embodiment, the content at the corresponding position in the image is found according to the location map. If 1001 is written there, then 1001 is "extracted". The chaotic decryption used is an existing technology, and it is only used here, so it will not be described in detail.
[0137] Preferably, after step S5, the method further includes:
[0138] S7: Based on the dense image, recover the original carrier image;
[0139] S71: Extract the image number, length_cNLM, cNLM, length_cCLM, cCLM, and decompress the non-embedded location map NLM and the compression method location map CLM.
[0140] S72: Based on the Non-Modifiable Block Location Map (NLM) and the Compression Method Location Map (CLM), obtain the original arrangement of blocks in the compressed image;
[0141] S73: Based on the number of the compressed image, add the corresponding pixel values of the images whose first two digits are 00, 01, 10, and 11 respectively to obtain reC′. fork ,reC′ circle ,reC′ triangle ,reC′ square ; where reC′ fork This represents the image obtained by adding the corresponding pixel values of the images whose first two digits are 00, reC′ circle This represents the image obtained by adding the corresponding pixel values of the images whose first two digits are 0 and 1, reC′ triangle This represents the image obtained by adding the corresponding pixel values of the first two images numbered 10, reC′ square This represents the image obtained by adding the corresponding pixel values of the first two images numbered 11;
[0142] S74: reC′ fork ,reC′ circle ,reC′ triangle ,reC′ square The original carrier image can be obtained by merging the sets corresponding to the corresponding image blocks in the image.
[0143] Preferably, after S74, the method further includes:
[0144] S75: If the set corresponding to a segmented image block is lost, extract the set corresponding to the other retained segmented image blocks according to the number of the compressed image, merge them to obtain the merged image, and set the pixel values corresponding to the lost segmented image blocks to 0.
[0145] Based on the merged image and the residual network used for pixel prediction, the prediction results of the set corresponding to the lost image blocks are obtained;
[0146] Specifically, the merged image is used as input to the intelligent predictor (a residual network for pixel prediction) designed in step 2 to obtain the prediction results of the missing parts.
[0147] S76: Based on the prediction results of the set corresponding to the lost image blocks, obtain the shared encrypted carrier image of the set corresponding to the lost image blocks; use S3 to obtain the shared prediction results for the missing parts.
[0148] S77: Obtain the prediction error of the set corresponding to the missing segmented image blocks;
[0149] Based on the compression method designed in steps 5.4 and 5.5, the prediction error of the missing part is extracted and the error is restored to the shared prediction results.
[0150] S78: Obtain the complete carrier image based on the prediction results of the set corresponding to the lost image blocks, the shared encrypted carrier image of the set corresponding to the lost image blocks, and the prediction error of the set corresponding to the lost image blocks.
[0151] Specifically, by adding the prediction results of the set corresponding to the lost image block, the shared encrypted carrier image of the set corresponding to the lost image block, and the prediction error of the set corresponding to the lost image block, we can obtain the original carrier image of the set corresponding to the lost image block. By supplementing it to the lost part, we can restore the original carrier image.
[0152] A specific embodiment of the present invention is as follows:
[0153] like Figure 1 As shown. In this embodiment of the invention, a 512×512 Lena image is used as an example, as follows. Figure 11a As shown. Follow these steps in sequence:
[0154] Step 1: Image Splitting
[0155] Step 1.1 Divide the original carrier image C of size H×W into non-overlapping 4×4 blocks. Number the 16 pixels within each 4×4 block according to the raster scan order as: p1, p2, p3, p4, p5, p6, p7, p8, p9, p 10 ,p 11 ,p 12 ,p 13 ,p 14 ,p 15 ,p 16 .
[0156] In this embodiment, the Lena image is used as an example, with H×W set to 512×512. The image selected from the Lena image is as follows... Figure 2As shown, a 4×4 pixel block with pixel values of 127, 126, 125, 124, 129, 128, 127, 126, 129, 129, 129, 128, 129, 130, 131, and 130 is used as an example for illustration.
[0157] Step 1.2 Next, C is split into 4 sets: the cross set C1, the circle set C2, the triangle set C3, and the square set C4. The cross set C1 contains all p1, p3, p9, and p1 within each block. 11 Pixel values at other locations are set to 0; the triangle set C2 contains all p2, p4, p... 10 ,p 12 Pixel values at other locations are set to 0; the triangle set C3 contains all p5, p7, and p6 pixels within the block. 13 ,p 15 , and the pixel values at other positions are set to 0. The square set C4 contains all p6, p8, and p in the block. 14 ,p 16 Pixel values at other locations are set to 0.
[0158] Specifically, the pixel values of C1 are 127, 0, 125, 0, 0, 0, 0, 0, 129, 0, 129, 0, 0, 0, 0, 0; the pixel values of C2 are 0, 126, 0, 124, 0, 0, 0, 0, 0, 129, 0, 128, 0, 0, 0, 0; the pixel values of C3 are 0, 0, 0, 0, 129, 0, 127, 0, 0, 0, 0, 0, 129, 0, 131, 0; and the pixel values of C4 are 0, 0, 0, 0, 0, 128, 0, 126, 0, 0, 0, 0, 0, 130, 0, 130.
[0159] Step 2: Train the intelligent pixel predictor (a residual network for pixel prediction):
[0160] This embodiment provides a type of residual network for pixel prediction, such as... Figure 3As shown, the network includes an input layer, a feature extraction layer, a prediction layer, and an output layer. The residual network also includes attention residual blocks and residual groups, each containing multiple residual blocks. Each residual block includes two or three convolutional layers and a skip connection. In a residual block with two convolutional layers, the first convolutional layer further extracts features, and the second convolutional layer maps the features to the spatial dimension of the target output. The padding is 2, the stride is 1, and the kernel size is 3×3. A residual block with three convolutional layers, compared to a two-layer residual block, only adds a pointwise convolutional layer with a 1×1 kernel on the skip connection to adjust the input feature dimension. The padding is 0, the stride is 1, and the kernel size is 3×3. Each attention residual block consists of a residual block with two convolutional layers and a channel attention machine module. The channel attention machine module first performs channel-based global max pooling and global average pooling to obtain two H×W×1 feature maps, and then applies a 7×7 convolution operation to reduce the dimension to one channel. The obtained features are used to generate spatial attention features through a sigmoid activation function. Finally, the spatial attention features are multiplied element-wise with the input of this module to obtain the final generated features. Each residual block consists of a residual block containing three convolutional layers and a residual block containing two convolutional layers connected together.
[0161] Step 2.1: Extract feature representations of the input image. A feature extraction module is deployed after the input layer. This module consists of 18 convolutional layers, each with multiple kernels of size 3×3. The first convolutional layer in the feature extraction module has a 1-channel input, a 4-channel output, a stride of 1, and padding of 1. Next, a batch normalization layer and a LeakyReLU activation function are connected. Then, an attention residual block is connected. The obtained features are then fed into a residual block. Finally, the final features are obtained through an attention residual block.
[0162] Step 2.2: Deploy the prediction module. Connect the prediction module after the feature extraction module. The prediction module contains seven residual blocks. The first residual block has 8 input channels and 16 output channels. The second residual block has 16 input channels and 32 output channels. The third residual block has 32 input channels and 64 output channels. The fourth residual block has 64 input channels and 32 output channels. The fifth residual block has 32 input channels and 16 output channels. The sixth residual block has 16 input channels and 8 output channels. The seventh residual block has 8 input channels and 1 output channel.
[0163] Table 1 shows the comparison results of the prediction accuracy of the intelligent predictor in this embodiment with that of related methods on the standard grayscale test image Lena of size 512×512. It can be seen that the prediction performance of this embodiment is better than that of traditional methods and related neural network-based predictors, with more pixels having a prediction error of 0.
[0164] Table 1
[0165]
[0166] Step 3: Generate the predicted image
[0167] Step 3.1: Combine the three sets obtained in Step 1—the cross set C1, the circle set C2, the triangle set C3, and the square set C4—into an input image.
[0168] Step 3.2: Input the data into the intelligent pixel predictor trained in Step 2 to obtain predicted pixels. The combination of C1, C2, and C3 as input yields prediction result C4′; the combination of C2, C3, and C4 as input yields prediction result C1′; the combination of C1, C3, and C4 as input yields prediction result C2′; and the combination of C1, C2, and C3 as input yields prediction result C4′. Combine C1′, C2′, C3′, and C4′ to obtain the predicted image C′ of the original carrier image C.
[0169] Specifically, such as Figure 4 As shown, the combination of cross set C1, circle set C2, and triangle set C3 as input 1 yields the prediction result output 1 for square set C4. The combination of cross set C1, circle set C2, and square set C4 as input 2 yields the prediction result output 2 for triangle set C3. The combination of cross set C1, triangle set C3, and square set C4 as input 3 yields the prediction result output 3 for circle set C2. The combination of circle set C2, triangle set C3, and square set C4 as input 4 yields the prediction result output 4 for cross set C1. Combining outputs 1, 2, 3, and 4 yields the predicted image pixel values of the original carrier image as: 127, 127, 126, 126, 127, 127, 125, 127, 129, 129, 129, 128, 129, 130, 130, 131.
[0170] Step 3.3: Replace the cross set C1, circle set C2, triangle set C3, and square set C4 in the original carrier image C with the corresponding positions in the predicted image C′, as follows: Figure 5 As shown, a carrier image C′ is obtained by retaining the true values of the set of forks C1, the set of circles C2, the set of triangles C3, and the set of squares C4. fork C′ circle C′ triangle C′ square .
[0171] Step 4: Additively secret share the encrypted original carrier image C and the predicted image C′ fork C′ circle C′ triangle C′square .
[0172] Step 4.1: Decompose all pixels of the original carrier image C and the predicted image C′ into 8 bit planes according to formula (1), where p nn (i,j) represents the nth pixel in the block located at (i,j). Let k represent its k-th bit plane, k = 7, 6, 5, 4, 3, 2, 1, 0.
[0173]
[0174] Specifically, such as Figure 6 As shown in Figure 8, Figure 6 Taking the top left pixel 127 as an example, splitting the bit plane yields 01111111.
[0175] Step 4.2: Randomly combine the bit planes according to formula (2-5), and the splitting result must satisfy formula 6, where Let represent the 1st to 4th bit plane combination of the nth pixel in the block located at (i,j), where a = 0, 1, 2, 3, 4, 5, 6, 7, q = 0, 1, 2, 3, 4, 5, 6, 7, t = 0, 1, 2, 3, 4, 5, 6, 7, and a ≤ q ≤ t.
[0176]
[0177]
[0178]
[0179]
[0180]
[0181] Specifically, in this embodiment, 4 is taken as the number of bit-plane groups, such as... Figure 6 As shown in Figure 8, Figure 6 Taking the top left pixel 127 as an example, the bit plane is split into 01111111. Each pair of adjacent bit planes is a group, resulting in 4 groups: 01000000, 110000, 1100, 11.
[0182] Step 4.3: Combine the bit planes according to formula (7) Randomly split into maxm addends Where add is a positive integer, m = 1, 2, ..., maxm;
[0183]
[0184] Specifically, in this embodiment, 3 is taken as the number of addends generated, such as... Figure 7 As shown in 9, Figure 7 Taking the first bit plane group 01000000 of the top left pixel 127 as an example, three addends a1:32, b1:19, and c1:13 are obtained by randomly splitting the data using random factors of 0.5 and 0.3.
[0185] Step 4.4: Combine in each bit plane according to formula (8) Randomly select an addend without repetition. Combine to generate n shared encrypted carriers. n n = 1, 2, 3... It's important to note that the same bit plane combination will not appear in different shares. The same addend within. The shared encrypted carrier generated from the original carrier image is denoted as ori_share. n Predict image C′ fork C c ′ ircle C t ′ riangle C s ′ quare The generated shared encrypted carrier is denoted as fork_share n ,circle_share n triangle_share n square_share n .
[0186]
[0187] Specifically, in this embodiment, 3 is taken as the number of shared elements generated, such as... Figure 7 As shown in 9, Figure 7 Taking the first bit plane group 01000000 with pixel 127 in the upper left corner as an example, the corresponding position pixel values in the three shared elements are 63, 36, and 28.
[0188] Step 5: Use an adaptive joint coding method to free up embedding space for the secret information;
[0189] Step 5.1: Divide all the shared images obtained in Step 4.4 into non-overlapping 2×2 blocks, ori_share n The pixels in each block are denoted as R, sp1, sp2, sp3, fork_share in the order of raster scan. n ,circle_share n triangle_share n square_share nThe pixels in each block are denoted as R, spp1, spp2, spp3 in the order of raster scanning.
[0190] Step 5.2: Share the encrypted carrier fork_share n ,circle_share n triangle_share n square_share n Each of the original and corresponding encrypted carrier images shares the same ori_share. n The comparison is performed by block. The maximum error value within a block is calculated using formula (9), where max(·) is the function for finding the maximum value.
[0191] maxe=max(|spp1-sp1|,|spp2-sp2|,|spp3-sp3|) (9)
[0192] Step 5.3: Decompose sp1, sp2, sp3 and spp1, spp2, spp3 into bit planes using Step 4.1. Use formula (10) to calculate the highest distinct bit within the block. Where k = 7, 6, 5, 4, 3, 2, 1, 0, dif(·) compares whether two elements are the same. If they are different, the result is the current value of k plus one. If k iterates from 7 to 0 and both elements are the same, then 0 is returned.
[0193]
[0194] Step 5.4: If the current block satisfies formula (11), then represent spp1-sp1, spp2-sp2, and spp3-sp3 using 4 bits, where the first bit 0 represents a positive number and 1 represents a negative number. Convert the other three bits |spp1-sp1|, |spp2-sp2|, and |spp3-sp3| to binary representation. Replace the obtained 4 bits in the high 4 bits of sp1, sp2, and sp3 respectively.
[0195] maxe≤7&&(3+3×maxd)>12 (11)
[0196] Specifically, such as Figure 10As shown, taking the 2×2 block on the left as an example, the black mark in the upper left corner is the true value of the cross set C1. The prediction errors of the other three pixels are 63-63=0, 63-64=-1, 63-64=-1, max(|0|,|-1|,|-1|)=1. 63 splits the bit plane to get 00111111, and 64 splits the bit plane to get 01000000. 63 and 64 are compared bit by bit using formula (10). When k=6, it is the first different bit, and the return value is k+1=7. 63 and 64 are compared bit by bit using formula (10). k traverses from 7 to 0. There are no different bits, and the return value is 0. maxd=max(7,0)=0. Formula (11) is used to determine the position where the secret information can be added to the image.
[0197] Step 5.5: If the current block satisfies formula (12), then represent maxd using three binary bits and record the lower maxd bit planes of sp1, sp2, and sp3. Replace the three binary representations of maxd with the lower maxd bit planes of sp1, sp2, and sp3 in sp1, sp2, and sp3 in sequence.
[0198] maxe≤7&&(3+3×maxd)≤12||maxd≤7&&maxe>7 (12)
[0199] Specifically, such as Figure 10 As shown, taking the 2×2 block on the right as an example, the black mark in the upper left corner is the true value of the cross set C1. The prediction errors of the other three pixels are 49-48=1, 49-49=0, 49-49=0, max(|1|,|0|,|0|)=1. 49 is split into bit planes to get 00110001, and 48 is split into bit planes to get 00110000. 49 and 48 are compared bit by bit using formula (10). When k=0, it is the first different bit, and the return value is k+1=1. 49 and 49 are compared bit by bit using formula (10). k traverses from 7 to 0. There are no different bits, and the return value is 0. maxd=max(1,0)=1. Using formula (11) to judge, 12<(3+3×1), so step 5.5 is used for compression. The block gets 18 bits freed up.
[0200] Step 5.6: Generate a non-embedded block location graph (NLM). If the current block satisfies formula (13), it is recorded as 1; otherwise, it is recorded as 0.
[0201] maxe>7&&maxd>7 (13)
[0202] Step 5.7 Generate the compression method location map (CLM). If the current block is compressed using step 5.4, it is recorded as 1; if the current block is compressed using step 5.5, it is recorded as 0. Otherwise, no record is made.
[0203] Step 5.8: Compress NLM and CLM into cNLM and cCLM using arithmetic coding compression, and express the length of cNLM as length_cNLM and the length of cCLM as length_cCLM using log2(512×512 / 4) bits respectively.
[0204] Step 5.9: Arrange all the compressed embeddable blocks obtained in Step 5.5 at the beginning of the image, then arrange all the compressed embeddable blocks obtained in Step 5.4 next, and arrange the non-embedded blocks at the end.
[0205] Step 5.10: Number the obtained compressed image. The first two digits represent the set to be retained, followed by the number n of shared elements generated within the set. 00 represents retaining the set of crosses C1, 01 represents retaining the set of circles C2, 10 represents retaining the set of triangles C3, and 11 represents retaining the set of squares C4.
[0206] Step 5.11: Embed the compressed image number, length_cNLM, cNLM, length_cCLM, cCLM, and the 32-bit coordinates of the last embeddable block into the freed-up space. This yields the final encrypted carrier image, as shown below. Figure 11b As shown, these are three shared images generated from the carrier image of the retained fork set C1.
[0207] Step 5.12: Distribute the shared image.
[0208] Step 6: Embed secret information:
[0209] Step 6.1: Encrypt the secret information to be embedded using a chaotic map;
[0210] Step 6.2: The hiding party receives the compressed carrier image, extracts the coordinates of length_cNLM, cNLM, length_cCLM, cCLM and the last embeddable block, and decompresses it.
[0211] Step 6.3: Using the location graph NLM and CLM, embed the length of the embedded secret information and the encrypted secret information into the carrier image share_C. k The process involves obtaining a dense image and sending it to the receiver. For example... Figure 11c The image shown is a 3-fold image of the fork set C1.
[0212] Table 3 shows the comparison results of the embedding capacity of the present invention with that of related methods. Overall, the average embedding capacity of the present invention is 3.11 bpp, which is higher than that of other methods.
[0213] Table 3
[0214]
[0215] Step 7: Extract secret information:
[0216] Step 7.1: Extract the length embedding and secret information of the embedded secret information from the received encrypted image, and perform chaotic decryption to obtain the original secret information.
[0217] Step 8: Restore the original carrier image;
[0218] Step 8.1: Extract the image numbers, length_cNLM, cNLM, length_cCLM, and cCLM, and decompress the two position images.
[0219] Step 8.2: Use the location maps NLM and CLM to restore the original arrangement of the blocks.
[0220] Step 8.3: Using the image number, add the corresponding pixel values of the images whose first two digits are 00, 01, 10, and 11 respectively to obtain reC′. fork ,reC′ circle ,reC′ triangle ,reC′ square .
[0221] Step 8.4: Add reC′ fork ,reC′ circle ,reC′ triangle ,reC′ square By merging the corresponding retained sets, the original carrier image can be obtained.
[0222] Step 8.5: If a retained set is missing, extract the other three retained sets based on the image number, merge them into one image, and set the pixel values of the missing set to 0. Use the merged image as input to the intelligent predictor designed in Step 2 to obtain the prediction result for the missing portion.
[0223] Step 8.6: Share the prediction results obtained in Step 4 for the missing parts.
[0224] Step 8.7: Based on the compression method designed in steps 5.4 and 5.5, extract the prediction error of the missing part and restore the error to the shared prediction results.
[0225] Step 8.8: Add the restored shared values to obtain the original value of the lost set, and supplement it into the image obtained in Step 8.5. At this point, the carrier image is lostlessly restored. Figure 11d As shown. It is important to note that since data embedding and image encryption are independent, data extraction and image reconstruction are also separable.
[0226] The reversible information hiding method in the ciphertext domain of this embodiment utilizes an intelligent predictor to efficiently and accurately predict pixel values, thereby improving the embedding capacity. Simultaneously, it employs additive secret sharing of the encrypted carrier image, providing security and resistance to collusion analysis. Furthermore, this invention proposes a joint compression coding strategy that fully utilizes the bit-plane distribution characteristics, using appropriate compression methods for pixel blocks with different characteristics, while also exhibiting fault tolerance for the loss of sets corresponding to segmented image blocks. At the receiving end, this embodiment can losslessly extract the secret information and recover the original image. It possesses advantages such as high prediction accuracy, strong embedding capability, complete reversibility, lossless extraction, and separability.
[0227] The intelligent pixel predictor proposed in this embodiment improves the accuracy and robustness of embedding and extracting hidden information by leveraging the powerful representational capabilities of deep learning and neural networks. It also proposes an adaptive compression coding strategy to fully utilize the characteristics of bit-plane distribution, and an image encryption based on additive secret sharing to achieve multi-party sharing of encrypted images. This approach offers advantages such as strong security, high embedding capacity, high image quality, and reversibility. It can improve the capacity of existing technologies while ensuring image information security and achieving lower computational complexity, providing a reversible information hiding method in the ciphertext domain based on intelligent predictors and additive secret sharing. This method can be widely applied in fields such as image steganography, copyright protection, and secure communication, providing a new solution for the development of information hiding technology. The combination of reversible information hiding in the ciphertext domain and neural network models in this embodiment has broad application prospects and can play an important role in protecting privacy and information security, such as in secure communication, encrypted storage, and digital rights management.
[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain, characterized in that, Includes the following steps: S1: Preprocess the original carrier image to obtain a preprocessed image that retains some of the original carrier image information; S2: Establish a residual network for pixel prediction to obtain a predicted image containing the real pixels of the original carrier image based on the preprocessed image that retains part of the original carrier image information; The residual network for pixel prediction in S2 includes an input layer, a feature extraction layer, a prediction layer and an output layer, an attention residual block and a residual block; The input layer is used to input the preprocessed image that retains some of the original carrier image information; The input end of the feature extraction layer is connected to the output end of the input layer, and is used to extract the feature representation of the fixed real pixel part in the preprocessed image after retaining some original carrier image information; The feature extraction layer includes multiple convolutional layers, batch normalization layers, LeakyReLU activation function, and attention residual blocks; Multiple convolutional layers are connected sequentially, with the first convolutional layer connected to the output of the input layer; the last convolutional layer is connected to a batch normalization layer, which is activated by the LeakyReLU activation function to obtain local features of the real pixel portion of the preprocessed image; the attention residual block is connected to the LeakyReLU activation function to obtain a feature representation of the real pixel portion of the preprocessed image, retaining some original carrier image information, based on the local features of the real pixel portion of the preprocessed image. The input of the prediction layer is connected to the output of the feature extraction layer, and is used to obtain a predicted image containing the real pixels of the original carrier image based on the feature representation of the real pixel part in the preprocessed image that retains some original carrier image information. The prediction layer comprises seven residual blocks. Assuming the number of input channels for the first residual block is x, the number of output channels for the first residual block is 2x. The number of input channels for the second, third, fourth, fifth, sixth, and seventh residual blocks are 2x, 3x, 4x, 3x, 2x, and x, respectively. The number of output channels for the second, third, fourth, fifth, sixth, and seventh residual blocks are 3x, 4x, 3x, 2x, x, and x, respectively. The input end of the output layer is connected to the output end of the prediction layer, and is used to output a predicted image containing the real pixels of the original carrier image; S3: Using an additive secret sharing algorithm, the original carrier image and the predicted image containing the real pixels of the original carrier image are encrypted to obtain a shared encrypted carrier image; S4: Based on the adaptive joint coding method and the shared encrypted carrier image, the shared encrypted carrier image is compressed to free up space for secret information, so as to obtain the carrier image obtained by encoding and compression; S5: Based on the carrier image obtained by the encoding and compression, acquire the encrypted image and send it to the secret receiver to complete the hiding of reversible information.
2. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 1, characterized in that, In step S3, the method for obtaining the shared encrypted carrier image is as follows: S31: Decompose all pixels in the original carrier image and the predicted image containing the real pixels of the original carrier image into 8 bit planes as follows: (1) In the formula: This represents the i-th row and j-th column of the segmented image block in the segmented image block matrix. k Units plane k =7,6,5,4,3,2,1,0; This represents the i-th row and j-th column of the segmented image block located in the segmented image block matrix. nn 1 pixel; This indicates the modulo operation; This indicates the integer rounding operation; S32: Randomly combine the 8 bit planes to obtain 4 sets of bit plane combinations as follows: (2) (3) (4) (5) (6) In the formula: , , , These represent the i-th row and j-th column of the segmented image block in the segmented image block matrix. nn The first group of bit plane combinations, the second group of bit plane combinations, the third group of bit plane combinations, and the fourth group of bit plane combinations for each pixel; a , q , t Both represent the bit plane number, and ; This represents the i-th row and j-th column of the segmented image block located in the segmented image block matrix. nn 1 pixel; S33: Randomly split the 4 sets of lattice combinations into maxm Addends as follows: (7) In the formula: maxm This represents the total number of addends, where m represents the addend number. This represents the m-th addend in the r-th bit plane. m =1,2... maxm ; add It is a positive integer; S34: Based on the bit plane combination and the addend, obtain the shared encrypted carrier image, so as to obtain the shared encrypted carrier image generated from the original carrier image and the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image; (8) In the formula: This represents a shared encrypted carrier image based on additive secrets; n represents the number of the shared encrypted carrier image.
3. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 1, characterized in that, In step S1, the method for preprocessing the original image is as follows: S11: Divide the original carrier image into a non-overlapping 4x4 matrix of image blocks, and number each image block in the matrix according to the raster scan order. p 1, p 2, p 3, p 4, p 5 , p 6, p 7, p 8, p 9, p 10 , p 11 , p 12 , p 13 , p 14 , p 15 , p 16 ; S12: The numbered... p 1, p 3, p 9, p 11 The segmented image blocks are used as a fork set; Number p 2, p 4, p 10 , p 12 The segmented image blocks are used as a set of circles; Number p 5 ,p 7, p 13 , p 15 The segmented image blocks are used as a triangular set; Number p 6 ,p 8, p 14 , p 16 The segmented image blocks are treated as a set of squares; S13: Set the pixel values of all block image blocks in the original carrier image except for the fork set to 0 to form the first preprocessed image; Set the pixel values of all image blocks in the original carrier image except for the circle set to 0 to form the second preprocessed image; Set the pixel values of all image blocks in the original carrier image except for the triangular set to 0 to form the third preprocessed image; The pixel values of all image blocks in the original carrier image, except for the square set, are set to 0 to form the fourth preprocessed image.
4. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 3, characterized in that, In step S2, the method for obtaining the predicted image containing the actual pixels of the original carrier image is as follows: S21: Combine the second preprocessed image, the third preprocessed image, and the fourth preprocessed image to form the first input image; The first preprocessed image, the third preprocessed image, and the fourth preprocessed image are combined to form the second input image; The first preprocessed image, the second preprocessed image, and the fourth preprocessed image are combined to form the third input image; The first preprocessed image, the second preprocessed image, and the third preprocessed image are combined to form the fourth input image; S22: Input the first input image, the second input image, the third input image, and the fourth input image into the residual network for pixel prediction to obtain the first prediction image, the second prediction image, the third prediction image, and the fourth prediction image, respectively; and combine the first prediction image, the second prediction image, the third prediction image, and the fourth prediction image to obtain the prediction image of the original carrier image; S23: Replace the image block at the position corresponding to the block image block in the fork set in the predicted image of the original carrier image with the block image block in the fork set to obtain the first predicted image containing the real pixels of the original carrier image. The image block at the position corresponding to the block image block in the circle set in the predicted image of the original carrier image is replaced with the block image block in the circle set to obtain a second predicted image containing the real pixels of the original carrier image. The image block at the position corresponding to the block image block in the triangle set in the predicted image of the original carrier image is replaced with the block image block in the triangle set to obtain a third predicted image containing the real pixels of the original carrier image. The image block at the position corresponding to the block image block in the square set in the predicted image of the original carrier image is replaced with the block image block in the square set to obtain a fourth predicted image containing the real pixels of the original carrier image.
5. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 1, characterized in that, In step S4, the method for obtaining the space to free up secret information and the carrier image obtained by encoding and compression is as follows: S41: The shared encrypted carrier image generated from the original carrier image and the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image are respectively split into non-overlapping parts. The blocks, and the pixels in the blocks of the shared encrypted carrier image generated from the original carrier image, are respectively denoted as follows according to the raster scan order: R , sp 1, sp 2, sp 3; The pixels in the blocks of the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image are respectively denoted as follows according to the raster scan order: R , spp 1, spp 2, spp 3; among which, S42: Obtain the maximum error value between the pixels in a block of the shared encrypted carrier image generated from the predicted image containing the true pixels of the original carrier image and the pixels in a block of the shared encrypted carrier image generated from the original carrier image. (9) In the formula: This represents the maximum error value of a pixel; max(·) is the function to find the maximum value. S43: Obtain the highest difference bit of the block between the shared encrypted carrier image generated from the predicted image containing the real pixels of the original carrier image and the shared encrypted carrier image generated from the original carrier image; (10) In the formula: The `dif` operator indicates the highest distinct bit of the block; `dif(·)` compares whether two elements are the same. Represents pixel value spp 1 of k Units plane; Represents pixel value sp 1 of k Units plane; Represents pixel value spp 2nd k Units plane; Represents pixel value sp 2nd k Units plane; Represents pixel value spp 3rd k Units plane; Represents pixel value sp 3rd k Units plane; S44: Based on the highest different bit and the maximum error value of the pixel of the block, obtain the compressed position map of the embeddable and non-embedded blocks; The method for obtaining the compressed location map of inlayable and non-inlayable blocks is as follows: When satisfied (11) At that time, execute S441 to obtain the first compressed embeddable block; If satisfied (12) At that time, execute S442 to obtain the second compressed embeddable block; where: This indicates a logical AND operation. If satisfied (13) At that time, execute S443 to obtain the non-intercalable block location map; S441: Obtain the first compressed embeddable block, as follows: Will All numbers are represented using 4 binary bits: the first bit (0) represents a positive number, and the first bit (1) represents a negative number. The second, third, and fourth bits of the four binary digits are used The converted binary representation; The second, third, and fourth bits of the four binary digits are used The converted binary representation; The second, third, and fourth bits of the four binary digits are used The converted binary representation; Therefore, the obtained The four binary bits are replaced to sp In the high 4 bit plane of 1; the obtained The four binary bits are replaced to sp In the high 4 bit plane of 2; the obtained The four binary bits are replaced to sp In the 3-high 4-bit bit plane, the first compressed embeddable block corresponding to the block is obtained; S442: Obtain the second compressed embeddable block, as follows: Will maxd It is represented using 3 binary bits, and replaced sp 1, sp 2, sp The three highest digits of 3; Will sp The maxd bit plane starting from the fourth bit of 1 uses sp 1 low maxd Replace the bit plane; Will sp The maxd bit plane starting from the fourth bit of 2 uses sp 2 low maxd Replace the bit plane; Will sp The maxd bit plane starting from the fourth bit of 3 uses sp 3 low maxd Replace the bit plane; That is, the second compressed embeddable block corresponding to the block is obtained; S443: Generate a Non-Modifiable Block Location Map (NLM) using the following method: When the block satisfies formula (13), it is marked as 1; otherwise, it is marked as 0, thus obtaining the non-embedded block location map. S45: Generate a compression method location map based on the first compressed insertable block and the second compressed insertable block. CLM ; S46: Compression of non-modular location maps based on arithmetic coding methods NLM Location diagram of compression method CLM Compress to cNLM and cCLM ;in, cNLM Non-block location map representing arithmetic coding compression NLM ; cCLM Location diagram of compression methods for arithmetic coding compression CLM ; S47: Based on the compressed embeddable and non-embedded block position map, arrange the second compressed embeddable block, the first compressed embeddable block, and the non-embedded block in sequence to obtain the compressed image; then obtain the number of the compressed image and the space freed up for secret information; S48: Number the compressed image. length_cNLM , cNLM , length_cCLM , cCLM The coordinates of the last embeddable block are embedded into the space freed up by the secret information for position replacement, resulting in an encoded and compressed carrier image; this encoded and compressed carrier image is then sent to the data hiding party; wherein... length_cNLM express cNLM Length; length_cCLM express cCLM The length.
6. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 5, characterized in that, The method for obtaining dense images is as follows: S51: Use chaotic mapping to encrypt the secret information to be embedded, and obtain the encrypted secret information; S52: The data hiding party receives the encoded and compressed carrier image and extracts the encoded and compressed carrier image. length_cNLM , cNLM , length_cCLM , cCLM Given the coordinates of the last embeddable block, obtain the decompressed image; S53: According to the non-intercalable block location map NLM Location diagram of compression method CLM The length of the secret information and the encrypted secret information are sequentially embedded into the carrier image obtained by encoding and compression, and the encrypted image is obtained and sent to the secret recipient.
7. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 6, characterized in that, After step S5, Includes: S6: Extracting secret information from a dense image, using the following method: Extract the length of the secret information and the encrypted secret information from the received encrypted image, and perform chaotic decryption to obtain the original secret information.
8. The method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 7, characterized in that, Following step S5, the method further includes: S7: Based on the dense image, recover the original carrier image; the method is as follows: S71: Extract the number of the compressed image. length_cNLM , cNLM , length_cCLM , cCLM And decompress the non-embedded block location map NLM Location diagram of compression method CLM ; S72: According to the non-intercalable block location map NLM Location diagram of compression method CLM, to obtain The original arrangement of blocks in the compressed image; S73: Based on the number of the compressed image, add the corresponding pixel values of the images whose first two digits are 00, 01, 10, and 11 respectively to obtain... , , , ;in, This represents the image obtained by adding the corresponding pixel values of the images whose first two digits are 00. This represents the image obtained by adding the corresponding pixel values of the images whose first two digits are 01. This represents the image obtained by adding the corresponding pixel values of the first two images numbered 10. This represents the image obtained by adding the corresponding pixel values of the first two images numbered 11; S74: Will , , , The original carrier image can be obtained by merging the sets corresponding to the corresponding image blocks in the image.
9. A method for reversible information hiding based on an intelligent predictor and additive secret shared ciphertext domain according to claim 8, characterized in that, Following S74, it also includes: S75: If the set corresponding to a segmented image block is lost, extract the set corresponding to the other retained segmented image blocks according to the number of the compressed image, merge them to obtain the merged image, and set the pixel values corresponding to the lost segmented image blocks to 0. Based on the merged image and the residual network used for pixel prediction, the prediction results of the set corresponding to the lost image blocks are obtained; S76: Based on the prediction results of the set corresponding to the lost image blocks, obtain the shared encrypted carrier image of the set corresponding to the lost image blocks; S77: Obtain the prediction error of the set corresponding to the missing segmented image blocks; S78: Obtain the complete carrier image based on the prediction results of the set corresponding to the lost image blocks, the shared encrypted carrier image of the set corresponding to the lost image blocks, and the prediction error of the set corresponding to the lost image blocks.
