A robust steganography method and device for privacy image protection

Abstract

The application discloses a privacy image protection-oriented robust steganography method and device, and belongs to the technical field of information security. The method comprises the following steps: scaling a cover image to obtain a carrier image capable of carrying private image information; performing frequency domain transformation on the carrier image to construct a frequency domain embedding domain; performing bit plane splitting processing on a private image to be protected to obtain private bit planes; designing a nested quantization index modulation algorithm to embed the private bit planes into the frequency domain embedding domain to generate a stego image. The method can not only realize complete lossless reconstruction of the private image in a lossless environment, but also reconstruct the private image with recognizable semantic information in a lossy environment, thereby effectively preventing sensitive images from leaking the original appearance in the process of transmission, storage and processing, and improving the security of the private image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security, and in particular to a robust steganography method and device for privacy image protection. BACKGROUND

[0002] The privacy image protection technology refers to a technology of securely hiding, encrypting or concealing the image content, so that the sensitive information of the image is not leaked in the process of transmission, storage and use.

[0003] The image steganography technology is a common privacy image protection method, and the basic idea is to embed the privacy image into a carrier image to form a stego image, so as to realize the information hiding transmission, and the original privacy image can be reconstructed after processing the stego image at the receiving end. However, the existing privacy image lossless reconstruction steganography technology generally lacks robustness. Once the stego image undergoes lossy operation in the actual application scenario, such as image compression and noise interference due to limited transmission bandwidth, the embedded privacy image information will be damaged, and the semantically recognizable privacy image cannot be effectively reconstructed. That is to say, the existing method can only ensure the complete recovery of the privacy image in a strict lossless environment, and it is difficult to maintain the recognizability and usability of the privacy image in the presence of compression or interference.

[0004] Therefore, there is an urgent need for a privacy image protection method and device that can have both lossless reconstruction capability and robust recovery capability, so as to realize lossless reconstruction of the privacy image in a lossless environment, and still be able to reconstruct a privacy image with semantically recognizable in a lossy environment, thereby improving the security and applicability of the privacy image. SUMMARY

[0005] The present application aims to provide a robust steganography method and device for privacy image protection, to solve the technical problem that the existing image steganography technology cannot simultaneously have lossless recovery and robust recovery. The present application improves the usability and recovery quality of the privacy image in a lossy environment while realizing lossless recovery of the privacy image, thereby enhancing the security and stability of the privacy image in the process of transmission, storage and processing.

[0006] To achieve the above-mentioned purpose, the present application provides a robust steganography method for privacy image protection, comprising the following steps:

[0007] In the first aspect, the present application provides a robust steganography method for privacy image protection, comprising:

[0008] scaling the cover image to obtain a carrier image capable of carrying privacy image information;

[0009] performing frequency domain transformation on the carrier image to construct a frequency domain embedding domain;

[0010] The privacy bit plane is obtained by performing bit plane splitting on the privacy image to be protected;

[0011] The privacy bit plane is embedded into the frequency domain embedding domain to obtain a dense image;

[0012] Perform the inverse process on the encrypted image to reconstruct the privacy image to be protected.

[0013] Preferably, scaling the cover image to obtain a carrier image capable of carrying privacy image information includes:

[0014] The resolution of the privacy image is determined, and an interpolation algorithm is used to scale the cover image to four times that of the privacy image to obtain the carrier image;

[0015] The interpolation algorithm employs any one of the following methods: nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation.

[0016] Preferably, the step of performing frequency domain transformation on the carrier image to construct a frequency domain embedding domain includes:

[0017] The carrier image is segmented into blocks to obtain... Image blocks;

[0018] Perform a discrete cosine transform on each image patch to obtain the corresponding frequency domain coefficient matrix. , ;

[0019] The frequency domain coefficient matrix of all image blocks is reorganized to obtain the frequency domain embedding domain.

[0020] Preferably, the reorganization of the frequency domain coefficient matrix of all image blocks to obtain the frequency domain embedding domain includes:

[0021] frequency domain coefficient matrix The frequency domain coefficients in the figure are expanded into a one-dimensional sequence according to Zigzag order, as follows:

[0022] ;

[0023] For the same position in the one-dimensional sequence of all frequency domain coefficient matrices The frequency domain coefficients at each point are aggregated to form the first... The dimensional aggregation coefficient vector is represented as:

[0024] , ,

[0025] in, Represents the frequency domain coefficient matrix Position of the unfolded one-dimensional sequence Elements at the location;

[0026] Finally, the complete frequency domain embedding domain is obtained .

[0027] Preferably, the privacy image to be protected is subjected to bit plane splitting processing to obtain privacy bit planes, including:

[0028] The binary bits of each pixel in the privacy image to be protected are extracted and aggregated bit by bit to obtain 8 privacy bit planes corresponding to the bit positions Each privacy bit plane is composed of the bits of all pixels of the privacy image to be protected at the same bit position.

[0029] Preferably, the privacy bit planes are embedded in the frequency domain embedding domain to obtain a stego image, including:

[0030] The bit information in the privacy bit plane is embedded in the frequency domain embedding domain using a nested quantization index modulation algorithm, which is divided into two layers of quantization index modulation processes:

[0031] The first layer of quantization index modulation embeds the first four privacy bit planes in the frequency domain embedding domain , and the embedding process is:

[0032] The first four privacy bit planes are flattened into one dimension, denoted as ; the frequency domain embedding domain is flattened into one dimension, denoted as ,

[0033] modulates and embeds in with a quantization step size , denoted as:

[0034] ,

[0035] where is the quantization center, is the th coefficient in the sequence , and is the th bit in the sequence ; when is 0, is quantized to an even multiple of the quantization step size ; when is 1, is quantized to an odd multiple of the quantization step size ;

[0036] The second layer of quantization index modulation embeds the last four privacy bit planes Embedding into the shifted quantization centers The embedding process is as follows:

[0037] The last 4 privacy bit planes are flattened into one dimension, denoted as ;

[0038] Embedding is done as follows:

[0039] ,

[0040] ,

[0041] where, , is the shift coefficient applied to the second layer, is the th coefficient in after embedding and , is the th bit in the sequence ;

[0042] The stego frequency domain coefficients are recombined into a frequency domain matrix and converted to the spatial domain to obtain the stego image.

[0043] Preferably, an inverse process is performed on the stego image to reconstruct the privacy image to be protected, including:

[0044] Performing block, discrete cosine transform and frequency domain coefficient recombination on the stego image to obtain a stego frequency domain embedding domain , and extracting two layers of bit information from the stego frequency domain embedding domain in turn using an inverse nested quantization index modulation algorithm, which is divided into two layers of inverse quantization index modulation processes:

[0045] The first layer of inverse quantization index modulation extracts the embedded first 4 privacy bit plane one-dimensional sequence from the stego frequency domain embedding domain , and the extraction process is represented as:

[0046] ,

[0047] ,

[0048] where, is the th coefficient of the stego frequency domain embedding domain , denotes the nearest even quantization center to , Indicates distance The most recent odd quantization center;

[0049] like Then determine the first layer of bits. If it is 0, otherwise determine the first layer bit. =1, One-dimensional sequence The Bit;

[0050] The second-layer inverse quantization index modulates the one-dimensional sequence extracted from the first-layer inverse quantization index. Building upon this foundation, we continue to embed the domain from the dense frequency domain. Extract one-dimensional sequence The extraction process is represented as follows:

[0051] ,

[0052] when Determine the second layer of bits =1; when Determine the second layer of bits The value is set to 0. One-dimensional sequence The Bit;

[0053] One-dimensional sequence and Converted to 8 reconstructed bit planes;

[0054] The eight reconstructed bit planes are reconstructed into pixel grayscale values ​​to generate a reconstructed privacy image.

[0055] Secondly, the present invention provides a robust steganography apparatus for privacy image protection, used to implement the aforementioned robust steganography method for privacy image protection, the apparatus comprising:

[0056] The carrier image construction module is used to scale the cover image to obtain a carrier image capable of carrying privacy image information;

[0057] The frequency domain processing module is used to perform frequency domain transformation on the carrier image and construct a frequency domain embedding domain;

[0058] The privacy image splitting module is used to perform bit-plane splitting processing on the privacy image to be protected, so as to obtain the privacy bit plane;

[0059] The steganography embedding module is used to embed the privacy bit plane into the frequency domain embedding domain to obtain a steganographic image;

[0060] a privacy image reconstruction module configured to perform an inverse process on the stego image to reconstruct the privacy image to be protected.

[0061] In a third aspect, a computer-readable storage medium storing one or more programs for execution by a computing device is provided, the one or more programs comprising instructions configured to cause the computing device to perform any of the robust steganography methods for privacy image protection described above.

[0062] In a fourth aspect, a computing device is provided, comprising one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising instructions for performing any of the robust steganography methods for privacy image protection described above.

[0063] Compared with the prior art, the present application has the following beneficial effects:

[0064] The present application provides a robust steganography method for privacy image protection, which constructs a steganography embedding domain in the frequency domain and uses nested quantization index modulation for privacy image embedding, so that the privacy image can be completely lossless reconstructed in a lossless environment, and still can recover a privacy image with semantically recognizable in a lossy environment such as compression and noise interference, effectively solving the technical problem that the prior art cannot balance losslessness and robustness. The present application significantly enhances the security and anti-distortion ability of the privacy image under the premise of maintaining the visual quality of the stego image, and is suitable for various privacy image protection scenes such as social media, medical images, and judicial evidence, and has high practical value and application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 is a robust steganography method for privacy image protection provided by the embodiment of the present application;

[0066] Figure 2 is a privacy image schematic diagram selected in the embodiment of the present application;

[0067] Figure 3 is a cover image schematic diagram obtained in the embodiment of the present application;

[0068] Figure 4 is a carrier image schematic diagram obtained in the embodiment of the present application;

[0069] Figure 5 is a frequency domain coefficient matrix schematic diagram obtained in the embodiment of the present application;

[0070] Figure 6 is a stego image schematic diagram obtained in the embodiment of the present application;

[0071] Figure 7 is a schematic diagram of a dense frequency domain embedding domain obtained in an embodiment of the present application;

[0072] Figure 8 is a schematic diagram of a dense image in performance analysis of an embodiment of the present application;

[0073] Figure 9 is a schematic diagram of a reconstructed privacy image in performance analysis of an embodiment of the present application. DETAILED DESCRIPTION

[0074] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail below with reference to embodiments and drawings. Here, the illustrative embodiments of the present application and their descriptions are used to explain the present application, but are not intended to limit the present application.

[0075] It should also be noted that, in order to avoid obscuring the present application due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present application are shown in the drawings, and other details not closely related to the present application are omitted.

[0076] It should be emphasized that the term "comprises / comprising" as used herein means the presence of a stated feature, element, step or component, but does not preclude the presence or addition of one or more other features, elements, steps or components.

[0077] It should also be noted that, unless otherwise specified, the term "connected" as used herein can not only mean direct connection, but also indirect connection in the presence of an intermediate.

[0078] In the following, embodiments of the present application will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts or the same or similar steps.

[0079] It should be emphasized here that the step labels mentioned in the following are not a limitation on the order of the steps, but it should be understood that the steps can be performed in the order mentioned in the embodiments, or in an order different from the embodiments, or several steps can be performed simultaneously.

[0080] An embodiment of the present application provides a robust steganography method for privacy image protection, referring to Figure 1 , comprising the following steps:

[0081] S1, scaling the cover image to obtain a carrier image capable of carrying privacy image information;

[0082] S2, performing frequency domain transformation on the carrier image to construct a frequency domain embedding domain;

[0083] S3, performing bit plane splitting processing on the to-be-protected privacy image to obtain a plurality of privacy bit planes;

[0084] S4, embedding the plurality of privacy bit planes into the frequency domain embedding domain constructed above by using a nested quantization index modulation algorithm to obtain a stego image;

[0085] S5, performing an inverse process based on the stego image to reconstruct the privacy image.

[0086] Based on the above technical solutions, the privacy image can be losslessly reconstructed in a lossless environment, and the privacy image with semantic recognizable can still be reconstructed in a lossy environment.

[0087] In step S1 of the embodiment, the cover image is scaled to obtain a carrier image capable of carrying privacy image information, specifically including:

[0088] In order to ensure that the carrier image has enough space to carry the complete privacy image, the resolution of the cover image is enlarged to four times of the privacy image by using an interpolation algorithm to obtain the carrier image. If the resolution of the cover image is , the resolution of the carrier image needs to be scaled to . and are the height and width of the cover image respectively, and are the height and width of the carrier image respectively.

[0089] The classical interpolation algorithms mainly include nearest neighbor interpolation, bilinear interpolation and cubic convolution interpolation. Here, the cover image is denoted as , and the carrier image is denoted as . The calculation formula for scaling the cover image is:

[0090] ,

[0091] In the formula, the parameter may be , or , which respectively represent scaling the cover image at a fixed resolution by using the nearest neighbor interpolation, bilinear interpolation or cubic convolution interpolation.

[0092] In step S2 of the embodiment, the carrier image is subjected to frequency domain transformation to construct a frequency domain embedding domain, specifically including:

[0093] S2.1, frequency domain transformation processing: performing discrete cosine transformation (DCT) on the carrier image to convert the spatial domain pixel data into frequency domain coefficients, including the following sub-steps:

[0094] S2.1.1, block processing: divide the carrier image into multiple image blocks; performing block processing to obtain multiple image blocks;

[0095] The following is an example of dividing the carrier image into multiple image blocks. Suppose that the carrier image is divided into image blocks, denoted as:

[0096] ,

[0097] wherein represents the th image block.

[0098] S2.1.2, performing discrete cosine transform (DCT): performing two-dimensional discrete cosine transform on each of the image blocks to obtain the corresponding frequency domain coefficient matrix , denoted as:

[0099] ,

[0100] wherein represents the frequency domain coefficient at coordinate in the th frequency domain coefficient matrix .

[0101] S2.2, reorganizing the frequency domain coefficients:

[0102] unfolding each of the frequency domain coefficient matrices obtained in step S2.1.2 into a one-dimensional sequence in a fixed Zigzag order:

[0103] ,

[0104] Then, aggregating the frequency domain coefficients at the same position in all the frequency domain coefficient matrices to form a th-dimensional aggregated coefficient vector:

[0105] , ,

[0106] wherein represents the element at position in the one-dimensional sequence unfolded from the th frequency domain coefficient matrix ;

[0107] ​​Finally, the complete frequency domain embedding domain is obtained. .

[0108] In step S3 of this embodiment, the privacy image to be protected is subjected to bit-plane splitting to obtain multiple privacy bit-planes, specifically including:

[0109] The 8-bit binary bits of each pixel in the privacy-protecting image are extracted bit by bit and aggregated to form 8 privacy bit planes. Each privacy bit plane consists of bits from all pixels that occupy the same bit position. The 8 privacy bit planes are denoted as... .

[0110] In step S4 of this embodiment, based on the embedding domain constructed in step S2, eight privacy bit plane information are written into the frequency domain coefficients to form dense frequency domain coefficients, and then converted to the spatial domain to obtain a dense image.

[0111] Specifically, it includes the following sub-steps:

[0112] S4.1 Nested Quantization Index Modulation: Two layers of nested quantization index modulation are used to embed the eight privacy bit planes into the frequency domain embedding domain respectively. The first layer embeds the privacy bit planes. The second layer embeds a privacy bit plane. .

[0113] S4.1.1 First-level quantization index modulation: The first four privacy bit planes are embedded into all coefficients in the frequency domain embedding domain using first-level quantization index modulation. The parity quantification center.

[0114] The first four privacy planes Flattened to one dimension, denoted as Its length is All coefficients in the frequency domain embedding domain It is also flattened into one dimension, denoted as Its length is Quantization step size Will Modulation embedding The process is as follows:

[0115] ,

[0116] in It is a quantitative center. When When it is 0, Quantized to quantization step size Even multiples of; when When it is 1, Quantized to quantization step size an odd multiple of For sequence the first coefficient in the sequence , the first bit in the sequence , the first bit in the sequence

[0117] S4.1.2, second layer quantization index modulation: on the basis of the first layer quantization result, the last 4 privacy bit planes are embedded into the offset quantization center. First, they are flattened into one dimension, denoted as , whose length is 1 x (4UV), and then the embedding is completed using the following process:

[0118] ,

[0119] ,

[0120] wherein , wherein is the offset coefficient applied by the second layer (relative to the proportion of ), is the first coefficient in the sequence , is the first bit in the sequence and after embedding and .

[0121] S4.2, inverse discrete cosine transform: the encrypted frequency domain coefficients after embedding the privacy image are recombined into frequency domain matrices, and inverse discrete cosine transform is performed to convert to the spatial domain. Subsequently, the spatial domain image blocks are spliced according to the spatial position to obtain the encrypted image.

[0122] In step S5 of the embodiment, the inverse process is performed based on the encrypted image to reconstruct the privacy image, specifically including the following steps:

[0123] S5.1, inverse nested quantization index modulation: the encrypted image is subjected to the same blocking, discrete cosine transform and frequency domain coefficient reorganization process as step S2 to obtain the encrypted frequency domain embedding domain , and two layers of bit information are extracted from each frequency domain coefficient in turn.

[0124] S5.1.1, first layer inverse quantization index modulation: the first 4 privacy bit plane one-dimensional sequence embedded in step S4.1.1 is extracted from the encrypted frequency domain embedding domain . The principle of this layer of inverse quantization index modulation is: according to each frequency domain coefficient ​​The distance relationship between the distance from the even quantization center and the distance from the odd quantization center is used to determine the first-layer bit information carried thereby. Specifically, if the distance from the even quantization center is closer, it is determined that the first-layer bit corresponding to the coefficient is 0; if the distance from the odd quantization center is closer, it is determined that the first-layer bit corresponding to the coefficient is 1. The specific determination process is as follows:

[0125] ,

[0126] ,

[0127] denotes the distance from the nearest even quantization center, denotes the distance from the nearest odd quantization center. If

[0128] , it is determined that the first-layer bit is 0, otherwise the first-layer bit is determined to be 1, is the i-th bit of the one-dimensional sequence . S5.1.2, second-layer inverse quantization index modulation: on the basis of the one-dimensional sequence extracted from the first-layer inverse quantization index, a one-dimensional sequence

[0129] is further extracted from the frequency domain embedding domain . The principle of the second-layer inverse quantization index modulation is as follows: if the frequency domain coefficient is located to the right of the quantization center corresponding to the first-layer bit, it is determined that the second-layer bit is 1; if it is located to the left of the quantization center, it is determined that the second-layer bit is 0. The specific process is as follows: ,

[0130] When , it is determined that the second-layer bit

[0131] is 1; when , it is determined that the second-layer bit is 0, is the i-th bit of the one-dimensional sequence . According to the embedding order, the two layers of bits extracted are respectively filled into the corresponding 8 privacy bit planes, so that 8 reconstructed privacy bit planes are obtained.

[0132]

[0133] ​​​​​​S5.2, privacy image reconstruction: reconstructing the privacy image from the one-dimensional sequence extracted in step S5.1 and reconstructing the privacy image.

[0134] S5.2.1, privacy bit reorganization: converting the one-dimensional sequence and into eight-dimensional reconstruction bit planes with dimensions of In a lossless environment, the reconstruction bit plane is consistent with the original privacy bit plane obtained in step S3; in a lossy environment, there is only a slight error.

[0135] S5.2.2, privacy image reconstruction: reconstructing the reconstruction bit plane into pixel grayscale values to generate a reconstructed privacy image The specific calculation formula is:

[0136] ,

[0137] wherein, is the pixel value of the reconstructed privacy image at position , and represents the bit value of the th reconstruction bit plane at position .

[0138] The above provides a kind of privacy image protection-oriented robust steganography method provided by the present application, and MATLAB2022 software is used for simulation. For ease of illustration, small size matrix is used for example demonstration. The privacy image and cover image selected in the embodiment are both gray matrix, as shown in Figure 2 and Figure 3 . Parameter settings are as follows: first layer quantization step , second layer offset coefficient .

[0139] The simulation process is as follows:

[0140] (1) obtain carrier image: according to the resolution of the privacy image, scale the cover image to four times the resolution of the privacy image to obtain the carrier image. In the present example case, the resolution of the privacy image is , so the resolution of the carrier image should be . Then the cover image is scaled using bilinear interpolation to obtain the carrier image as shown in Figure 4 . ​

[0141] (2) Constructing the frequency domain embedding domain: for carrier images conduct Obtain a block The block is then subjected to a two-dimensional discrete cosine transform to obtain, as shown below. Figure 5 The frequency domain coefficient matrix shown .

[0142] For the frequency domain coefficient matrix Expand according to Zigzag order to form a one-dimensional sequence. Then, for the frequency domain coefficient matrix... The frequency domain coefficients at the same position are aggregated to obtain the frequency domain embedding domain. .

[0143] because There is only one in the middle. matrix Therefore, each There is only one coefficient, and for .

[0144] (3) Privacy bit plane splitting process: splitting the privacy image Each pixel is split into 8 privacy bit planes using 8 bits, represented as follows: ,

[0145] =[0 1 0 1;1 0 1 1;0 0 0 0;1 1 1 0];

[0146] =[1 1 1 0;0 0 0 0;1 1 0 1;0 0 0 0];

[0147] =[0 1 1 0;0 0 0 0;0 0 0 1;0 0 0 1];

[0148] =[1 0 1 1;0 1 0 0;1 0 1 1;0 1 0 1];

[0149] =[1 0 1 1;0 1 0 1;1 0 1 1;0 0 0 0];

[0150] =[1 1 1 1; 0 1 0 0; 0 0 1 1; 1 1 0 0];

[0151] =[0 1 1 0;1 1 0 0;0 1 0 1;0 1 0 0];

[0152] =[0 0 1 0;0 0 1 1;0 0 0 0;1 1 1 1].

[0153] (4) Steg embedding: using nested quantization index modulation to embed 8 privacy bit planes Embedded sequentially in the frequency domain and embedding domain Each coefficient in the [database name]. The first 4 privacy bit planes. Embedded in the first layer, the last 4 privacy bit planes Embedded in the second layer.

[0154] Assuming the first layer quantization step size Second layer offset coefficient For ease of demonstration, the following example shows two bits nested within the first two frequency domain coefficients:

[0155] =1017.5, =0, then , =1, then .

[0156] =13.5, =1, then , =0, then .

[0157] Will All bitstreams are embedded The dense frequency domain coefficients obtained in the middle and later are recombined as Frequency domain matrix, and for each The coefficient matrix is ​​subjected to inverse discrete cosine transform to reconstruct each image patch. The reconstructed image patches are then stitched together according to their original spatial positions to obtain a dense image. Since there is only one such image in this example... block, so get one Dense images like Figure 6 .

[0158] (5) Reconstructing privacy-preserving images: For images containing privacy... conduct Block partitioning and discrete cosine transform are used to obtain a dense frequency domain embedding domain. This example case involves densely packed images. Dense frequency domain embedding obtained by block partitioning and discrete cosine transform like Figure 7 .

[0159] Then, inverse nested quantization index modulation is performed on each of the above frequency domain coefficients to extract 8 privacy bit planes, and recombine them into pixel values to obtain the privacy image. In this example, the bit positions extracted from the first two frequency domain coefficients are taken as an example to illustrate the extraction of the privacy image:

[0160]

[0161] First, the first layer of bit positions is extracted. Since , i.e. is closer to . Therefore, .

[0162] Then, the second layer of bit positions is extracted. Since , i.e. , so .

[0163]

[0164] First, the first layer of bit positions is extracted. Since , i.e. is closer to . Therefore, . Then, the second layer of bit positions is extracted. Since , i.e. , so When all 8 privacy bit planes are extracted and converted into pixel values to obtain the reconstructed privacy image, it is consistent with Figure 2 .

[0165] The performance of the privacy protection method based on robust steganography proposed in this embodiment is analyzed as follows.

[0166] I. Analysis of cover image quality

[0167] At present, the most widely used image quality evaluation index is the Peak Signal-to-Noise Ratio (PSNR), where the larger the PSNR value, the better the image quality. Here, a resolution of hornbill image is selected as the privacy image, and a resolution of baboon image is selected as the cover image. The cover image is scaled to a carrier image with a resolution of , and then the privacy image is hidden in it to obtain the stego image, as shown in Figure 8 . Figure 8 ​​​​It can be found that the stego image in the embodiment has no obvious visible distortion, can ensure the secure transmission of the stego image in the network, and ensures the security of the private image. In addition, Table 1 shows the quality of the stego image.

[0168] Table 1 stego image quality

[0169]

[0170] It can be found from Table 1 that the PSNR of the stego image in the embodiment is 31.3761 dB, which means that the quality is close to that of the carrier image, can ensure the secure transmission of the stego image in the network, and ensures the security of the private image.

[0171] II. Quality of reconstructed private image

[0172] The quality of the reconstructed private image directly determines its usability. An excellent robust steganography method should be able to realize lossless reconstruction of the private image in a lossless environment, and recover a private image that is still semantically clear and usable in a lossy environment. Table 2 shows the quality comparison of the reconstructed private image under the condition that the stego image is not subjected to degradation and is subjected to different degradation. In addition, the visualization result of the reconstructed private image under the condition of no degradation is as shown in Figure 9 .

[0173] Table 2 quality comparison of reconstructed private image

[0174]

[0175] As can be seen from Table 2, under the condition of no degradation, the PSNR of the private image reconstructed from the stego image reaches infinity, and the visualization effect is as shown in Figure 9 , without any visible distortion. This shows that in a lossless environment, the present application can realize true pixel-by-pixel lossless recovery. In the degradation scenarios of JPEG compression, noise, etc., the PSNR of the reconstructed private image is still higher than 30 dB, which shows that in the presence of degradation, the present application can still obtain a private image with actual usable visual quality. Therefore, the present application can provide more reliable security protection for medical images, judicial evidence, and cross-platform private image transmission applications, and has clear engineering practical value.

[0176] Based on the same inventive concept, one embodiment of the present application also provides a robust steganography device for private image protection, which is used to implement the robust steganography method for private image protection of the above-mentioned embodiments. The device comprises:

[0177] a carrier image construction module configured to scale the cover image to obtain a carrier image capable of carrying private image information;

[0178] a frequency domain processing module, configured to perform frequency domain transformation on the carrier image to construct a frequency domain embedding domain;

[0179] a privacy image splitting module, configured to perform bit plane splitting processing on the privacy image to be protected to obtain privacy bit planes;

[0180] a steganographic embedding module, configured to embed the privacy bit planes into the frequency domain embedding domain to obtain a stego image;

[0181] a privacy image reconstruction module, configured to perform an inverse process on the stego image to reconstruct the privacy image to be protected.

[0182] It is worth noting that the apparatus embodiment is corresponding to the method embodiment described above, and the implementation manners of the method embodiment are applicable to the apparatus embodiment and can achieve the same or similar technical effects, and thus will not be described herein.

[0183] Based on the same inventive concept, one embodiment of the present application further provides a computer readable storage medium storing one or more programs, the one or more programs comprising instructions which, when executed by a computing device, cause the computing device to perform the robust steganographic method for privacy image protection of the above-described embodiments.

[0184] Based on the same inventive concept, one embodiment of the present application further provides a computing device comprising one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs comprise instructions for performing the robust steganographic method for privacy image protection of the above-described embodiments.

[0185] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0186] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0187] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0188] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0189] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the technical solutions of the present application. Although the present application is described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced, and any modification or replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A robust steganography method for privacy image protection, characterized in that, The method comprises: scaling the cover image to obtain a carrier image capable of carrying private image information; The carrier image is subjected to a frequency domain transformation to construct a frequency domain embedding domain, including: the carrier image is subjected to a block processing to obtain image blocks; each image block is subjected to a discrete cosine transformation to obtain a corresponding frequency domain coefficient matrix , ; the frequency domain coefficient matrices of all the image blocks are reorganized to obtain the frequency domain embedding domain; Bit plane splitting is performed on the privacy image to be protected to obtain a privacy bit plane, including: expanding the frequency domain coefficients in a frequency domain coefficient matrix in a Zigzag order into a one-dimensional sequence, denoted as: ； aggregating the frequency domain coefficients at the same position in the one-dimensional sequence of all the frequency domain coefficient matrices, to form a first dimensional aggregated coefficient vector, denoted as: aggregating the frequency domain coefficients at the same position in the one-dimensional sequence of all the frequency domain coefficient matrices, to form a first dimensional aggregated coefficient vector, denoted as: ， ， wherein denotes a frequency domain coefficient matrix position of the one-dimensional sequence element at position resulting in a complete frequency domain embedding domain ; embedding the private bit plane into the frequency domain embedding domain to obtain a stego image; performing an inverse process on the stego image to reconstruct the private image to be protected.

2. The robust steganography method for privacy-oriented image protection according to claim 1, wherein, The scaling of the cover image to obtain a carrier image capable of carrying private image information comprises: determining the resolution of the private image, and scaling the cover image to four times the resolution of the private image by using an interpolation algorithm to obtain the carrier image; The interpolation algorithm uses any one of the nearest neighbor interpolation method, the bilinear interpolation method and the cubic convolution interpolation method.

3. The robust steganography method for privacy-oriented image protection according to claim 1, wherein, The bit plane splitting processing of the private image to be protected comprises: The binary bits of each pixel in the privacy image to be protected are extracted and aggregated bit by bit respectively to obtain 8 privacy bit planes corresponding to the bit positions Each privacy bit plane is composed of the bits of all pixels of the privacy image to be protected at the same bit position.

4. The robust steganography method for privacy-oriented image protection according to claim 3, characterized in that, embedding the private bit plane into the frequency domain embedding domain to obtain a stego image comprises: embedding the bit information in the private bit plane into the frequency domain embedding domain by using a nested quantization index modulation algorithm, which comprises two layers of quantization index modulation processes: The first layer quantization index modulation embeds the first 4 privacy bit planes into the frequency domain embedding domain In one embodiment, the embedding process is: The first four privacy planes Flattened to one dimension, denoted as Embedding the frequency domain into the domain Flattened to one dimension, denoted as , with quantization step size will be modulation embedding wherein is represented as: ， in It is a quantitative center. For sequence The first in One coefficient, For sequence The first in Position; when When it is 0, Quantized to quantization step size Even multiples of; when When it is 1, Quantized to quantization step size An odd multiple of; The second layer quantization index modulates the first layer quantization result based on the last 4 privacy bit planes Embedding into the offset quantization center The embedding process is as follows: The last four privacy planes Flattened to one dimension, denoted as ; The embedding is performed in the following manner: ， ， wherein, , is an offset coefficient applied to the second layer, is is the coefficient embedded in and after the frequency domain coefficients containing the secret, is the bit in the sequence ; The frequency domain coefficients are recombined into a frequency domain matrix and converted to the spatial domain to obtain the stego image.

5. The robust steganography method for privacy-oriented image protection according to claim 4, characterized in that, performing an inverse process on the stego image to reconstruct the private image to be protected comprises: The dense image is subjected to block segmentation, discrete cosine transform, and frequency domain coefficient reconstruction to obtain the dense frequency domain embedding domain. The inverse nested quantization index modulation algorithm is used to embed the domain from the dense frequency domain. The process involves extracting two layers of bit information sequentially, divided into two layers of inverse quantization index modulation: The first layer inverse quantization index modulation extracts the embedded first four privacy bit planes one-dimensional sequence from the stego frequency domain embedding domain The extraction process is represented as:​ ， ， in, For dense frequency domain embedding domain The first in One coefficient, Indicates distance The most recent even quantization center, Indicates distance The most recent odd quantization center; If , then determine that the first layer bit is 0, otherwise determine that the first layer bit is 1, is the th bit of a one-dimensional sequence ; The second-layer inverse quantization index modulates the one-dimensional sequence extracted from the first-layer inverse quantization index. Building upon this foundation, we continue to embed the domain from the dense frequency domain. Extract one-dimensional sequence The extraction process is represented as follows: ， When , the second layer bit is determined to be 1; when , the second layer bit is determined to be 0, is the first bit of a one-dimensional sequence ; transforming a one-dimensional sequence and into 8 reconstructed bit-planes; reconstructing the eight reconstructed bit planes into pixel gray scale values to generate a reconstructed private image.

6. A robust steganography device for privacy image protection, characterized in that, The method comprises: a carrier image construction module configured to scale the cover image to obtain a carrier image capable of carrying private image information; A frequency domain processing module is configured to perform frequency domain transformation on the carrier image to construct a frequency domain embedding domain, and the specific process is as follows: performing block processing on the carrier image to obtain a plurality of image blocks; performing discrete cosine transformation on each image block to obtain a corresponding frequency domain coefficient matrix , ; and reorganizing the frequency domain coefficient matrices of all the image blocks to obtain the frequency domain embedding domain.​ The privacy image splitting module is used to perform bit-plane splitting on the privacy image to be protected, obtaining the privacy bit-plane. The specific process is as follows: the frequency domain coefficient matrix is ​​split... The frequency domain coefficients in the figure are expanded into a one-dimensional sequence according to Zigzag order, as follows: ； aggregating the frequency domain coefficients at the same position in the one-dimensional sequence of all the frequency domain coefficient matrices to form a first dimensional aggregated coefficient vector, denoted as: ​ ， ， wherein denotes a frequency domain coefficient matrix position of the one-dimensional sequence element at position resulting in a complete frequency domain embedding domain ; a steganographic embedding module configured to embed the private bit plane into the frequency domain embedding domain to obtain a stego image; a private image reconstruction module configured to perform an inverse process on the stego image to reconstruct the private image to be protected.

7. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method of any of claims 1-6. The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the robust steganographic methods for private image protection according to claims 1-5.

8. A computing device, comprising: The one or more processors, the memory, and the one or more programs are configured to perform any of the robust steganographic methods for private image protection according to claims 1-5.

Images