Robust video steganography method and system based on neighbor block difference cost

Abstract

The invention belongs to the field of information security, and relates to a robust video steganography method and system based on neighbor block difference cost, which can resist video transcoding of a social platform and is used for covert communication of a lossy channel. According to the method, a cost distribution method based on neighbor block difference is designed to evaluate the robustness of an embedded unit, a global robust modulation mode is selected by using a minimization distortion frame, and the embedded unit is modulated by adopting quantization index modulation according to the global robust modulation mode to embed secret information. The robust video steganography method and system provided by the invention integrate strategies such as adjacent block difference cost, an optimal robust modulation mode and a minimization distortion frame, and can effectively support robust video steganography communication under a social network lossy channel.

Description

Technical Field

[0001] The present invention belongs to the field of information security and is a robust video steganography method and system based on neighboring block difference cost. The method can resist video transcoding on social platforms and is used for covert communication in lossy channels. Background Art

[0002] Information hiding technology in the field of information security consists of two parts: steganography and steganalysis. Steganography primarily aims to embed secret information into media such as images, audio, and video to achieve covert communication. Steganalysis, on the other hand, uses machine learning and pattern recognition methods to distinguish between ordinary media and steganographic files.

[0003] In today's information age, digital video, with its rich content, high real-time performance, and strong visual performance, has gradually replaced digital images as the most influential communication medium. Using digital video as a covert communication carrier has many advantages:

[0004] 1) High embedding capacity: Compared with other carriers such as images and audio, videos have rich coding domain elements and can accommodate more secret information. Steganographers can split secret information into several data segments based on the application scenario and video characteristics, and embed them into different areas and embedding domains of the carrier video.

[0005] 2) Universality and frequency: Video transmission has gradually become the mainstream form of network communication, especially with the widespread use of social media, video sharing platforms and instant messaging tools. The transmission of video content is becoming increasingly popular. Hiding covert communications in video transmission is less likely to arouse suspicion, thereby improving the security of covert communications.

[0006] Generally speaking, when designing a steganography algorithm, the following factors need to be considered comprehensively:

[0007] 1) Imperceptibility. This mainly includes sensory indistinguishability and statistical undetectability. The former refers to the visual difference between the stego video and the target video being indistinguishable to the human sensory system; the latter refers to the statistical properties of the stego video and the target video being approximately the same for a given feature.

[0008] 2) Embedding Capacity: This refers to the maximum amount of data that can be embedded into the carrier video while ensuring a certain level of imperceptibility.

[0009] 3) Robustness. This reflects the ability to resist lossy digital signal processing. This refers to the ability to reliably and completely recover the embedded information even after the steganographic video has undergone lossy digital signal processing (such as blurring or lossy compression).

[0010] Continuous advances in video coding and decoding technologies, high-performance computing, and computer networking have enabled digital videos to be rapidly produced and transmitted in real time over the internet while maintaining high coding performance and visual fidelity. The rise of numerous social networks and video-sharing sites has provided new application scenarios for video-based steganography, but it has also brought significant challenges. Most social networks perform lossy digital signal processing, such as visual enhancement, resampling, and secondary compression, on videos uploaded and shared by users. The types of operations and parameters involved are generally confidential. Recovering embedded message data from files processed by social network black boxes is crucial for implementing highly secure, contactless steganographic communication based on social networks.

[0011] Currently, robust video steganography technology is still in its early stages of development and can be divided into the following two categories based on the type of embedding domain:

[0012] 1) Pixel-Domain Robust Video Steganography: The main idea is to embed message data by directly modulating the pixel values ​​of video frames. However, such methods often use repeated embedding to enhance the robustness of the method, but this sacrifices the embedding capacity and results in limited robustness.

[0013] 2) Spatial Transform Domain Robust Video Steganography: This involves performing one or more data transformations (such as discrete cosine transforms) on the pixel values ​​of a video frame, then modulating the resulting transform domain elements to embed the message data. Spatial transform domains exhibit strong robustness against lossy processing on social media platforms, and current robust video steganography methods often employ this type of embedding domain.

[0014] After patent search, the relevant patent applications in the field of this invention are as follows:

[0015] 1) Chinese patent application number 202410107837X, "Robust Video Steganography Method and Apparatus Based on Feature Point Synchronization," discloses a robust video steganography method and apparatus based on feature point synchronization. The technical approach of the method described in this patent is: feature point identification is performed by using the Speeded-Up Robust Features (SURF) algorithm on the luminance component (Y) of the steganographic video, and a robust frame is selected for embedding using a priori robust video frame optimization strategy based on the proportion of the first principal component of the video frame and the response strength of the feature points. In contrast, the present invention designs a method based on the difference cost of neighboring blocks, which optimizes the modulation mode for each embedding unit to embed secret information. Based on this, a distortion minimization framework is used to minimize the modulation cost, so that the modulation mode is the globally optimal robust modulation. Therefore, the design concept and implementation method of this patent application are significantly different from those of the present invention.

[0016] 2) The Chinese patent application number 202211246630.8, "Robust Video Steganography Method and Device Based on Adaptive Quantization", discloses a robust video steganography method and device based on adaptive quantization. The technical route of the method described in this patent is: by extracting the principal component analysis (PCA) features of the video luminance component, establishing a mapping relationship between the PCA features and the quantization step required for quantization index modulation (QIM), and then realizing adaptive steganography. In contrast, the present invention adopts a new cost evaluation framework, which optimizes the selection of modulation mode under the distortion minimization framework by designing the neighboring block difference cost and the modulation coefficient distortion cost. Therefore, the present invention has significant differences from the above-mentioned patents in terms of technical route and implementation mechanism.

[0017] 3) Chinese patent application number 202011388479.2, "Robust Video Steganography Method and Apparatus Based on Audio Side Channel," discloses a robust video steganography method and apparatus based on an audio side channel. The method described in this patent employs a technical approach: by separating the video stream and the accompanying audio stream, a steganographic channel and a side channel are constructed, respectively, to achieve synchronous transmission of message data and side information. In contrast, the present invention dispenses with complex side channel design and directly employs an optimal robust modulation scheme to embed information, simplifying the system architecture and improving steganography efficiency. Therefore, the design philosophy and implementation of this patent application differ significantly from those of the present invention. Summary of the Invention

[0018] The purpose of the present invention is to provide a robust video steganography method and system that integrates strategies such as neighboring block difference cost, optimal robust modulation method, and distortion minimization framework, so as to effectively support robust video steganography communication under lossy channels of social networks.

[0019] Compared to other robust video steganography methods, this invention proposes a cost allocation method based on neighboring block differences to evaluate the robustness of each embedding unit (EU). Based on this, a globally robust modulation scheme is selected for the embedding unit using a distortion minimization framework to achieve robust steganographic embedding of message data. This technical concept, therefore, is distinct from any previous robust video steganography method and is particularly suitable for high-reliability robust video steganographic communication over lossy channels.

[0020] According to research, most current robust video steganography methods require embedding secret information into robust regions and constructing side channels to synchronize robust regions, that is, synchronizing robust frames by transmitting identification information of robust frames or synchronizing robust regions through robust features. However, the lossy processing of social media will impose noise on the side channels, which makes it difficult to achieve complete synchronization of robust regions. Once the synchronization information changes, the extractor will extract secret information from the wrong region, which limits the success rate of covert communication under lossy channels. In response to the above shortcomings and deficiencies, the present invention has the following improvements: First, a cost allocation method based on neighboring block differences is designed to evaluate the robustness of the embedding unit; second, a distortion minimization framework is used to select a globally robust modulation method; finally, quantized index modulation is used to modulate the embedding unit according to the globally robust modulation method to embed secret information, thereby providing reliable technical support for robust video steganography communication under lossy channels of social networks.

[0021] Specifically, the technical solution adopted by the present invention is as follows:

[0022] A robust video steganography method based on neighboring block difference cost, wherein the embedding process of message data includes the following steps (if not otherwise specified, the following steps are executed by a computer), such as Figure 1 As shown:

[0023] 1) Preprocessing: Decode the carrier video to obtain a YUV video frame sequence;

[0024] 2) Divide the embedding unit: Divide the U component of the video frame into non-overlapping blocks, each of which is an embedding unit;

[0025] 3) Obtain carrier coefficients: For each embedding unit, obtain its carrier coefficients in the DWT-SVD domain;

[0026] 4) Obtaining modulation cost: For each embedded unit, obtain its modulation coefficient distortion cost and neighboring block difference cost;

[0027] 5) Obtaining the optimal robust modulation scheme: Based on the modulation cost of each embedded unit, the global optimal robust modulation scheme is calculated in combination with the distortion minimization framework;

[0028] 6) Robust modulation: Based on the globally optimal robust modulation scheme, the secret information encoded by the error correction code (ECC) is embedded by modulating the carrier coefficient of each embedding unit;

[0029] 7) Embedded unit reconstruction: Based on the modulated density coefficients, inverse singular value decomposition (ISVD) and inverse wavelet transform (IDWT) are used to obtain embedded units, and the embedded units are sequentially organized into a YUV video frame sequence;

[0030] 8) Post-processing: Encoding the above-mentioned YUV video frame sequence that has undergone steganographic processing to obtain the steganographic video (i.e., the encrypted video).

[0031] Furthermore, the above method also includes a message data extraction process (if not otherwise specified, the following steps are all executed by a computer), such as Figure 2 As shown, the following steps are included:

[0032] 1) Preprocessing: Decode the steganographic video to obtain a YUV video frame sequence;

[0033] 2) Divide the embedding unit: Divide the U component of the video frame into non-overlapping blocks, each of which is an embedding unit;

[0034] 3) Obtaining the density coefficient: For each embedded unit, obtain its density coefficient in the DWT-SVD domain;

[0035] 4) Steganographic decoding: Based on the interval category of the cryptanalysis coefficient of each embedded unit, steganographic codes (STCs) are used to decode and obtain the secret information;

[0036] 5) Post-processing: Perform error correction code decoding, decompression and other post-processing on the secret information to finally obtain the reconstructed message data.

[0037] Furthermore, the above method of dividing the U component of the video frame into non-overlapping blocks can be divided into 16×16 blocks or other values. Preferably, the blocks are divided into 16×16 blocks to balance the performance among robustness, visual effect and embedding capacity.

[0038] Based on the above introduction to the method, the neighboring block difference cost, optimal robust modulation method and corresponding message extraction strategy proposed in the present invention are now described in detail as follows.

[0039] [1] Modulation coefficient distortion cost and neighboring block difference cost

[0040] This method evaluates the modulation cost of each embedding unit by calculating the modulation coefficient distortion cost and the neighboring block difference cost of the embedding unit.

[0041] The modulation coefficient distortion cost is used to evaluate the coefficient distortion generated by the carrier coefficient of the embedding unit during the quantization index modulation process. Due to the different bits to be embedded, the coefficient distortion generated by the modulation is also different. In order to distinguish the modulation modes that generate different coefficient distortions, the present invention defines the modulation that introduces a large coefficient distortion from the modulation embedding unit to the midpoint of the nearest interval as LCDM (Large Coefficient Distortion Modulation). Conversely, the modulation that introduces a small coefficient distortion from the modulation embedding unit to the midpoint of the nearest interval is defined as SCDM (Small Coefficient Distortion Modulation). Figure 3 Different from SCDM, LCDM modulates the carrier coefficients to another type of interval.

[0042] Specifically, given an embedding unit P i , first obtain its carrier coefficient c in the DWT-SVD domain i , that is, for P i First, discrete wavelet transform (DWT) is used to obtain the low-frequency subband LL, and then singular value decomposition (SVD) is used to process the low-frequency subband LL to obtain the singular value diagonal matrix, and the maximum singular value is taken as the carrier coefficient c of the embedding unit. i Assuming that the quantization index modulation step size is q, the coefficient distortion caused by SCDM is:

[0043]

[0044] Among them, [] indicates rounding. The coefficient distortion caused by LCDM is:

[0045]

[0046] The neighboring block difference cost is used to evaluate the difference between the neighboring blocks of an embedded unit and its adjacent embedded units. Since neighboring block differences affect the amount of transcoding noise introduced during video transcoding, and thus the robustness of the embedded unit, the neighboring block difference cost is also taken into account when designing the modulation cost.

[0047] Specifically, for the embedding unit P c , and its neighboring block difference cost is defined as:

[0048]

[0049] Where m represents P c The number of adjacent embedding units, MAE represents the absolute mean error of the two matrices, P jIndicates P c The adjacent embedding units of (x, y) represent the index value of the matrix, h represents the height of the matrix, and w represents the width of the matrix. Since only the coding block will be used for prediction during the video encoding process, P is selected. c The left and upper neighboring blocks of are used as neighboring blocks, such as Figure 4 shown.

[0050] The modulation cost is composed of the modulation coefficient distortion cost and the difference cost of adjacent blocks. When the code rate control mode is CRF (Constant Rate Factor), the greater the difference between adjacent blocks, the greater the probability of error in the embedded unit. Combined with this feature, for the embedded unit P i , and its modulation cost is defined as follows:

[0051]

[0052] Among them, c i is from the embedding unit P i Obtained carrier coefficient, s i It is c i The density coefficient after modulation, IDC i Indicates c i The interval category of the quantitative index, ids i Indicates s i The interval category of the quantitative index, Indicates that modulation does not change IDC i The interval category of the quantization index, Indicates that modulation will be idc i The quantization index is modulated to other intervals. α is a parameter that controls the weight of the modulation coefficient distortion cost and the difference cost between neighboring blocks. Indicates the use of maximum minimization regularization To the interval [0,1], take and They are The maximum and minimum values ​​of Denotes the embedding unit P i The difference cost of neighboring blocks is n, and n represents the total number of embedding units.

[0053] [2] Optimal robust modulation scheme

[0054] According to the modulation cost of each embedded unit calculated based on the difference cost of neighboring blocks, the optimal robust modulation method is used to select the most robust modulation method globally based on the distortion minimization framework to modulate all embedded units.

[0055] Given a video, divide it into 16×16 non-overlapping blocks, denoted as embedding unit P = {P i|i∈[1,n]}, n is the total number of embedding units in the video. Let the secret information to be embedded be x={x i |i∈[1,m]}, where m is the bit length of the information to be embedded. The quantization step size of the quantization index modulation is q. The modulation process is as follows:

[0056] 1) Obtain the carrier coefficient of the embedding unit: Perform discrete wavelet transform on each embedding unit, take the low-frequency subband LL, perform singular value decomposition on the low-frequency subband LL, obtain the singular value diagonal matrix, and extract the maximum singular value as the carrier element of the embedding unit. Repeat the above operation until all embedding units are extracted, denoted as c = {c i |i∈[1,n]};

[0057] 2) Get the interval category of the carrier coefficient quantization index: For the carrier coefficient c i , get the interval category idc of its quantization index i , as shown below:

[0058]

[0059] Repeat the above process until the quantization index of all carrier coefficients is obtained, which is recorded as idc = {idc i |i∈[1,n]}.

[0060] 3) Obtaining the modulation cost of the embedding unit: The modulation cost of each embedding unit is calculated using the difference cost of the neighboring blocks, denoted as ρ = {ρ i |i∈[1,n]}.

[0061] 4) Minimizing the modulation cost: Using the stego code STCs to encode based on the minimization of the modulation cost, the quantized index interval category after stego-coding is obtained, which is denoted as:

[0062] ids=Enc(idc,ρ,x),

[0063] Where ids = {ids i |i∈[1,n]} is the quantized index interval category after steganalytic encoding.

[0064] 5) Obtain the optimal robust modulation mode: According to idc and ids, the optimal robust modulation mode is obtained, which is recorded as:

[0065] md=idc^ids,

[0066] Where md={md i |i∈[1,n]}∈{0,1} n , where md i Indicates IDC iThe modulation mode is 0, which indicates SCDM and does not change the interval category of the carrier coefficient quantization index. 1 indicates LCDM and changes the interval category of the carrier coefficient quantization index to another interval category. ^ indicates bitwise exclusive OR.

[0067] 6) Modulate the carrier coefficient: According to the optimal robust modulation method, modulate each carrier coefficient in turn. i , its optimal robust modulation mode is md i , the modulated density coefficient is s i , the modulation process is as follows:

[0068]

[0069] in:

[0070]

[0071] Repeat the above process until all carrier coefficients are modulated and obtain the carrier density coefficient s = {s i |i∈[1,n]}.

[0072] [3] Extraction process based on optimal robust modulation

[0073] The extraction process is the inverse process of the optimal robust modulation, combined with the steganographic video's encryption coefficient s = {s i |i∈[1,n]}, the extraction process based on optimal robust modulation includes the following steps:

[0074] 1) Get the interval category of the quantization index of the density coefficient: For the density coefficient s i , get the interval category ids of its quantization index i , as shown below:

[0075]

[0076] Repeat the above operation until all the interval categories of the density coefficients are obtained. i |i∈[1,n]}.

[0077] 2) Steganographic code decoding: Use STCs to perform steganographic code decoding on ids to obtain the secret information, as shown below

[0078] y=Dec(ids),

[0079] Where y={y i |i∈[1,m]} represents the extracted secret information.

[0080] The present invention also provides a robust video steganography system based on neighboring block difference cost, which includes a message data embedding module and / or a message data extraction module; the message data embedding module uses the message data embedding process in the above-mentioned method of the present invention to embed message data to generate a steganographic video; the message data extraction module uses the message data extraction process in the above-mentioned method of the present invention to extract message data from the steganographic video.

[0081] The robust video steganography method of the present invention has the following beneficial effects on the relevant technical fields:

[0082] 1) No need for synchronization of robust region identification information. Currently, the way to improve the robustness of most robust video steganography methods is to propose a robust region optimization strategy to embed the secret information into the robust region. In order to achieve blind extraction of secret information, previous methods generally construct a side channel to transmit the identification information of the robust region together. The extracting party obtains the robust region based on the robust region identification of the side channel, and then obtains the secret information from the robust region. The robustness of this type of method depends on the information identifying the robust region to be accurately synchronized, but due to lossy compression, the side channel cannot accurately synchronize the identification information, which limits the robustness of the method. The method proposed in the present invention does not require the synchronization of robust region identification information, and embeds the secret information by selecting the optimal robust modulation method, which solves the limitations of previous methods.

[0083] 2) Robustness is enhanced and can be used for social network channels and various encoders. Combining the modulation coefficient distortion and the difference between adjacent blocks, the present invention designs a cost function to evaluate the modulation cost of each embedded unit to assist the optimal robust modulation method in selecting the globally optimal robust modulation to modulate the embedded unit. In this way, global robustness is achieved, and the robustness is further improved compared to the usual methods. Moreover, the cost function is applicable to the CRF rate control mode, and most encoders and social network channels use content-adaptive video coding. Therefore, the cost function can be applied to most encoders and social network channels, and the robustness and channel support are further enhanced and expanded.

[0084] 3) Adaptive control of security and robustness. The cost function consists of two parts: modulation coefficient distortion and neighboring block dissimilarity. Neighboring block dissimilarity is related to video transcoding noise, while modulation coefficient distortion is related to modulation intensity. In other words, neighboring block dissimilarity is related to the robustness of the algorithm, while modulation coefficient distortion is related to the security of the algorithm. By setting different weights for these two costs, the security and robustness of the algorithm can be regulated. When the algorithm requires higher security, the weight of modulation coefficient distortion can be increased to enhance the security of the algorithm. Similarly, the robustness of the algorithm can be enhanced by increasing the weight of neighboring block dissimilarity. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 It is a schematic diagram of message data embedding of the present invention;

[0086] Figure 2 It is a schematic diagram of message data extraction of the present invention;

[0087] Figure 3 Schematic diagram of two modulation modes of the present invention;

[0088] Figure 4 Schematic diagram of the difference between adjacent blocks of the present invention;

[0089] Figure 5 It is a message data embedding flow chart of the present invention;

[0090] Figure 6 It is a message data extraction flow chart of the present invention. DETAILED DESCRIPTION

[0091] The present invention will be described below in conjunction with the embodiments and drawings.

[0092] The present invention supports embedding bitstream into a specified video, so the message data can be in any format, such as documents, images, audio and video, and the message data can be pre-processed with compression, encryption, error correction coding and other security and robustness enhancement methods. The message data embedding process of the robust video steganography method based on neighboring block difference cost proposed by the present invention is as follows: Figure 5 The specific operation details are as follows.

[0093] 1) Video Preprocessing: If the input video format is MP4, MKV, or AVI, use the Python toolkit to decode the video into YUV components. If the input video format is less common, such as WebM or TS, use FFmpeg to transcode the video into MP4 and then use the Python toolkit to decode the video into YUV components. If the input video is YUV, use the Python toolkit to directly read it and proceed to the next step.

[0094] 2) Divide the embedding unit: Extract all U components of the YUV sequence in the video and divide the U components into non-overlapping 16×16 pixel blocks, i.e., embedding units. Arrange all the embedding units in the video in order and record them as P = {P i |i∈[1,n]}, n is the total number of embedding units in the video.

[0095] 3) Extract the carrier coefficients of the embedding unit: Perform discrete wavelet transform on each embedding unit, take the low-frequency subband for singular value decomposition, and obtain the singular value diagonal matrix. On this basis, take the maximum singular value as the carrier coefficient of the embedding unit. This constitutes the chrominance component embedding domain carrier element set c = {c i |i∈[1,n]}.

[0096] 4) Calculate the modulation cost: extract the modulation coefficient distortion cost and the neighboring block difference cost.

[0097] 4.1) Obtain the modulation coefficient distortion cost, for the embedding unit P i , calculate the modulation coefficient distortion cost caused by LCDM as:

[0098]

[0099] Where q is the pre-agreed quantization index modulation step size, and [] indicates rounding to the nearest integer. The modulation coefficient distortion cost caused by SCDM is calculated as:

[0100]

[0101] 4.2) Calculate the difference cost of neighboring blocks, for the embedding unit P i , if P i If the embedding units on the left and above exist, the formula for calculating the difference between adjacent blocks is:

[0102]

[0103] Among them, P i,upper is the embedded unit P i Upper side embedded unit, P i,left is the embedded unit P i Left side embedded unit.

[0104] If P i If only the embedding unit on the left exists, the formula for calculating the difference between neighboring blocks is:

[0105]

[0106] If P i If only the upper embedding unit exists, the formula for calculating the difference between adjacent blocks is:

[0107]

[0108] If P i If the embedding units on the left and above do not exist, the formula for calculating the difference between adjacent blocks is:

[0109]

[0110] 4.3) Set the modulation cost, for each embedding unit P i , calculate the modulation coefficient distortion cost and the neighboring block difference cost respectively, then the modulation cost is:

[0111]

[0112] Among them, c i is from the embedding unit P i Obtained carrier coefficient, s i It is c i The density coefficient after modulation, IDC i Indicates c i The interval category of the quantitative index, ids i Indicates s i The interval category of the quantitative index, Indicates that modulation does not change IDC i The interval category of the quantization index, Indicates that modulation will be idc i The quantization index is modulated to other intervals. α is a parameter that controls the weight of the modulation coefficient distortion cost and the difference cost between neighboring blocks. Indicates the use of maximum minimization regularization To the interval [0,1], take and They are The maximum and minimum values ​​of .

[0113] 4.4) Repeat steps 4.1) to 4.3) until the modulation costs of all embedded units are obtained, that is, ρ = {ρ i |i∈[1,n]}.

[0114] 5) Get the interval category of the carrier coefficient quantization index: For the carrier coefficient c i , get the interval category idc of its quantization index i , as shown below:

[0115]

[0116] Repeat this process until the quantization index of all carrier coefficients is obtained, recorded as idc = {idc i |i∈[1,n]}.

[0117] 6) Minimize the modulation cost: Use the stego code STCs to encode based on the minimum modulation cost, and obtain the quantized index interval category after stego coding, which is recorded as

[0118] ids=Enc(idc,ρ,x),

[0119] Where ids = {ids i |i∈[1,n]} is the quantized index interval category after stego-coding. x={x i |i∈[1,m]} is the bit to be embedded with encryption information, and m is the length of the bit to be embedded with encryption information.

[0120] 7) Obtain the optimal robust modulation method: According to idc and ids, the optimal robust modulation method is obtained, which is recorded as

[0121] md=idc^ids,

[0122] Where md={md i |i∈[1,n]}∈{0,1} n , where 0 indicates SCDM, which does not change the interval category of the carrier coefficient quantization index. 1 indicates LCDM, which changes the interval category of the carrier coefficient quantization index to another interval category. ^ indicates bitwise exclusive OR.

[0123] 8) Modulate the carrier coefficient: According to the optimal robust modulation method, modulate each carrier coefficient in turn. i , its optimal robust modulation mode is md i , the modulated density coefficient is s i , the modulation process is as follows:

[0124]

[0125] in:

[0126]

[0127] Repeat the above process until all carrier coefficients are modulated and obtain the carrier density coefficient s = {s i |i∈[1,n]}.

[0128] 9) Reconstructing the embedding unit: For each encrypted coefficient, perform inverse singular value decomposition and inverse discrete wavelet transform operations in sequence to obtain the embedding unit with embedded encrypted bits.

[0129] 10) Video encoding: The embedded units are sequentially concatenated to obtain the U component of the video after embedding the secret information. The U component is then combined with the Y and V components of the original video and encoded to obtain the steganographic video.

[0130] The message data extraction process of the robust video steganography method based on neighboring block difference cost proposed in this invention is as follows: Figure 6 The specific operation details are as follows:

[0131] 1) Encrypted Video Preprocessing: If the input video format is MP4, MKV, or AVI, use the Python toolkit to decode the video into YUV components and read it. If the input video format is less common, such as WebM or TS, use FFmpeg to transcode the video to MP4, then use the Python toolkit to decode the video into YUV components and read it. If the input video is YUV, use the Python toolkit to read it directly and proceed to the next step.

[0132] 2) Divide the embedding unit: Extract all U components of the YUV sequence in the video and divide the U components into non-overlapping 16×16 pixel blocks, i.e., embedding units. Arrange all the embedding units in the video in order and record them as P = {P i |i∈[1,n]}, n is the total number of embedding units in the video.

[0133] 3) Obtain the embedding unit's carrier coefficient: Perform discrete wavelet transform on each embedding unit, take the low-frequency subband for singular value decomposition, and obtain the singular value diagonal matrix. On this basis, take the maximum singular value as the embedding unit's carrier coefficient. This constitutes the chrominance component embedding domain carrier element set s = {s i |i∈[1,n]}.

[0134] 4) Get the interval category of the quantization index of the density coefficient: For the density coefficient s i , get the interval category ids of its quantization index i , as shown below:

[0135]

[0136] Repeat the above operation until all the interval categories of the density coefficients are obtained. i |i∈[1,n]}.

[0137] 5) Steganographic code decoding: Use STCs to perform steganographic code decoding on ids to obtain the secret information, as shown below

[0138] y=Dec(ids),

[0139] Where y={y i |i∈[1,m]} represents the extracted secret information.

[0140] 6) Post-processing: The extracted secret bits are reassembled and post-processed such as error correction decoding and decompression as needed to finally obtain the reconstructed message data.

[0141] As can be seen from the above specific implementations, the embedding process of the present invention allows for adjustment of the preset parameter α required for controllable parameters of security and robustness according to algorithm requirements. Furthermore, the embedding unit block size can be modulated to control the embedding capacity. Therefore, the present invention possesses strong adaptability and flexibility.

[0142] The present invention also provides a robust video steganography system based on neighboring block difference cost, which includes a message data embedding module and / or a message data extraction module; the message data embedding module uses the message data embedding process in the above-mentioned method of the present invention to embed message data to generate a steganographic video; the message data extraction module uses the message data extraction process in the above-mentioned method of the present invention to extract message data from the steganographic video.

[0143] The message data embedding module can be further divided into a preprocessing submodule, an embedding unit division submodule, a carrier coefficient acquisition submodule, a modulation cost acquisition submodule, an optimal robust modulation mode acquisition submodule, a robust modulation submodule, an embedding unit reconstruction submodule, and a post-processing submodule. Similarly, the message data extraction module can be further divided into a preprocessing submodule, an embedding unit division submodule, a carrier coefficient acquisition submodule, a steganalysis decoding submodule, and a post-processing submodule. The specific operating procedures of each of these modules and submodules can be referenced in the corresponding procedures in the aforementioned method embodiments and will not be further elaborated here.

[0144] It should be understood that the methods and systems disclosed in the above embodiments provided by the present invention can be implemented in other ways. For example, the division of the above modules can have other division methods in actual implementation, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. The various modules in the present invention can be implemented in the form of software functional units, which can be stored in a computer-readable storage medium, including several instructions for causing a computer device to perform some or all steps of the method described in the present invention. For example, one embodiment of the present invention provides a computer device (computer, server, smart phone, etc.), which includes a memory and a processor, the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program includes instructions for executing each step in the method of the present invention. For example, another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, CD, etc.), the computer-readable storage medium stores a computer program, and when the computer program is executed by a computer, it implements each step of the method of the present invention.

[0145] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Those skilled in the art may modify or make equivalent substitutions for the technical solutions of the present invention without departing from the spirit and scope of the present invention. The scope of protection of the present invention shall be based on the claims.

Claims

1. A robust video steganography method based on neighboring block difference cost, characterized in that: The method includes a message data embedding process, wherein the message data embedding process includes the following steps: Decode the carrier video to obtain a YUV video frame sequence; Divide the U component of the video frame into non-overlapping blocks, each of which is an embedding unit; For each embedding unit, obtain its carrier coefficient in the DWT-SVD domain; For each embedding unit, obtain its modulation cost, where the modulation cost includes the modulation coefficient distortion cost and the neighboring block difference cost; According to the modulation cost of each embedded unit, the global optimal robust modulation method is calculated in combination with the distortion minimization framework; According to the globally optimal robust modulation method, the secret information encoded by the error correction code is embedded by modulating the carrier coefficient of each embedding unit; According to the modulated density coefficient, the inverse singular value decomposition and inverse wavelet transform are used to obtain the embedded units to achieve embedded unit reconstruction, and the embedded units are sequentially composed into a YUV video frame sequence; The YUV video frame sequence reconstructed by the embedding unit is encoded to obtain the steganographic video.

2. The method according to claim 1, characterized in that The modulation coefficient distortion cost is used to evaluate the coefficient distortion generated by the carrier coefficient of the embedded unit during the quantization index modulation process; the modulation with a large coefficient distortion introduced from the modulation embedded unit to the midpoint of the nearest interval is defined as LCDM, and the modulation with a small coefficient distortion introduced from the modulation embedded unit to the midpoint of the nearest interval is defined as SCDM; The coefficient distortion caused by the SCDM is: The coefficient distortion caused by the LCDM is: Among them, [] means rounding to the nearest integer, c i is the embedded unit P i The carrier coefficients in the DWT-SVD domain, q is the quantization index modulation step size.

3. The method according to claim 2, characterized in that The neighboring block difference cost is used to evaluate the difference between the embedding unit and the adjacent embedding units; for the embedding unit P c , and its neighboring block difference cost is defined as: Where m represents P c The number of adjacent embedding units, MAE represents the absolute mean error of the two matrices, P j Indicates P c The adjacent embedding units of , (x, y) represents the index value of the matrix, h represents the height of the matrix, and w represents the width of the matrix.

4. The method according to claim 3, characterized in that The modulation cost is composed of the modulation coefficient distortion cost and the neighboring block difference cost. i , and its modulation cost is defined as follows: Among them, c i is from the embedding unit P i Obtained carrier coefficient, s i It is c i The steganographic coefficient after modulation, IDC i Indicates c i The interval category of the quantitative index, ids i Indicates s i The interval category of the quantitative index, Indicates that modulation does not change IDC i The interval category of the quantization index, Indicates that modulation will be IDC i The quantization index is modulated to other intervals, and α is a parameter that controls the weight of the modulation coefficient distortion cost and the difference cost of the neighboring blocks. Indicates the use of maximum minimization regularization To the interval [0,1], take and They are The maximum and minimum values ​​of Denotes the embedding unit P i The difference cost of neighboring blocks is n, and n represents the total number of embedding units.

5. The method according to claim 4, characterized in that The global optimal robust modulation scheme is calculated based on the modulation cost of each embedded unit and combined with the distortion minimization framework, including: Get the carrier coefficients of the embedded unit; Obtain the interval category of the carrier coefficient quantization index, and then obtain the quantization index idc of all carrier coefficients; Get the modulation cost of the embedded unit; Use the stego code STCs to encode based on minimizing the modulation cost, and obtain the quantized index interval category after stego-coding, recorded as ids; The optimal robust modulation scheme is obtained according to idc and ids, which is recorded as: md=idc^ids, Where md={md i |i∈[1,n]}∈{0,1} n , where md i Indicates IDC i Modulation mode, 0 represents SCDM, which does not change the interval category of the carrier coefficient quantization index; 1 represents LCDM, which changes the interval category of the carrier coefficient quantization index to other interval categories; ^ represents bit-by-bit XOR.

6. The method according to claim 5, characterized in that The method of embedding secret information encoded by the error correction code by modulating the carrier coefficient of each embedding unit according to the globally optimal robust modulation method includes: According to the optimal robust modulation method, each carrier coefficient is modulated in turn; for the carrier coefficient c i , its optimal robust modulation mode is md i , the modulated density coefficient is s i , the modulation process is as follows: in, Repeat the above process until all carrier coefficients are modulated and obtain the carrier density coefficient s = {s i |i∈[1,n]}.

7. A robust video steganography method based on neighboring block difference cost, characterized in that: The method comprises a message data extraction process, wherein the message data extraction process extracts the message data from the steganographic video generated by the method according to any one of claims 1 to 6; The message data extraction process includes the following steps: Decode the steganographic video to obtain a YUV video frame sequence; Divide the U component of the video frame into non-overlapping blocks, each of which is an embedding unit; For each embedded unit, obtain its density coefficient in the DWT-SVD domain; According to the interval category of the steganographic coefficient of each embedded unit, the steganographic code STCs is used to decode and obtain the secret information; The secret information is post-processed to obtain the reconstructed message data.

8. A robust video steganography system based on neighbor block difference cost, characterized in that: It includes a message data embedding module and / or a message data extraction module; the message data embedding module adopts the method described in any one of claims 1 to 6 to embed message data to generate a steganographic video; the message data extraction module adopts the method described in claim 7 to extract message data from the steganographic video.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program comprises instructions for executing the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a computer, the method according to any one of claims 1 to 7 is implemented.

Images