CN108322757B - Video Steganography Method Based on Intra-frame Predictive Coding - Google Patents

Abstract

The present invention provides a video steganography method based on intra-frame prediction coding. By constructing a suitable distortion function and then using STC for actual embedding, it can ensure that the distortion caused by modifying the prediction mode is close to the minimum. In the present invention, when the sender embeds the message, it starts to embed from the first key frame _I1 , and obtains the optimal group set according to the mapping rule

the optimal set

The secret message sequence m _1,1 and the embedded perturbation set are sent to the first-layer embedder to obtain the secret group set

will contain secret group set

The secret message sequence m _{1, 2} and the embedded perturbation set are sent to the second-layer embedder to obtain the secret group set

According to the set of secret groups

Select the final prediction mode to generate video code; finally generate the encrypted video X′; when the receiver extracts the message, obtain the key frame after intra-frame prediction encoding from the encrypted video; start the message extraction from the first key frame, use The decoder obtains the prediction mode; obtains the dense set according to the mapping rule, and extracts the private message.

Description

Video Steganography Method Based on Intra-frame Predictive Coding

technical field

The invention belongs to the video information hiding technology, in particular to a video steganography method based on intra-frame prediction coding.

Background technique

Steganography is an important means to ensure the confidentiality of message transmission. Although message encryption can ensure the privacy of messages, the behavior of message transmission will be known to attackers, and the encrypted garbled ciphertext will also attract the interest of attackers; in contrast, steganography can ensure message transmission. The behavior is invisible, and the attacker cannot distinguish whether it is a normal transmission or an encrypted transmission, so steganography is an important supplement to encryption technology. Steganography can be applied to all kinds of high-security message transmission fields to effectively prevent attackers from attacking.

Video steganography can modify different coding links such as motion vectors, quantized DCT coefficients, variable length coding and intra-frame prediction mode (IPM) during video coding, so as to hide secret messages. The sender hides the secret message in the video carrier by using the video steganography technology, the receiver can obtain the secret smoothly, and the third party will not discover the existence of the secret.

Today's video steganography schemes based on intra-frame prediction generally assume that the impact of modifying each 4x4 prediction block is the same, and reducing the number of modifications as much as possible can effectively reduce video distortion; however, this kind of thinking is likely to lead to actual modification. description is insufficient. Therefore, it is necessary to incorporate new schemes to better describe the overall embedding effect, so as to achieve secret embedding with minimal perturbation.

SUMMARY OF THE INVENTION

In order to solve the problems existing in the existing video steganography technology, the present invention provides a video steganography method based on intra-frame prediction coding. The method uses adaptive steganographic coding technology (STC) for actual embedding after constructing a suitable distortion function. , which ensures that the distortion caused by modifying the prediction mode is nearly minimal.

The present invention adopts the following technical solutions to realize: a video steganography method based on intra-frame prediction coding, comprising the following steps:

The sender and receiver negotiate the mapping rules together before steganographic transmission, and the mapping rules divide all possible prediction modes for 4x4 blocks into 3 groups;

The sender embeds the message: selects the carrier video X, decodes it into the original video frame sequence, and obtains the key frames I ₁ , I ₂ , ...;

Start embedding from the first key frame I ₁ , and let the secret messages embedded in I ₁ be m _1,1 and m _1,2 ; use intra-frame prediction coding on I ₁ to get all pixel blocks that will be coded with intra-frame 4x4 block prediction , and the optimal prediction mode obtained by using intra-frame prediction coding for these pixel blocks; further obtain the optimal group set according to the mapping rule

计算出每个4x4块的加1嵌入扰动和减1嵌入扰动，和计算时所使用的预测模式；Use the optimal set

Calculate the plus-1 embedded perturbation and minus-1 embedded perturbation for each 4x4 block, and the prediction mode used in the calculation;

作为载体向量，秘密消息序列m_1，1和对应的4x4块嵌入扰动集送入自适应编码的第一层嵌入器，得到含秘组集

the optimal set

As a carrier vector, the secret message sequence m ₁ , 1 and the corresponding 4x4 block embedding perturbation set are sent to the first layer embedder of adaptive coding, and the set containing the secret group is obtained.

作为载体向量，秘密消息序列m_1，2和对应的4x4块嵌入扰动集送入自适应编码的第二层嵌入器，得到含秘组集

will contain secret group set

As the carrier vector, the secret message sequence m _{1, 2} and the corresponding 4x4 block embedded perturbation set are sent to the second layer embedder of adaptive coding, and the set containing the secret group is obtained.

选用每个4x4块的最终预测模式，生成视频编码；According to the set of secret groups

Select the final prediction mode of each 4x4 block to generate a video code;

Embed messages for the next key frame until the embedding of all secret messages is completed, or all key frames are used up; use the standard encoder to complete the rest of the video encoding, and finally generate the encrypted video X';

The receiver extracts the message: uses the decoder to obtain the key frames I′ ₁ , I′ ₂ , .

Start message extraction from the first key frame I' ₁ , use the decoder to obtain the prediction mode of each 4x4 block in I ₁ ; further obtain the dense group set according to the mapping rule

作为含密向量，送入自适应编码的第一层提取器，得到秘密消息序列m_1，1，送入自适应编码的第二层提取器，得到秘密消息序列m_1，2；Dense set

As a dense vector, it is sent to the first layer extractor of adaptive coding to obtain the secret message sequence m _1,1 , and sent to the second layer extractor of adaptive coding to obtain the secret message sequence m _1,2 ;

Message extraction is performed on the next keyframe until all secret messages have been extracted, or all keyframes have been used.

π(i)∈{0，1，...，8}；所述最优组集

第一层嵌入器得到的含秘组集

第二层嵌入器得到的含秘组集

Preferably, the prediction mode is md _π(1) , md _π(2) , ...,

π(i)∈{0,1,...,8}; the optimal set

The set of secret groups obtained by the first layer embedder

The secret group set obtained by the second layer embedder

Preferably, in the first layer of the embedder, the embedding perturbation of each 4x4 block is:

为每个4x4块的加1嵌入扰动和减1嵌入扰动。in

Add 1 and subtract 1 embedding perturbation for each 4x4 block.

Preferably, in the second layer embedder, the embedding perturbation of each 4x4 block is:

为每个4x4块的加1嵌入扰动和减1嵌入扰动。in

Add 1 and subtract 1 embedding perturbation for each 4x4 block.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. The present invention utilizes adaptive steganographic coding (STC) to design a steganographic scheme, and defines a distortion function based on an intra-frame prediction mode; after constructing a suitable distortion function, the adaptive steganographic coding technology (STC) is used to carry out actual embedding, This ensures that the distortion caused by modifying the prediction mode is close to minimal.

As a video steganography technology based on the intra prediction mode, the present invention can better preserve the video quality, while the existing IPM-based method only embeds information for the appropriate 4×4 intra-frame prediction mode, and the non-optimal selection rules are serious The statistical properties of IPM are destroyed, making the steganography method based on IPM calibration (IPMC) easy to be detected by steganalysis tools.

2. A mapping rule is introduced to expand the range of selectable modes of each block, so as to increase the selection range of modified prediction modes and further reduce the distortion caused by modification. This ensures that the optimality of the IPM is maintained to the maximum at a given embedding rate during the embedding process, thus effectively resisting the detection of steganalysis methods based on IPM calibration (IPMC).

3. This method embeds information in the intra-frame prediction link, so the embedding capacity is higher and more information can be transmitted. This method uses the video compression quality to design the embedding position, and combines the mainstream steganographic coding methods, so the security is higher. The method can provide a high-capacity, high-security-level video steganography function, and can be used in secret information communication disguised as a video application, such as video on demand, video recording, video cloud storage, video recommendation, etc. Embedding video data to realize reliable and undetected transmission of secret information has good practical application value.

Description of drawings

Fig. 1 is a schematic diagram of 9 intra-frame prediction modes selected by the embedding scheme;

Fig. 2 is a flow chart of message embedding;

Figure 3 is a flow chart of message extraction.

Detailed ways

The present invention will be described in further detail below with reference to the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

The invention combines adaptive steganographic coding (STC) and mapping rules to design a steganographic scheme, constructs a perturbation function to describe the impact of embedding, adopts adaptive steganographic coding technology (STC) to achieve embedding, and combines mapping rules to further Minimize disturbance effects. Since the addition of the sum of absolute errors (SAD) leads to more frequent occurrence of the prediction mode recovery phenomenon, we consider the SAD prediction bias (SPD) when defining the perturbation function. Using the sum of absolute errors (SAD) of the prediction modes as a key factor to construct the perturbation function, the embedding scheme should traverse the entire mode space as much as possible to find an appropriate embedding mode, so here we introduce the difference between prediction modes and mode groups Mapping rules to complete the traversal; finally STC is used to implement the actual embedding. Furthermore, mapping rules are constructed using prediction modes and prediction groups to expand the range of selectable modes for each 4x4 block. The present invention can effectively maintain the predicted optimal characteristics of the video, so as to prevent the existence of secrets from being discovered by the calibration-based steganalysis method.

The present invention is provided with a sender and a receiver, and before the steganographic transmission, they jointly negotiate the mapping rules. The mapping rules are as follows: The intra-frame prediction coding of H.264 divides the key frame into 4x4 and 16x16 blocks, and the mapping rules divide the prediction modes that may be used for all 4x4 blocks into 3 groups, as follows:

gp ₀ ={md ₀ , md ₁ , md ₄ }

gp ₁ ={md ₃ , md ₅ , md ₈ }

gp ₂ ={md ₂ , md ₆ , md ₇ }

Where md _i represents the prediction mode i, and the prediction scan order corresponding to each prediction mode is shown in FIG. 1 . The mapping rule is a better grouping method obtained by analyzing a large number of video frame sequences.

As shown in Figure 1, among the 9 prediction modes, the gray part represents a 4x4 prediction block. During the intra prediction process, the 4x4 prediction block performs prediction values according to its left and upper neighbor luminance units, The prediction mode determines the direction of the value of the prediction block, and the meaning of each prediction mode is shown in Table 1.

Table 1. Nine prediction modes used by the scheme

In No. 0 of FIG. 1 , when the prediction mode 0 (Vertical) is used for the value, the luminance unit of each column in the 4×4 prediction block is equal to the luminance value of the adjacent side in the vertical direction of the column. The more special prediction mode is the average prediction (DC) in Figure 2, and the value of the prediction block is equal to the adjacent edge (A+B+C+D+I+J+K+L+M)/7.

If during intra prediction, the original optimal prediction mode is md _i , but we use mode md _j , the SAD prediction bias (SPD) resulting from this mode replacement is:

SPD _i,j =|SAD(md _i )-SAD(md _j )|

SAD(md _i ) is the predicted SAD generated using mode md _i .

实验结果如表2所示，该表显示了用不同候选模式代替最优模式产生的SPD。We select several video sequences as experimental sequences and observe the resulting SPD when the optimal prediction mode is replaced by the other 8 modes. Assuming that the total number of 4x4 prediction modes with the best mode md ₀ is K, the average SPD of each 4x4 prediction mode is

The experimental results are shown in Table 2, which shows the SPD produced by substituting different candidate modes for the optimal mode.

Since the patterns in the same group are not replaced with each other during steganography, we group the patterns with larger SPD values into the same group to reduce the overall embedded perturbation value. For example, as can be seen from Table 2, replacing the prediction mode md ₀ in the sequence I frame with the remaining 8 candidate prediction modes, the resulting average SPD is shown in the first row, where md ₀ is replaced by the average generated by md ₁ The SPD value is the largest, so md ₀ and md ₁ are placed in the same group. And so on, finally gp ₀ ={md ₀ , md ₁ , md ₄ }.

Table 2. Average SPD per 4x4 prediction mode

The message embedding process of the sender is shown in Figure 2, and the specific process is as follows:

Step 1: Select the carrier video X, decode it into the original video frame sequence, and obtain the key frames I ₁ , I ₂ , . . .

以及这些像素块使用标准H.264的帧内预测编码得到的最优预测模式md_π(1)，md_π(2)，...，

π(i)∈{0，1，…，8}。进一步根据映射规则得到最优组集

Step 2: Embedding from the first keyframe I ₁ . Assume that the secret messages embedded in I ₁ are m _1,1 and m _1,2 . Use H.264 intra prediction coding for I ₁ , get all pixel blocks x ₁ ,...,

and the optimal prediction modes md _π(1) , md _π(2) , ...

π(i)∈{0,1,…,8}. Further obtain the optimal group set according to the mapping rule

计算出每个I4B块的加1嵌入扰动和减1嵌入扰动

和计算时所使用的预测模式

Step 3: Use the optimal group set

Calculate the plus-1 embedded perturbation and minus-1 embedded perturbation for each I4B block

and the prediction mode used in the calculation

subject tomd _j ∈ gp _k , k=[g(i)+1]mod 3

subject tomd _j ∈ gp _k , k=[g(i)-1]mod 3

where SAD(md _i ) represents the sum of absolute errors generated when the prediction mode md _i is used.

Step 4: Set the embedding perturbation for each I4B block as:

作为载体向量，秘密消息序列m_1，1和对应的嵌入扰动集{ρ_i，，i＝1，...，N₁}送入自适应编码±1 STC的第一层嵌入器，得到含秘组集

the optimal set

As a vector vector, the secret message sequence m ₁ , 1 and the corresponding embedded perturbation set {ρ _i ,, i= ₁ , . Secret group set

Step 5: Set the embedding perturbation for each I4B block as:

作为载体向量，秘密消息序列m_1，2和对应的嵌入扰动集{ρ_i，，i＝1，...，N₁}送入自适应编码±1 STC的第二层嵌入器，得到含秘组集

will contain secret group set

As a carrier vector, the secret message sequence m _{1, 2} and the corresponding embedded perturbation set {ρ _i ,, i= ₁ , . Secret group set

选用每个I4B块的最终预测模式，生成H.264编码。选用规则为：Step 6: According to the secret group set

The final prediction mode of each I4B block is selected to generate the H.264 code. The selection rules are:

Step 7: Repeat steps 2-6 to embed messages for the next key frame until all secret messages are embedded, or all key frames are used up. The rest of the video encoding is done using the standard H.264 encoder. Finally, a dense H.264 encoded video X' is generated.

The message extraction process of the receiver is shown in Figure 3, and the detailed steps are as follows:

Step 1: Using the H.264 decoder, obtain the intra-prediction-coded key frames I' ₁ , I' ₂ , .

π(i)∈{0，1，…，8}。进一步根据映射规则得到含密组集

Step 2: Start message extraction from the _first key frame I'1. Using the H.264 decoder to obtain the prediction mode md _π(1) , md _π(2) , ..., of each I4B block in I ₁

π(i)∈{0,1,…,8}. Further obtain the dense set according to the mapping rule

作为含密向量，送入自适应编码±1STC的第一层提取器，得到秘密消息序列m_1，1，送入自适应编码±1STC的第二层提取器，得到秘密消息序列m_1，2。Step 3: Set the dense group

As a dense vector, it is sent to the first layer extractor of adaptive coding ±1STC to obtain the secret message sequence m _1,1 , and sent to the second layer extractor of adaptive coding ±1STC to obtain the secret message sequence m _1,2 .

Step 4: Repeat steps 2-3 to extract messages for the next key frame until all secret messages are extracted, or all key frames are used up.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above-mentioned embodiments, and any other changes, modifications, substitutions, combinations, The simplification should be equivalent replacement manners, which are all included in the protection scope of the present invention.

Claims (1)

1. the video steganography method based on intra-frame prediction coding, is characterized in that, comprises the following steps: The sender and receiver negotiate the mapping rules together before steganographic transmission, and the mapping rules divide all possible prediction modes for 4x4 blocks into 3 groups; The sender embeds the message: selects the carrier video X, decodes it into the original video frame sequence, and obtains the key frames I ₁ , I ₂ , ...; Start embedding from the first key frame I ₁ , and let the secret messages embedded in I ₁ be m _1,1 and m _1,2 ; use intra-frame prediction coding on I ₁ to get all pixel blocks that will be coded with intra-frame 4x4 block prediction , and the optimal prediction mode obtained by using intra-frame prediction coding for these pixel blocks; further obtain the optimal group set according to the mapping rule

Use the optimal set

Calculate the plus-1 embedded perturbation and minus-1 embedded perturbation for each 4x4 block, and the prediction mode used in the calculation; the optimal set

As a carrier vector, the secret message sequence m ₁ , 1 and the corresponding 4x4 block embedding perturbation set are sent to the first layer embedder of adaptive coding, and the set containing the secret group is obtained.

will contain secret group set

As the carrier vector, the secret message sequence m _{1, 2} and the corresponding 4x4 block embedded perturbation set are sent to the second layer embedder of adaptive coding, and the set containing the secret group is obtained.

According to the set of secret groups

Select the final prediction mode of each 4x4 block to generate a video code; Embed messages for the next key frame until the embedding of all secret messages is completed, or all key frames are used up; use the standard encoder to complete the rest of the video encoding, and finally generate the encrypted video X'; The receiver extracts the message: uses the decoder to obtain the key frames I′ ₁ , I′ ₂ , . Start message extraction from the first key frame I' ₁ , use the decoder to obtain the prediction mode of each 4x4 block in I ₁ ; further obtain the dense group set according to the mapping rule

Dense set

As a dense vector, it is sent to the first layer extractor of adaptive coding to obtain the secret message sequence m _1,1 , and sent to the second layer extractor of adaptive coding to obtain the secret message sequence m _1,2 ; Perform message extraction on the next key frame until all secret messages are extracted, or all key frames are used; The prediction mode used in the calculation

for:

subject tomd _j ∈ gp _k , k=[g(i)+1]mod 3

subject tomd _j ∈ gp _k , k=[g(i)-1]mod 3 where SAD(md _i ) represents the sum of absolute errors generated when the prediction mode md _i is used; The prediction mode is

md _π(i) is the optimal prediction mode obtained by using the standard H.264 intra-frame prediction coding for the i-th I4B block, π(i)∈{0,1,...,8}; the optimal group set

The set of secret groups obtained by the first layer embedder

The secret group set obtained by the second layer embedder

According to the set of secret groups

The final prediction mode of each I4B block is selected to generate H.264 encoding; the selection rules are:

In the first layer of the embedder, the embedding perturbation for each 4x4 block is:

in

plus-1 and minus-1 embedding perturbations for each 4x4 block;

In the second layer of the embedder, the embedding perturbation for each 4x4 block is:

in

plus-1 and minus-1 embedding perturbations for each 4x4 block; The 3 groups of prediction modes are: gp ₀ ={md ₀ , md ₁ , md ₄ } gp ₁ ={md ₃ , md ₅ , md ₈ } gp ₂ ={md ₂ , md ₆ , md ₇ } Among them, md ₀ is the vertical prediction, md ₁ is the horizontal prediction, md ₂ is the average prediction, md ₃ is the diagonal prediction in the lower left direction, md ₄ is the diagonal prediction in the lower right direction, and md ₅ is the vertical right prediction, md ₆ is a horizontal downward prediction, md ₇ is a vertical left prediction, and md ₈ is a horizontal upward prediction; If during intra prediction, the original optimal prediction mode is md _i , but mode md _j is used, the SAD prediction bias (SPD) generated by this mode replacement is: SPD _i,j =|SAD(md _i )-SAD(md _j )| SAD(md _i ) is the predicted SAD generated using mode md _i ; select several video sequences as experimental sequences, and observe the SPD generated when the optimal prediction mode is replaced by the other 8 modes; the mode with larger SPD value is divided into the same group to reduce the overall embedded perturbation value.

Images