Multi-image steganography method based on two-dimensional compressed sensing and maximum energy block

Abstract

The invention discloses a novel multi-image steganography, and belongs to the field of information security. According to the method, an optimal hidden area of a carrier is identified by establishing a concept of a maximum energy block, so that a plurality of secret images are hidden under the condition of almost not changing visual perception. This concept is based on the fact that a human vision system is insensitive to high energy regions and complex edges as well as gray scale variations of the texture. In order to further minimize the visual influence of a secret image on a carrier and expand an available steganography space, a video is adopted as the carrier, and discrete wavelet transform is utilized. A high-frequency component of a highest energy block in a video frame is screened to serve as a final hiding space, meanwhile, two-dimensional compressed sensing is used for compressing a plurality of secret images, and the secret images are embedded into specified channels of different video frames according to a specially-designed mapping relation, so that the amount of data needing to be hidden is reduced, and the steganography capacity is improved. The secret image can be accurately extracted and reconstructed based on the correct secret key in the inverse process.

Description

Technical Field

[0001] The present invention belongs to the field of image steganography for information security, and more specifically, is a multi-image steganography method based on two-dimensional compressed sensing and maximum energy blocks. Background Art

[0002] In the field of information security, image steganography is widely used in military, medical and commercial fields as an important means to protect the security of information transmission. Traditional image steganography is mainly divided into spatial domain methods (such as least significant bit LSB embedding) and frequency domain methods (such as discrete cosine transform DCT, discrete wavelet transform DWT, etc.). Although these methods can achieve information hiding, they will significantly reduce the visual quality of the carrier image when embedding a large amount of data. In recent years, although multi-image steganography has improved the security and capacity of data hiding, the existing technology still has limitations. Compressed sensing technology has been introduced into steganography, which improves the concealment and security of steganography by reducing the amount of data and increasing the hidden capacity. However, the existing steganography methods based on compressed sensing still have shortcomings, such as the quality of the carrier image decreases when the number of secret images increases. Therefore, a steganography method is needed to improve the steganography capacity and security while ensuring the visual quality of the carrier image. Summary of the invention

[0003] The present invention aims to improve the steganographic capacity and security while ensuring the visual quality of the carrier image. A new image steganographic method based on two-dimensional compressed sensing and maximum energy block selection strategy is proposed.

[0004] The method of the present invention comprises two stages: embedding a secret image and extracting and reconstructing a secret image, and specifically comprises the following steps: (1) Image embedding stage: The algorithm first uses two measurement matrices to compress multiple secret images to reduce the amount of data and facilitate subsequent hiding. At the same time, the secret key is used to establish a mapping relationship between the secret image and the video frame to determine the video frame position where each secret image should be embedded. Subsequently, the energy score map of the video frame is obtained based on the predefined block area size, step size and energy scoring strategy, and the block with the highest energy is selected from the video frame as the embedding position. The selected blocks are subjected to discrete wavelet transform, and the compressed secret image is embedded into the high-frequency coefficients of these blocks to generate a steganographic block. The steganographic block is reintegrated back into the selected channel through inverse discrete wavelet transform to form a steganographic channel, which is combined with other channels to form a steganographic frame. Finally, the original video frame is replaced by the steganographic frame to generate the final steganographic video. The entire embedding process ensures the hiding of secret information while maintaining the quality of the video. The above two measurement matrices and the secret key are used as decryption keys. Only the receiver who has the key can extract and restore the secret image. (2) Image extraction and reconstruction phase: The algorithm uses the stego video, secret key, and measurement matrix as input. First, the corresponding stego frame is extracted according to the key and the stego channel is obtained. Then, the stego blocks are extracted according to the score map, and these blocks are subjected to discrete wavelet transform to extract the embedded compressed secret image from the high-frequency coefficients. Subsequently, the secret image is restored through the reconstruction algorithm using the measurement matrix. Finally, multiple secret images are recovered.

[0005] The specific implementation process of the above step (1) is as follows: (1a) Generate a pair of measurement matrices and , perform two-dimensional compressed sensing on each image to obtain the compressed corresponding image. At the same time, given a secret key value, the hash value of each image is calculated in the following way, and then the number of frames and channels required to hide each image in the video frame are calculated according to the hash value of the image. , (1) in is an image The hash value of This is the initial key. , (2) , (3) in is the total number of video frames. (1b) Find the corresponding video frame based on the obtained mapping relationship, use the proposed maximum energy block selection strategy according to the predefined block area size and step size to obtain the energy distribution of the corresponding video frame, and select the block with the largest energy in the video frame as the embedding position. The detailed selection strategy is defined as follows: , (4) in , and It is used to control the weight of each feature. is the energy score of the block, is the score of the edge feature, is the score of the texture feature. The scores of the three are defined as follows: , (5) in , is the coordinate in the video frame The gradient amplitude of the pixel, is the total number of pixels in the video frame. , (6) in , , are the edge detection scores for each channel. They are obtained using the Canny operator. , , Represents the weight of each channel. , (7) The features of the gray-level co-occurrence matrix are used to derive a texture score for the region. , Used to control the weights of two features. The formulas for the two features are as follows: , (8) , (9) in , Represents the grayscale value in the image. It means the gray value is The pixel and gray value are The probability that pixels of appear at the same time in a specific direction and distance. (1c) DWT operation is performed on the selected blocks to embed the compressed secret image into the high-frequency coefficients of these blocks. (1d) After embedding is completed, an inverse discrete wavelet transform is performed, and the modified block is integrated back into the selected channel to obtain the steganographic channel, which is then combined with other channels to generate a steganographic frame, and finally the steganographic frame is used to replace the original frame to construct the steganographic video.

[0006] The specific implementation process of the above step (2) is as follows: (2a) Extract the corresponding steganographic frame according to the key and obtain the steganographic channel. Then, extract the corresponding steganographic blocks according to the score map, perform discrete wavelet transform on these blocks, and extract the embedded compressed secret image from the high-frequency coefficients. (2b) Using the measurement matrix, the secret image is recovered through the reconstruction algorithm. During the reconstruction process, key parameters (such as error tolerance and maximum number of iterations) are initialized to control the convergence and performance of the algorithm. By performing an initial solution estimate, the solution is optimized using gradient descent and bivariate shrinkage methods, and the energy difference is calculated. By judging whether the iteration has converged, the iteration is stopped when the difference between the current solution and the previous solution is less than the set tolerance or when the maximum number of iterations is reached. The wavelet coefficients are thresholded in the wavelet domain, and finally a high-quality reconstructed image is obtained through iterative optimization and regularization.

[0007] Compared with the prior art, the advantages of the present invention are as follows: (1) By using two-dimensional compressed sensing technology to compress the secret image data, more secret information can be embedded in a limited carrier space, significantly improving the information hiding capacity of image steganography. (2) The maximum energy block selection strategy is used to hide the secret information in the high-frequency components of the carrier image with the maximum energy. These areas have less impact on visual perception and are more difficult to be detected by steganalysis tools, thereby enhancing the concealment and security of the steganography. (3) Using video frames as carriers and establishing a dedicated mapping relationship between secret images and carriers not only provides more abundant carrier choices for steganography, but also increases the flexibility and practicality of steganography, and can better adapt to different application scenarios and needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Attached Figure 1 The figure is a schematic diagram of the image embedding process of the method of the present invention.

[0009] Attached Figure 2 The figure is a flow chart of the image extraction and reconstruction phase of the method of the present invention.

[0010] Attached Figure 3 (a) to 3(b) are respectively visual effect diagrams before and after embedding the video frame in the embodiment of the present invention.

[0011] Attached Figure 4 (a) to Figure 4 (d) are four secret grayscale images reconstructed in the embodiments of the present invention. DETAILED DESCRIPTION

[0012] The present invention is further described below with reference to the accompanying drawings, and a specific embodiment of the present invention is described.

[0013] In this embodiment, a 256×256 grayscale secret image is selected from the database of the image processing site to implement the proposed steganography method. A 107-frame cover video is selected as a steganographic carrier, which includes the embedding and extraction of the secret image, which are respectively realized through the steganalysis embedding stage and the image extraction and reconstruction stage.

[0014] The entire embedding and extraction stage can be implemented as follows: (1) Embedding process: Generate a pair of 192×192 measurement matrices based on a compression ratio of 0.75 and , according to the set key Key, use formula (1) to calculate the hash value of each image, and then calculate the frame number and channel of the corresponding video of each image according to formulas (2) and (3). Find the corresponding steganographic channel, and calculate the energy distribution of the frame in the form of a sliding window according to formula (4) according to the set block area size and step size, where the three feature weights are 0.3, 0.3, and 0.4 respectively. Select the block with the largest energy in the frame to perform the DWT operation, embed the compressed secret image into the high-frequency coefficients of the block, and then perform an inverse discrete wavelet transform to integrate the modified block back into the selected channel to obtain the steganographic channel, combine it with other channels to generate a steganographic frame, and finally replace the original frame with the steganographic frame to construct the steganographic video. (2) Extraction process: according to Figure 2 Specific steps: S1. Input the steganographic video. S2. Extract the video frame containing the secret information and the corresponding channel according to the key. S3. Find the maximum energy block according to the energy distribution diagram of the frame. S4. Perform DWT operation on the steganographic block to obtain a compressed image. S5. Use the measurement matrix to reconstruct the secret image through the reconstruction algorithm.

[0015] like Figure 3 The following are visual effects before and after embedding the secret image into the video frame according to the embodiment of the present invention. It can be seen that the visual changes before and after embedding the secret image into the video frame are very small, and generally speaking, it maintains good visual and structural quality. Figure 4 The results of the reconstructed four secret grayscale images are shown. This clearly shows that the reconstruction effect is good. The experiment verifies that the method of the present invention effectively utilizes the energy concentration area in each frame as a carrier of hidden information by using video frames. Through a carefully designed algorithm, it can achieve high-precision hiding and high-quality reconstruction of secret information. The experimental results show that the method excels in imperceptibility and reconstructed image quality, meeting the needs of the modern field of information hiding. Compared with previous methods, the proposed scheme shows that when hiding a large amount of data, the scheme can better preserve the original structure and visual quality of the video, and make more precise and reasonable modifications to the video.

Claims

1. A multi-image steganography method based on two-dimensional compressed sensing and maximum energy block, characterized in that: The following steps are involved: (1) Image embedding stage: The algorithm first uses two measurement matrices to compress multiple secret images to reduce the amount of data and facilitate subsequent hiding. At the same time, the secret key is used to establish a mapping relationship between the secret image and the video frame to determine the video frame position where each secret image should be embedded. Then, the energy score map of the video frame is obtained based on the predefined block area size, step size, and energy scoring strategy, and the block with the highest energy is selected from the video frame as the embedding position. The selected blocks are subjected to discrete wavelet transform, and the compressed secret image is embedded into the high-frequency coefficients of these blocks to generate a steganographic block; the steganographic block is reintegrated back into the selected channel through inverse discrete wavelet transform to form a steganographic channel, and it is combined with other channels to form a steganographic frame; finally, the original video frame is replaced by the steganographic frame to generate the final steganographic video; the entire embedding process ensures the hiding of secret information while maintaining the quality of the video. The above two measurement matrices and the secret key are used as decryption keys. Only the receiver who has the key can extract and restore the secret image; (2) Image extraction and reconstruction stage: The algorithm uses the stego video, secret key, and measurement matrix as input. First, the corresponding stego frame is extracted according to the key and the stego channel is obtained. Then, the stego blocks are extracted according to the score map, and discrete wavelet transform is performed on these blocks to extract the embedded compressed secret image from the high-frequency coefficients. Subsequently, the secret image is recovered through a reconstruction algorithm using the measurement matrix; Eventually, multiple secret images were recovered.

2. According to claim 1, a multi-image steganography method based on two-dimensional compressed sensing and maximum energy block is characterized in that A new method for finding embedded regions is proposed to calculate the energy score of the video frame. The specific implementation process of step (1) is as follows: (1a) Generate a pair of measurement matrices and , perform two-dimensional compressed sensing on each image to obtain the compressed corresponding image; at the same time, given a secret key value, use the following method to calculate the hash value of each image, and then calculate the number of frames and channels required to hide each image in the video frame according to the hash value of each image; in is an image The hash value of is the initial set key; in is the total number of video frames; (1b) Find the corresponding video frame according to the obtained mapping relationship, and use the proposed maximum energy block selection strategy to obtain the energy distribution of the corresponding video frame according to the predefined block area size and step size, and select the block with the largest energy in the video frame as the embedding position. The detailed selection strategy is defined as follows: in , and It is used to control the weight of each feature. is the energy score of the block, is the score of the edge feature, is the score of the texture feature; the scores of the three are defined as follows: in , is the coordinate in the video frame The gradient amplitude of the pixel, is the total number of pixels in the video frame; in , , are the edge detection scores for each channel; they are obtained using the Canny operator; and , , Represents the weight of each channel; The features of the gray-level co-occurrence matrix are used to derive a texture score for the region. , Used to control the weights of two features. The formulas for the two features are as follows: in , Represents the grayscale value in the image; It means the gray value is The pixel and gray value are The probability that pixels of appear at the same time in a specific direction and distance; (1c) Performing DWT operations on the selected blocks to embed the compressed secret image into the high-frequency coefficients of these blocks; (1d) After embedding is completed, an inverse discrete wavelet transform is performed, and the modified block is integrated back into the selected channel to obtain the steganographic channel, which is then combined with other channels to generate a steganographic frame, and finally the steganographic frame is used to replace the original frame to construct the steganographic video.

3. The multi-image steganography method based on two-dimensional compressed sensing and maximum energy block according to claim 1 is characterized in that The secret image is extracted and a high-quality reconstructed image is obtained. The specific implementation of step (2) is as follows: (2a) Extract the corresponding steganographic frame according to the key and obtain the steganographic channel; then, extract the corresponding steganographic blocks according to the score map, and perform discrete wavelet transform on these blocks to extract the embedded compressed secret image from the high-frequency coefficients; (2b) Recovering the secret image through a reconstruction algorithm using the measurement matrix; initializing key parameters (such as error tolerance and maximum number of iterations) during the reconstruction process to control the convergence and performance of the algorithm; optimizing the solution using gradient descent and bivariate shrinkage methods by performing an initial solution estimate, and calculating the energy difference; By judging whether the iteration has converged, the iteration is stopped when the difference between the current solution and the previous solution is less than the set tolerance or when the maximum number of iterations is reached; the wavelet coefficients are thresholded in the wavelet domain, and finally a high-quality reconstructed image is obtained through iterative optimization and regularization processing.

Images