Large-capacity image steganography method and system based on multi-stage learning and medium

Abstract

The invention discloses a high-capacity image steganography method and system based on multi-stage learning and a medium. The method comprises the following steps: acquiring an image to be processed; and inputting the to-be-processed image into a pre-trained multi-stage learning high-capacity depth image steganography model, and obtaining a target image output by the multi-stage learning high-capacity depth image steganography model. From two perspectives of contrast refinement and multi-scale adjustment, a multi-stage adjustment high-capacity image steganography network is designed, and by designing an optimization stage from coarse to fine in a hiding process, multi-stage iterative optimization is realized, and the modeling capability of a model in a high-capacity steganography condition is improved.

Description

Technical Field

[0001] The present invention relates to a large-capacity image steganography method, system and medium based on multi-stage learning, and belongs to the technical field of image processing. Background Art

[0002] In recent years, with the rapid development of the internet and multimedia technologies, the field of information technology has experienced unprecedented rapid iteration. However, this has also been accompanied by the increasing prominence of information security issues. These issues are not only emerging in an endless stream but also becoming increasingly complex and dynamic, placing more stringent and comprehensive requirements on information security protection. Image steganography, as an effective means of protecting information security, can achieve covert communication of secret information in an imperceptible manner. It has a wide range of applications in defense, medicine, and commerce, and is of vital importance in the theoretical research and practical application of information security.

[0003] For image steganography, the size of the secret information transmitted in a single transmission is the stegocapacity, which represents the efficiency of image steganography. As the stegocapacity increases, the difficulty of image steganography also increases, making it difficult to guarantee the concealment and recovery effectiveness of the stegocapacity. Since the 21st century, with the rapid development of the internet and big data technologies, the primary medium for information transmission has gradually evolved from text alone to multimedia forms including images. This shift has placed higher demands on the efficiency of covert communication. Therefore, research on large-capacity image steganography methods is of great significance for the practical application of image steganography technology.

[0004] Traditional image steganography methods use traditional machine learning methods such as least significant bit (LSB) or content-adaptive steganography to hide binary secret information. Due to limitations in modeling capabilities, these methods have very limited steganographic capabilities, with a stegocapacity typically ranging from 0.1 to 0.4 bits per pixel. For a carrier image of 128 × 128 pixels and a stegocapacity of 0.4 bpp, the hidden secret information is 128 × 128 × 0.4 = 6,553.6 bits. To hide a complete secret image of the same size within a single carrier image, each carrier pixel must contain the information of an entire pixel, or 24 bpp (8 × (R,G,B) bpp). This results in a transmitted secret information of 128 × 128 × 24 = 393,216 bits. Traditional image steganography methods struggle to generate high-quality stegocapacities and recover secret images with such a large stegocapacity.

[0005] Leveraging the powerful representation capabilities of convolutional neural networks, the field of image steganography has rapidly advanced, shifting the type of secret data from binary secret information to secret images. Baluja et al. first proposed hiding an RGB color secret image within a single carrier image. This method uses a hidden network to encode the preprocessed secret information and the carrier image, generating a secret image. A universal recovery network then reconstructs the recovered secret image from the secret image, enabling covert transmission of the secret image. Unlike Deep-Stego, the UDH method encodes the secret image through a hidden network and then directly adds the encoded secret information to the original carrier image to generate the secret image. Leveraging the reversible nature of reversible neural networks, ISN utilizes the forward and backward flows of reversible neural networks to hide and recover secret image information, respectively. Similarly, HiNet and DeepMIH employ reversible neural networks to hide and recover secret images in the wavelet domain. However, these image steganography methods have not performed well in large-scale steganographic scenarios involving hiding multiple secret images. This situation can be attributed to the fact that the secret hiding process adopted by the above methods is a single stage, resulting in insufficient feature learning of the carrier and secret information, and lack of necessary refinement steps, thus limiting the further improvement of steganalysis concealment and recovery effectiveness when dealing with large-scale steganalysis tasks;

[0006] It can be seen that in order to solve the above technical problems, there is an urgent need for a large-capacity image steganography method, system and medium based on multi-stage learning. Summary of the Invention

[0007] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a large-capacity image steganography method, system and medium based on multi-stage learning. To address the problem of insufficient feature learning in the large-capacity secret hiding process, a multi-stage adjustment large-capacity image steganography network is designed from the two perspectives of contrast refinement and multi-scale adjustment. By designing a coarse-to-fine optimization stage in the hiding process, multi-stage iterative optimization is achieved, thereby improving the modeling ability of the model in large-capacity steganography situations.

[0008] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0009] In a first aspect, the present invention provides a large-capacity image steganography method based on multi-stage learning, comprising:

[0010] Get the image to be processed;

[0011] The image to be processed is input into a pre-trained multi-stage learning large-capacity deep image steganography model, and the following processing is performed:

[0012] Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales;

[0013] The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

[0014] Furthermore, the types of the images to be processed include carrier images, secret images and secret-carrying images;

[0015] When the image to be processed is a carrier image and a secret image, a hiding process is performed;

[0016] When the image to be processed is a confidential image, restoration processing is performed.

[0017] Furthermore, during the hidden processing, multi-scale input pairs are obtained from the image to be processed through downsampling, and two "1 × 1" convolutional layers are used to transform the number of channels of the half-scale and quarter-scale input image pairs in the multi-scale input pairs;

[0018] According to the number of iterative optimizations, the transformed multi-scale input pairs are subjected to basic encryption processing or iterative encryption processing to obtain encrypted images of different scales;

[0019] When the number of iterations is 1, basic encryption processing is performed. When the number of iterations is greater than 1, iterative encryption processing is performed.

[0020] In the basic encryption process, the target image is generated through the coarse hidden network;

[0021] During the iterative encryption process, a full-scale initial encrypted image is first generated based on the coarse hidden network. The intermediate recovered encrypted image and secret image are then generated based on the initial encrypted image through the encrypted image recovery network and the secret image recovery network. Difference pairs are obtained based on the input pairs and the intermediate recovered pairs. The initial encrypted image is iteratively adjusted through the difference pairs to obtain encrypted images of different scales.

[0022] During the recovery process, the recovered secret graphs of different scales are obtained from the input secret graph through the secret graph refinement recovery network.

[0023] Furthermore, in the generation of the target image through the coarse hidden network, the hiding process is expressed by the following formula:

[0024] (1);

[0025] Where, 、 and They are the final full-scale, 1 / 2-scale and 1 / 4-scale density maps, is the coarse hidden network, is the carrier image, For the secret picture, , , and They are and The height and width of and The number of channels; and is a 1 / 2 scale vector-secret graph input pair, , ,; and is a 1 / 4 scale vector-secret graph input pair, , ;

[0026] In the iterative encryption process, a coarse hidden network is used to perform basic encryption to obtain an initial encrypted image. The initial encrypted image is the final full-scale encrypted image obtained in the basic encryption process. , the generation process of the initial encrypted image is expressed as:

[0027] (2);

[0028] exist Iterative process, recovering the network from the carrier graph and secret graph recovery network The intermediate recovered carrier map is generated based on the initial carrier map and secret map , by input vector-secret graph pair and intermediate recovery carrier-secret graph pair Calculate the difference between and , the calculation process is as follows:

[0029] (3);

[0030] The secret image is recovered by the difference carrier-secret image pair in the subsequent iteration process. Continuously refine and adjust The encrypted map generated during the iteration process Obtained by the following formula:

[0031] (4);

[0032] Through formulas (3) and (4), the encrypted image is iteratively adjusted and optimized until ,when hour, and Obtained from formula (3);

[0033] Different sizes of secret maps , and By Refining hidden network during iteration According to The difference pairs are obtained in the iterative process, and the hidden process is expressed as follows:

[0034] (5);

[0035] Where, It is a full-size encrypted map. It is a 1 / 2 size dense map. It is a 1 / 4 size encrypted image;

[0036] The method of obtaining recovered secret graphs of different scales from the input secret graph through the secret graph refinement recovery network includes:

[0037] (6);

[0038] Where, For the full-size secret recovery map, The secret recovery image is 1 / 2 size. The secret recovery image is 1 / 4 size. Refine the recovery network for the secret graph, It is a full-size confidential map.

[0039] Furthermore, the secret image restoration network and carrier image restoration network The convolution kernel size of the middle convolution layer is set to 3×3, and the number of channels of the convolution layer is set to {( , 64), (64, 128), (128, 256),(256,128), (128, 64), (64, )};

[0040] The coarse hidden network , refine the hidden network and image thinning and restoration network The number of channels in the convolutional layer is set to {( , 64), (64, 128),(128, 256), (256, 128), (256,64), (128, )};

[0041] The coarse hidden network , refine the hidden network and image thinning and restoration network The convolution kernel size of the middle convolution layer and the transposed convolution layer is set to 4 × 4.

[0042] Furthermore, the multi-stage optimization of the secret-carrying graphs or secret recovery graphs of different scales according to the multi-scale information includes:

[0043] Different loss functions are selected based on multi-scale information to optimize the secret-carrying images or secret recovery images of different scales. The optimization goal is to minimize the loss function, where:

[0044] When half-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows:

[0045] (7);

[0046] Where, and represents the hiding and recovery loss weights, To hide the losses, To recover the losses, Represents the loss weight of the scale image used;

[0047] When the intermediate images of the full-scale and half-scale pairs are used to generate the target image, the loss function as follows:

[0048] (8);

[0049] When the full-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows:

[0050] (9).

[0051] In a second aspect, the present invention provides a large-capacity image steganography system based on multi-stage learning, comprising:

[0052] Acquisition unit: used to acquire the image to be processed;

[0053] Processing unit: used to input the image to be processed into a pre-trained multi-stage learning large-capacity deep image steganography model and perform the following processing:

[0054] Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales;

[0055] The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

[0056] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0057] In a fourth aspect, the present invention provides a computer device, comprising:

[0058] Memory, used to store computer programs / instructions;

[0059] A processor, configured to execute the computer program / instructions to implement the steps of any one of the methods described in the first aspect.

[0060] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention provides a large-capacity image steganography method based on multi-stage learning. To address the problem of insufficient feature learning in the large-capacity secret hiding process, a multi-stage adjustment large-capacity image steganography network is designed from the two perspectives of contrast refinement and multi-scale adjustment. By designing a coarse-to-fine optimization stage in the hiding process, multi-stage iterative optimization is achieved, thereby improving the modeling ability of the model in large-capacity steganography situations. A multi-stage optimization module is designed, which utilizes the multi-stage contrast information of carrier-carrier secret and secret-recovery secret graph pairs to iteratively optimize and adjust the generated carrier secret and recovery secret graphs. A multi-stage pyramid optimization module is designed, which directly adjusts and enhances the intermediate feature representation of the hiding and recovery process through the multi-stage and multi-scale information of carrier-carrier secret and secret-recovery secret graph pairs of different scales, thereby improving the modeling ability and steganography performance of the model in large-capacity steganography situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a system framework diagram of a large-capacity image steganography method based on multi-stage learning provided by an embodiment of the present invention;

[0064] Figure 2 This is a detailed diagram of the system framework of a large-capacity image steganography method based on multi-stage learning provided by an embodiment of the present invention;

[0065] Figure 3 Schematic diagram comparing the visualization effects of this method, UDH, and DEEPMIH when a secret image is hidden in a carrier image;

[0066] Figure 4 Schematic diagram comparing the visualization effects of this method with those of UDH and DEEPMIH when two secret images are hidden in one carrier image;

[0067] Figure 5 A schematic diagram comparing the visualization effects of this method and UDH when five secret images are hidden in one carrier image;

[0068] Figure 6 This is a flowchart of a large-capacity image steganography method based on multi-stage learning in Example 1 of the present invention. DETAILED DESCRIPTION

[0069] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. Unless there is a conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0070] The term "and / or" in this document simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. Additionally, the character " / " in this document generally indicates an "or" relationship between the related objects.

[0071] Example 1:

[0072] Figure 6 This is a flowchart of a large-capacity image steganography method based on multi-stage learning in the first embodiment of the present invention. The large-capacity image steganography method based on multi-stage learning provided in this embodiment can be applied to a terminal and can be executed by a large-capacity image steganography system based on multi-stage learning. The system can be implemented in software and / or hardware. The system can be integrated into a terminal, such as any smartphone, tablet computer or computer device with communication functions. Figure 6 , the method of this implementation specifically includes the following steps:

[0073] Get the image to be processed;

[0074] The image to be processed is input into a pre-trained multi-stage learning large-capacity deep image steganography model, and the following processing is performed:

[0075] Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales;

[0076] The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

[0077] The application process of the large-capacity image steganography method based on multi-stage learning provided in this embodiment specifically involves the following steps:

[0078] Furthermore, the types of the images to be processed include carrier images, secret images and secret-carrying images;

[0079] When the image to be processed is a carrier image and a secret image, a hiding process is performed;

[0080] When the image to be processed is a confidential image, restoration processing is performed.

[0081] Furthermore, during the hidden processing, multi-scale input pairs are obtained from the image to be processed through downsampling, and two "1 × 1" convolutional layers are used to transform the number of channels of the half-scale and quarter-scale input image pairs in the multi-scale input pairs;

[0082] According to the number of iterative optimizations, the transformed multi-scale input pairs are subjected to basic encryption processing or iterative encryption processing to obtain encrypted images of different scales;

[0083] When the number of iterations is 1, basic encryption processing is performed. When the number of iterations is greater than 1, iterative encryption processing is performed.

[0084] In the basic encryption process, the target image is generated through the coarse hidden network;

[0085] During the iterative encryption process, a full-scale initial encrypted image is first generated based on the coarse hidden network. The intermediate recovered encrypted image and secret image are then generated based on the initial encrypted image through the encrypted image recovery network and the secret image recovery network. Difference pairs are obtained based on the input pairs and the intermediate recovered pairs. The initial encrypted image is iteratively adjusted through the difference pairs to obtain encrypted images of different scales.

[0086] During the recovery process, the recovered secret graphs of different scales are obtained from the input secret graph through the secret graph refinement recovery network.

[0087] Furthermore, in the generation of the target image through the coarse hidden network, the hiding process is expressed by the following formula:

[0088] (1);

[0089] Where, 、 and They are the final full-scale, 1 / 2-scale and 1 / 4-scale density maps, is the coarse hidden network, is the carrier image, For the secret picture, , , and They are and The height and width of and The number of channels; and is a 1 / 2 scale vector-secret graph input pair, , ,; and is a 1 / 4 scale vector-secret graph input pair, , ;

[0090] In the iterative encryption process, a coarse hidden network is used to perform basic encryption to obtain an initial encrypted image. The initial encrypted image is the final full-scale encrypted image obtained in the basic encryption process. , the generation process of the initial encrypted image is expressed as:

[0091] (2);

[0092] exist Iterative process, recovering the network from the carrier graph and secret graph recovery network The intermediate recovered carrier map is generated based on the initial carrier map and secret map , by input vector-secret graph pair and intermediate recovery carrier-secret graph pair Calculate the difference between and , the calculation process is as follows:

[0093] (3);

[0094] The secret image is recovered by the difference carrier-secret image pair in the subsequent iteration process. Continuously refine and adjust The encrypted map generated during the iteration process Obtained by the following formula:

[0095] (4);

[0096] Through formulas (3) and (4), the encrypted image is iteratively adjusted and optimized until ,when hour, and Obtained from formula (3);

[0097] Different sizes of secret maps , and By Refining hidden network during iteration According to The difference pairs are obtained in the iterative process, and the hidden process is expressed as follows:

[0098] (5);

[0099] Where, It is a full-size encrypted map. It is a 1 / 2 size dense map. It is a 1 / 4 size encrypted image;

[0100] The method of obtaining recovered secret graphs of different scales from the input secret graph through the secret graph refinement recovery network includes:

[0101] (6);

[0102] Where, For the full-size secret recovery map, The secret recovery image is 1 / 2 size. The secret recovery image is 1 / 4 size. Refine the recovery network for the secret graph, It is a full-size confidential map.

[0103] Furthermore, the secret image restoration network and carrier image restoration network The convolution kernel size of the middle convolution layer is set to 3×3, and the number of channels of the convolution layer is set to {( , 64), (64, 128), (128, 256),(256,128), (128, 64), (64, )};

[0104] The coarse hidden network , refine the hidden network and image thinning and restoration network The number of channels in the convolutional layer is set to {( , 64), (64, 128),(128, 256), (256, 128), (256,64), (128, )};

[0105] The coarse hidden network , refine the hidden network and image thinning and restoration network The convolution kernel size of the middle convolution layer and the transposed convolution layer is set to 4 × 4.

[0106] Furthermore, the multi-stage optimization of the secret-carrying graphs or secret recovery graphs of different scales according to the multi-scale information includes:

[0107] Different loss functions are selected based on multi-scale information to optimize the secret-carrying images or secret recovery images of different scales. The optimization goal is to minimize the loss function, where:

[0108] When half-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows:

[0109] (7);

[0110] Where, and represents the hiding and recovery loss weights, To hide the losses, To recover the losses, Represents the loss weight of the scale image used;

[0111] When the intermediate images of the full-scale and half-scale pairs are used to generate the target image, the loss function as follows:

[0112] (8);

[0113] When the full-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows:

[0114] (9).

[0115] The following describes the contents involved in the above embodiment in conjunction with a preferred embodiment.

[0116] The proposed Multi-Process Adjustment Large Capacity Steganography Network (MPALSH-Net) includes a Multi-Process Comparative Optimization (CPO) module and a Multi-Process Pyramidal (CPP) module. It uses a coarse-to-fine strategy to iteratively adjust and optimize the image quality of the secret image and the secret image in terms of feature learning and image scale, respectively. Figure 1 The multi-stage learning large-capacity steganography method proposed in this invention includes five main components, namely the coarse hidden network , refine the hidden network , carrier image restoration network , Secret Graph Recovery Network , Secret Graph Thinning Recovery Network Assume that the proposed multi-stage learning high-capacity steganography method is composed of ( ) iterative optimization process, 、 and Will repeat ( )Second-rate.

[0117] Figure 1 Demonstrated The overall structure of CPCO is an iterative process. Figure 1 In the hidden process, the input is the carrier image and secret map . and They are and The height and width of the and The number of channels is obtained by downsampling pooling operation. , The vector-secret graph input pair gets a 1 / 2 size vector-secret graph input pair , and a 1 / 4 size vector-secret graph input pair , Correspondingly, the steganographic image also has full size, half size and 1 / 4 size. , , Only the full-size encrypted image is transmitted to the receiver to recover the secret image. Similarly, the output of the recovery process has three sizes: full size, half size and quarter size. , and .

[0118] There are two cases in the hiding process. and .when When the final secret map , and Only the coarse hidden network Without any adjustment and refinement process, the hidden process can be expressed by the following formula:

[0119] (1)

[0120] when When the final secret map , and go through Adjustment and refinement. First, the coarse hidden network Generate the encrypted image in the 0th iteration , the process can be expressed as:

[0121] (2)

[0122] Then, in Iterative process, recovering the network from the carrier graph and secret graph recovery network Generate the intermediate recovered carrier graph and secret graph and After that, the input vector-secret graph pair and intermediate recovery carrier-secret graph pair Calculate the difference between and , the calculation process is as follows:

[0123] (3)

[0124] The secret image is recovered by the difference carrier-secret image pair in the subsequent iteration process. Continuously refine and adjust.

[0125] In the The encrypted map generated during the iteration process Obtained by the following formula:

[0126] (4)

[0127] Through formulas (3) and (4), the encrypted image is iteratively adjusted and optimized until .when hour, and From formula (3), we can get the following: , and By Refining hidden network during iteration get:

[0128] (5)

[0129] During the recovery process, full-size, half-size, and quarter-size recovery secret images , and Recovery Network via Secret Graph Refinement Recover from the encrypted image. The process is as follows:

[0130] (6)

[0131] In the proposed MPALS-Net, since secret image recovery depends on the secret-carrying image, continuous optimization of the secret-carrying image during the secret hiding process relies on intermediate carrier-secret image pairs. Therefore, the multi-stage optimization module not only iteratively adjusts the secret-carrying image but also performs multi-stage optimization on the recovered secret image. Therefore, appropriately increasing the number of iterative processes during the hiding phase can improve the image quality of the recovered secret image to a certain extent.

[0132] In addition, Secret Image Restoration Network and carrier image restoration network The convolution kernel size of the convolution layer is set to "3×3". At the same time, the secret image restoration network and carrier image restoration network The convolution kernel size of the middle convolution layer is set to 3×3, and the number of channels of the convolution layer is set to {( , 64), (64, 128), (128, 256),(256,128), (128, 64), (64, )};The coarse hidden network , refine the hidden network and image thinning and restoration network The number of channels in the convolutional layer is set to {( , 64), (64, 128),(128,256), (256, 128), (256, 64), (128, )};. Coarse hidden network , refine the hidden network and image thinning and restoration network The convolution kernel size of the middle convolution layer and the transposed convolution layer is set to "4 × 4".

[0133] To further adjust and optimize the secret image and the recovered secret image, the present invention proposes a multi-stage pyramid optimization module based on the multi-stage optimization module. It uses the multi-stage and multi-scale information of multi-scale carrier-secret and secret-recovered secret image pairs to enhance the intermediate feature representation of MPALS-Net. The structure of the multi-stage pyramid optimization module proposed by the present invention is as follows: Figure 2 As shown in Figure 2, the CPCO and CPP modules are marked with green and blue boxes, respectively. Dark blue boxes represent convolutional layers, and red boxes represent transposed convolutional layers. Furthermore, in the multi-stage pyramid optimization module during iteration 0, two 1×1 convolutional layers are used to transform the number of channels of the half-scale and quarter-scale input image pairs. Specifically, 1×1 convolutional layers are used after the downsampling operation to adjust the number of channels. Correspondingly, 1×1 convolutional layers are also used to adjust the number of channels during the generation of the half-scale and quarter-scale encrypted and recovered secret images.

[0134] During the hiding process, the full-scale, half-scale, and quarter-scale carrier-secret input pairs are gradually added to the coarse hidden network. Correspondingly, in the last iteration of the refined hidden network Generate full-scale, half-scale and quarter-scale density maps. Similarly, during the recovery process, The full-scale, half-scale and quarter-scale secret images are recovered. In addition, the half-scale and quarter-scale carrier-carrier and secret-recovered secret image pairs are used to generate the full-scale carrier and recovered secret images with better image quality. and secret - recover secret loss Combined, the optimization goal is to minimize the loss function, which is defined as follows:

[0135] (7)

[0136] in, and denotes the hiding and recovery loss weights. Indicates the loss weights of half-size and quarter-size. In addition, the multi-scale loss with only full-scale and half-scale pairs in the loss function is as follows:

[0137] (8)

[0138] In addition, when only full-scale and quarter-scale image pairs are used for steganography, the loss function can be expressed as:

[0139] (9)

[0140] This multi-scale CPP architecture ensures that fine-grained details and long-term dependencies from coarse scales can be preserved;

[0141] The specific implementation of the multi-stage learning large-capacity deep image steganography network is as follows:

[0142] 1. Initialization parameters: Carrier graph sample: and secret image samples ;

[0143] Number of secret images: And the number of carrier images: ;

[0144] Iterative optimization times: ;

[0145] Weights in the loss function 、 as well as ;

[0146] Initial learning rate: lr;

[0147] Number of batch samples: B;

[0148] Maximum number of iterations: T;

[0149] 2. Repeat the following steps (steps within an epoch) until the network converges:

[0150] If the number of iterations is greater than or equal to 1 and less than T:

[0151] Then do the following:

[0152] if =1, the final encrypted image is obtained according to the following formula , and :

[0153] ;

[0154] Otherwise, the encrypted image of the 0th iteration is obtained according to the following formula: :

[0155] ;

[0156] If the current number of iterations is greater than or equal to 0 and less than or equal to n-2, the difference pair is generated according to the following formula and :

[0157] ;

[0158] If the current number of iterations is not equal to, then the following formula is used to obtain :

[0159] ;

[0160] The recovered secret images of full size, half size and quarter size are obtained according to the following formula , and :

[0161] ;

[0162] The overall loss is calculated according to the following formula:

[0163] ;

[0164] Update model parameters using gradient descent algorithm;

[0165] In summary, the steganography method provided by the present invention aims at the problem of insufficient feature learning in the process of large-capacity secret hiding. From the two perspectives of contrast refinement and multi-scale adjustment, a multi-stage adjustment large-capacity image steganography network is designed. By designing a coarse-to-fine optimization stage in the hiding process, multi-stage iterative optimization is achieved, thereby improving the modeling ability of the model in large-capacity steganography situations. A multi-stage optimization module is designed, which utilizes the multi-stage contrast information of the carrier-carrier secret and secret-recovery secret graph pairs to iteratively optimize and adjust the generated carrier secret and recovery secret graphs. A multi-stage pyramid optimization module is designed, which directly adjusts and enhances the intermediate feature representation of the hiding and recovery process through the multi-stage and multi-scale information of the carrier-carrier secret and secret-recovery secret graph pairs of different scales, thereby improving the modeling ability and steganography performance of the model for large-capacity steganography situations, combined with Figures 3 to 5 As can be seen from the comparison diagram, the method provided by the present invention has significant improvements over existing methods in various steganographic situations.

[0166] Embodiment 2: This embodiment provides a processing device, including:

[0167] Acquisition unit: used to acquire the image to be processed;

[0168] Processing unit: used to input the image to be processed into a pre-trained multi-stage learning large-capacity deep image steganography model and perform the following processing:

[0169] Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales;

[0170] The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

[0171] The specific functional implementation of each of the above modules can be found in the relevant content of the method in Example 1 and will not be elaborated on here.

[0172] Example 3: This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods described in Example 1 are implemented.

[0173] Embodiment 4: This embodiment provides a computer device, including:

[0174] Memory, used to store computer programs / instructions;

[0175] A processor, configured to execute the computer program / instructions to implement the steps of any one of the methods described in Example 1.

[0176] Example 5: This embodiment provides a computer program product, including a computer program / instruction, which implements the steps of any method described in Example 1 when executed by a processor.

[0177] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0178] Those skilled in the art will appreciate that embodiments of the present disclosure may be provided as methods, systems, or computer program products. Thus, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0179] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0180] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0181] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure and are not intended to limit its scope of protection. Although the present disclosure has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the disclosed claims to be approved.

Claims

1. A large-capacity image steganography method based on multi-stage learning, characterized in that: include: Get the image to be processed; The image to be processed is input into a pre-trained multi-stage learning large-capacity deep image steganography model, and the following processing is performed: Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales; The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

2. The large-capacity image steganography method based on multi-stage learning according to claim 1 is characterized in that: The types of images to be processed include carrier images, secret images and secret-carrying images; When the image to be processed is a carrier image and a secret image, a hiding process is performed; When the image to be processed is a confidential image, restoration processing is performed.

3. The large-capacity image steganography method based on multi-stage learning according to claim 2 is characterized in that: In the hidden processing process, multi-scale input pairs are obtained from the image to be processed through downsampling operation, and two 1 × 1 convolutional layers are used to transform the number of channels of half-scale and quarter-scale input image pairs in the multi-scale input pairs; According to the number of iterative optimizations, the transformed multi-scale input pairs are subjected to basic encryption processing or iterative encryption processing to obtain encrypted images of different scales; When the number of iterations is 1, basic encryption processing is performed. When the number of iterations is greater than 1, iterative encryption processing is performed. In the basic encryption process, the target image is generated through the coarse hidden network; During the iterative encryption process, a full-scale initial encrypted image is first generated based on the coarse hidden network. The intermediate recovered encrypted image and secret image are then generated based on the initial encrypted image through the encrypted image recovery network and the secret image recovery network. Difference pairs are obtained based on the input pairs and the intermediate recovered pairs. The initial encrypted image is iteratively adjusted through the difference pairs to obtain encrypted images of different scales. During the recovery process, the recovered secret graphs of different scales are obtained from the input secret graph through the secret graph refinement recovery network.

4. The large-capacity image steganography method based on multi-stage learning according to claim 3 is characterized in that: In the generation of the target image through the coarse hidden network, the hidden process is expressed by the following formula: （1）； Where, 、 and They are the final full-scale, 1 / 2-scale and 1 / 4-scale density maps, is the coarse hidden network, is the carrier image, For the secret picture, , , and They are and The height and width of and The number of channels; and is a 1 / 2 scale vector-secret graph input pair, , ,; and is a 1 / 4 scale vector-secret graph input pair, , ; In the iterative encryption process, a coarse hidden network is used to perform basic encryption to obtain an initial encrypted image. The initial encrypted image is the final full-scale encrypted image obtained in the basic encryption process. , the generation process of the initial encrypted image is expressed as: （2）； exist Iterative process, recovering the network from the carrier graph and secret graph recovery network The intermediate recovered carrier map is generated based on the initial carrier map and secret map , by input vector-secret graph pair and intermediate recovery carrier-secret graph pair Calculate the difference between and , the calculation process is as follows: (3)； The secret image is recovered by the difference carrier-secret image pair in the subsequent iteration process. Continuously refine and adjust The encrypted map generated during the iteration process Obtained by the following formula: (4)； Through formulas (3) and (4), the encrypted image is iteratively adjusted and optimized until ,when hour, and Obtained from formula (3); Different sizes of secret maps , and By Refining hidden network during iteration According to The difference pairs are obtained in the iterative process, and the hidden process is expressed as follows: (5)； Where, It is a full-size encrypted map. It is a 1 / 2 size dense map. It is a 1 / 4 size encrypted image; The method of obtaining recovered secret graphs of different scales from the input secret graph through the secret graph refinement recovery network includes: （6）； Where, For the full-size secret recovery map, The secret recovery image is 1 / 2 size. The secret recovery image is 1 / 4 size. Refine the recovery network for the secret graph, It is a full-size confidential map.

5. The large-capacity image steganography method based on multi-stage learning according to claim 4 is characterized in that: The Secret Image Restoration Network and carrier image restoration network The convolution kernel size of the middle convolution layer is set to 3×3, and the number of channels of the convolution layer is set to {( , 64), (64, 128), (128, 256),(256, 128), (128, 64), (64, )}; The coarse hidden network , refine the hidden network and image thinning and restoration network The number of channels in the convolutional layer is set to {( , 64), (64, 128),(128, 256), (256, 128), (256,64), (128, )}; The coarse hidden network , refine the hidden network and image thinning and restoration network The convolution kernel size of the middle convolution layer and the transposed convolution layer is set to 4 × 4.

6. The large-capacity image steganography method based on multi-stage learning according to claim 5 is characterized in that: The multi-stage optimization of the secret-carrying graphs or secret recovery graphs of different scales according to the multi-scale information includes: Different loss functions are selected based on multi-scale information to optimize the secret-carrying images or secret recovery images of different scales. The optimization goal is to minimize the loss function, where: When half-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows: （7）； Where, and represents the hiding and recovery loss weights, To hide the losses, To recover the losses, Represents the loss weight of the scale image used; When the intermediate images of the full-scale and half-scale pairs are used to generate the target image, the loss function as follows: (8)； When the full-scale and quarter-scale intermediate images are used to generate the target image, the loss function as follows: (9)。 7. A large-capacity image steganography system based on multi-stage learning, characterized in that: include: Acquisition unit: used to acquire the image to be processed; Processing unit: used to input the image to be processed into a pre-trained multi-stage learning large-capacity deep image steganography model and perform the following processing: Performing hiding processing or recovery processing on the image to be processed according to the type of the image to be processed, and obtaining secret images or secret recovery images of different scales; The secret-carrying images or secret recovery images of different scales are optimized in multiple stages according to the multi-scale information to obtain and output a target image.

8. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer device comprising: Memory, used to store computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instruction, which implements the steps of the method according to any one of claims 1 to 6 when executed by a processor.

Images