Image processing method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses an image processing method, an image processing device, electronic equipment, a computer storage medium and a computer program product, and the method comprises the steps: obtaining a secret image and a carrier image, and enabling the size of the secret image to be larger than the size of the carrier image; calling an image hiding model, and hiding the secret image into the carrier image to obtain a secret-carrying image; performing image merging processing on reference image areas contained in the secret-carrying image and the background image to obtain a merged image area; and replacing the reference image area of the background image with the combined image area to obtain a steganographic image. According to the technical scheme, the flexibility of image steganography can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer and communication technology, and in particular, to an image processing method, an image processing device, an electronic device, a computer storage medium, and a computer program product. Background Art

[0002] Image steganography refers to embedding a secret image into a carrier image under the premise of being imperceptible. In the relevant technology, there are three main types of image steganography schemes, namely, steganography schemes based on image transform domain (such as discrete Fourier transform, discrete cosine transform, etc.), steganography schemes based on image space domain (such as least significant bit, histogram shift, etc.) and steganography schemes based on deep learning. Among them, the first and second types of schemes have low embedding rates and are easy to detect secret images in the carrier medium; while the third type of scheme, although it improves the embedding rate to a certain extent, can only hide one image into another image of the same size at most, resulting in a lack of flexibility in the location of the secret image. Therefore, how to improve the flexibility of image steganography is a problem that needs to be solved urgently. Summary of the invention

[0003] The embodiments of the present application provide an image processing method, an image processing device, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the flexibility of image steganography.

[0004] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present application.

[0005] In a first aspect, an embodiment of the present application provides an image processing method, comprising:

[0006] Acquire a secret image and a carrier image, wherein the size of the secret image is larger than the size of the carrier image;

[0007] Calling an image hiding model to hide the secret image in the carrier image to obtain a secret image; wherein the image hiding model comprises a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract a reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to perform feature merging of the reduced dimension feature map with the carrier image to obtain a target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image;

[0008] Performing image merging processing on the reference image area contained in the secret image and the background image to obtain a merged image area;

[0009] The reference image area of ​​the background image is replaced by the merged image area to obtain a steganographic image.

[0010] In a second aspect, an embodiment of the present application provides an image processing device, the device comprising an acquisition unit and an image hiding unit, wherein:

[0011] The acquisition unit is used to acquire a secret image and a carrier image, wherein the size of the secret image is larger than the size of the carrier image;

[0012] The image hiding unit is used to call the image hiding model to hide the secret image in the carrier image to obtain the secret image; wherein the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract a reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to perform feature merging of the reduced dimension feature map with the carrier image to obtain a target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image;

[0013] The image hiding unit is further used to perform image merging processing on the reference image area contained in the secret image and the background image to obtain a merged image area;

[0014] The image hiding unit is further used to replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising one or more processors; a storage device for storing one or more computer programs, wherein when the one or more computer programs are executed by the one or more processors, the electronic device implements the data processing method for push data recall as described above.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of an electronic device, the electronic device executes the image processing method as described above.

[0017] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, wherein the computer program is stored in a computer-readable storage medium, and a processor of an electronic device reads and executes the computer program from the computer-readable storage medium, so that the electronic device performs the image processing method as described above.

[0018] In the technical solution provided in the embodiment of the present application, a reduced-dimensional feature map of a secret image of the same size as the carrier image can be extracted through the feature extraction network in the image hiding model, so that the subsequent merging network can realize the preliminary merging of the secret image and the carrier image by merging the reduced-dimensional feature map and the carrier image. Finally, since the target hidden feature map is only a feature map, it is necessary to activate the network to perform activation mapping processing on it, and finally obtain the secret image that completely hides the secret image. Therefore, the embodiment of the present application can embed the secret image into a carrier image smaller in size than the secret image through the image hiding model, thereby improving the embedding rate of image steganography. At the same time, the smaller secret image can be flexibly embedded in the reference image area in the background image, thereby improving the flexibility of image steganography and the robustness to interference such as cropping.

[0019] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0021] Figure 1 This is a schematic diagram of an application scenario of image steganography provided by an embodiment of the present application;

[0022] Figure 2 is a structural schematic diagram of an image processing system provided in an embodiment of the present application;

[0023] Figure 3 It is a flowchart of an image processing method provided in an embodiment of the present application;

[0024] Figure 4 is a flowchart of another image processing method provided in an embodiment of the present application;

[0025] Figure 5 It is a flowchart of another image processing method provided in an embodiment of the present application;

[0026] Figure 6 It is a schematic diagram of an image steganography process provided by an embodiment of the present application;

[0027] Figure 7 It is a flowchart of another image processing method provided in an embodiment of the present application;

[0028] Figure 8 It is a schematic diagram of the training process of an image steganography model provided in an embodiment of the present application;

[0029] Fig. 9 is a structural block diagram of an image processing device provided in an embodiment of the present application;

[0030] Fig.10 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present application is shown. DETAILED DESCRIPTION

[0031] Here, exemplary embodiments will be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0032] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0033] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all contents and operations, nor must they be executed in the order described. For example, some operations may be decomposed, while some operations may be combined or partially combined, so the actual execution order may change according to actual conditions.

[0034] It should also be noted that the "multiple" mentioned in this application refers to two or more than two. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship.

[0035] With the rapid development of the Internet, digital media such as images and videos have become important media for transmitting information. These information are generated and transmitted on the Internet every day, which brings new challenges to information security. In some scenarios where secret messages are shared, the sender of the message does not want the secret message to be obtained by unauthorized third parties, so the sender of the message will choose to encrypt the message.

[0036] At present, encryption methods such as cryptography and information hiding technology have been proposed to improve information security. Among them, cryptography methods can directly protect the content of information by encryption; information hiding technology includes image steganography technology and digital watermarking technology, both of which are used to hide confidential information in harmless media such as images, videos, texts, and audios. Digital watermarking technology is usually used for the authentication and copyright protection of digital information. Image steganography technology is the art of hiding information. It can hide secret images in carrier media in an invisible way without arousing suspicion from third parties, and at the same time, the original secret information can be extracted from it with high quality.

[0037] For details, please see the attached Figure 1 , shows a schematic diagram of an application scenario of image steganography. Figure 1 As shown in the figure, the sender of the message embeds the secret image 101 containing the secret information "123" into the ordinary landscape image 102, and then obtains the image 103 containing the secret image. If the steganographic image 103 is transmitted in an insecure channel, even if a third party intercepts the image 103, they will only think that it is an ordinary landscape photo; only the receiver authorized by the sender can recover the real secret image 104 from the landscape photo.

[0038] In the related art, there are three main types of image steganography schemes; the first type is the steganography scheme based on the image transform domain, which usually cleverly hides binary information in the coefficients of certain image transforms (such as discrete cosine transform, discrete wavelet transform, discrete Fourier transform). Although the steganography scheme based on the image transform domain shows good robustness to common interference, the computational complexity is very high, and the information embedding rate (which can be measured by the number of bits of secret information per pixel) is very low.

[0039] The second category is steganographic schemes based on the image space domain, such as the least significant bit-based steganographic method, the histogram shift-based steganographic method, the differential expansion-based steganographic method, etc.; although the steganographic schemes based on the image space domain can achieve a higher embedding rate than the methods based on the image transform domain, the embedding rate is still at a low level, and it is easy for steganalysis to detect the presence of secret information in the image.

[0040] The third category is steganography schemes based on deep learning, such as steganography schemes based on convolutional neural networks, steganography schemes based on generative adversarial networks, etc. Compared with the two traditional steganography schemes based on image transformation domain and image space domain, the embedding rate of steganography schemes based on deep learning has been improved to a certain extent. However, at present, steganography schemes based on deep learning can only embed secret images into carrier images of the same size, and the overall embedding rate is not high, which makes the location of secret image hiding lack of flexibility.

[0041] Among them, deep learning is an important part of machine learning (ML), and machine learning is an important research direction of artificial intelligence (AI) technology.

[0042] The so-called artificial intelligence technology refers to the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science; it mainly produces a new intelligent machine that can respond in a similar way to human intelligence by understanding the essence of intelligence, so that the intelligent machine has multiple functions such as perception, reasoning and decision-making. Accordingly, AI technology is a comprehensive discipline, which mainly includes several major directions such as computer vision technology (Computer Vision, CV), speech technology, natural language processing technology and machine learning / deep learning.

[0043] The so-called machine learning is a multi-disciplinary cross-discipline involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of AI and the fundamental way to make computer devices intelligent; the so-called machine learning is a multi-disciplinary cross-discipline involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines; it specializes in studying how computer devices simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance. Deep learning is a technology that uses deep neural network systems to perform machine learning; machine learning / deep learning can usually include artificial neural networks, reinforcement learning (RL), supervised learning, unsupervised learning and other technologies; the so-called supervised learning refers to the processing method of using training samples with known categories (with labeled categories) for model training, and unsupervised learning refers to the processing method of using training samples with unknown categories (not labeled) for model training.

[0044] Based on this, an embodiment of the present application provides an image processing solution, which can hide the secret image in a carrier image smaller than the secret image by calling an image hiding model, thereby obtaining a secret image; then, the secret image is merged with the reference image area contained in the background image to obtain a merged image area; finally, the reference image area of ​​the background image is replaced with the merged image area to obtain a steganographic image.

[0045] Among them, the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence; the feature extraction network is used to extract the dimensionality reduction feature map of the secret image, and the size of the dimensionality reduction feature map is the same as the size of the carrier image; the merging network is used to merge the dimensionality reduction feature map with the carrier image to obtain the target hidden feature map; the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image.

[0046] In addition, the secret image refers to the secret information to be embedded; the carrier image refers to the image embedded with the secret image; and the steganographic image refers to the image containing the secret image. In addition, the size of the background image is greater than or equal to the size of the carrier image; the reference image area of ​​the background image includes a preset image area or a random image area in the background image, that is, the reference image area can be a pre-set image area in the background image or a randomly selected image area in the background image.

[0047] It is not difficult to see that this scheme can extract the reduced-dimensional feature map of the secret image with the same size as the carrier image through the feature extraction network in the image hiding model, so that the subsequent merging network can realize the preliminary merging of the secret image into the carrier image by merging the reduced-dimensional feature map and the carrier image. Finally, since the target hidden feature map is only a feature map, it is necessary to activate the network to perform activation mapping processing on it, and finally obtain the secret image that completely hides the secret image. At the same time, this scheme merges and replaces the reference image area of ​​the secret image with the background image, so that the position of the secret image in the final stego image is more flexible and has stronger robustness to cropping. It can be seen that this scheme can embed the secret image into the carrier image with a smaller size than the secret image through the image hiding model, thereby improving the embedding rate of image steganography; at the same time, the obtained secret image can be flexibly embedded into the background image to form a stego image, which improves the flexibility of image steganography.

[0048] Based on the above image processing solution, the present application embodiment provides an image processing system, which can be seen in Figure 2 , Figure 2The image processing system shown may include multiple terminal devices 201 and multiple servers 202. Among them, a communication connection is established between any terminal device and any server. For example, the terminal device 201 may include any one or more of a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart car-mounted device, and a smart wearable device. Multimedia playback clients, social clients, browser clients, information stream clients, educational clients, etc. can be run in the terminal device 201. The server 202 can be a server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms. Any terminal device 201 and any server 202 can be directly or indirectly connected in communication via wired or wireless communication, and this application is not limited here.

[0049] In some embodiments, the above image processing method can be performed only by Figure 2 The terminal device 201 in the image processing system shown in the figure executes, and the specific execution process is as follows: the terminal device 201 can first obtain the secret image and the carrier image, wherein the size of the secret image is larger than the size of the carrier image. Then, the terminal device 201 can call the image hiding model to hide the secret image in the carrier image to obtain the secret image; wherein the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract the reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to merge the reduced dimension feature map with the carrier image to obtain the target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image. Afterwards, the terminal device 201 can perform image merging processing on the reference image area contained in the secret image and the background image to obtain the merged image area. Finally, the terminal device 201 can replace the reference image area of ​​the background image with the merged image area to obtain the steganographic image.

[0050] Optionally, the terminal device 201 may also output a steganographic image. Optionally, the terminal device 201 may also obtain a receiving terminal identifier of the steganographic image and send the steganographic image to a terminal device indicated by the receiving terminal identifier. Optionally, the above-mentioned image processing method may also be performed by only Figure 2 The server 202 in the image processing system shown in the figure executes, and the specific execution process thereof can refer to the specific execution process of the terminal device 201, which will not be repeated here.

[0051] In some other embodiments, the above image processing method may be run in an image processing system, which may include a terminal device and a server. Figure 2 The terminal device 201 and the server 202 included in the image processing system shown are jointly completed. The specific execution process is as follows: the terminal device 201 can obtain the secret image and the carrier image, and then send the secret image and the carrier image to the server 202; after receiving the secret image and the carrier image, the server 202 can call the image hiding model to hide the secret image in the carrier image to obtain the secret image; wherein the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract the reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to merge the reduced dimension feature map with the carrier image to obtain the target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image. Afterwards, the server 202 can perform image merging processing on the reference image area contained in the secret image and the background image to obtain the merged image area. Finally, the server 202 can replace the reference image area of ​​the background image with the merged image area to obtain the stego image.

[0052] Optionally, the server 202 may send the stego image to the terminal device 201. Optionally, the server 202 may also directly obtain the receiving terminal identifier of the stego image, and send the stego image to the terminal device indicated by the receiving terminal identifier.

[0053] It should be noted that the embodiments of the present application can process various images that require secure transmission, and can be applied to various scenarios, including but not limited to cloud technology, AI (Artificial Intelligence), smart home, assisted driving and other scenarios. It can also be used in financial applications, social applications, shopping applications and other applications that require secure information transmission, without limitation.

[0054] Specifically, the technical solution of the embodiment of the present application can be applied to a smart home scene, such as the terminal device can be set as a camera device in a smart home scene, the server can be a server that provides image processing services for smart devices in the smart home scene, and the secret image can be an image of the living room captured by the camera device. After capturing the image of the living room, the camera device can upload the image of the living room to the server; then, the server can randomly select two pre-stored images from an image database storing multiple pre-stored images of different sizes as a carrier image and a background image, respectively. Afterwards, the server can call an image hiding model to hide the living room image into the selected pre-stored image to obtain a secret image containing the living room image. After obtaining the secret image containing the living room image, the server can perform image merging processing on the secret image containing the living room image and the reference image area in the selected background image to obtain a merged image area; finally, the server can replace the reference image area of ​​the background image with the above-mentioned merged image area to obtain the corresponding stego image.

[0055] For another example, the technical solution of the embodiment of the present application is applied to a social application, and the terminal device can be a terminal device running a social application. The object using the social application can select an image as a secret image, select an image as a carrier image, and select an image as a background image in the conversation interface of the social application; then, the terminal device can call the image hiding model to hide the selected secret image into the selected carrier image to obtain the corresponding secret image. After the corresponding secret image, the terminal device can perform image merging processing on the corresponding secret image and the reference image area contained in the selected background image to obtain a merged image area; finally, the terminal device can replace the reference image area of ​​the selected background image with the merged image area obtained above to obtain the corresponding steganographic image.

[0056] Optionally, after obtaining the corresponding stego image, the terminal device may send the corresponding stego image to another terminal device used by the conversation object of the above-mentioned conversation interface; further, after receiving the corresponding stego image, the other terminal device may perform image recovery processing on the stego image to obtain a restored image of the selected secret image; finally, the other terminal device may also display the restored image.

[0057] It should be noted that in the specific implementation of the present application, if the secret image, carrier image, background image and other related image data are related to the object, when the embodiment of the present application is applied to a specific product or technology, it is necessary to obtain the permission or consent of the object, and the collection, use and processing of the relevant image data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0058] The following is a detailed description of various implementation details of the technical solution of the embodiment of the present application:

[0059] like Figure 3 As shown, Figure 3 is a flowchart of an image processing method according to an embodiment of the present application, which can be applied to Figure 2 The image processing system shown in the figure, the method can be executed by the terminal device or the server, or can be executed by the terminal device and the server together. In the embodiment of the present application, the method is executed by the server as an example for explanation, wherein the image processing method may include S301 to S304, which are described in detail as follows:

[0060] S301, obtaining a secret image and a carrier image.

[0061] In the embodiment of the present application, the size of the secret image is larger than the size of the carrier image. The method of obtaining the secret image may be: the server receives the secret image uploaded by the terminal device; wherein the secret image may be an image selected by the user of the terminal device from multiple images displayed by the terminal device, or may be an image input into the terminal device by the user of the terminal device through downloading or mobile hard disk, and there is no limitation on this.

[0062] Optionally, the server may also store the secret image in a database, and the server may directly obtain the secret image from the corresponding database.

[0063] In one embodiment, the method for obtaining the carrier image can be: the server receives the carrier image uploaded by the terminal device; wherein the carrier image can be an image selected by the user of the terminal device from multiple images displayed by the terminal device, or it can be an image input into the terminal device by the user of the terminal device through downloading or mobile hard disk, and there is no limitation on this.

[0064] Optionally, the server may also pre-store a plurality of preset images of different sizes in a certain database, and the server may directly select a preset image whose size is smaller than the secret image from the corresponding database as the carrier image.

[0065] S302: Call the image hiding model to hide the secret image in the carrier image to obtain the secret-carrying image.

[0066] In the embodiment of the present application, the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence. Among them, the feature extraction network is used to extract the dimension reduction feature map of the secret image, and the size of the dimension reduction feature map is the same as the size of the carrier image; the merging network is used to merge the dimension reduction feature map with the carrier image to obtain the target hidden feature map; the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image.

[0067] In one embodiment, the feature extraction network may include a dimensionality reduction convolution layer, or multiple dimensionality reduction convolution layers connected in sequence; then, the specific process of extracting the dimensionality reduction feature map of the secret image through the feature extraction network may be: performing dimensionality reduction feature extraction processing on the secret image through one or more dimensionality reduction convolution layers to obtain the dimensionality reduction feature map of the secret image. Among them, the one or more dimensionality reduction convolution layers scale the size of the secret image (also referred to as a scaling factor) by the same multiple as the size multiple between the secret image and the carrier image, so that the size of the finally obtained dimensionality reduction feature map is the same as the size of the carrier image.

[0068] In one embodiment, the size of the target hidden feature map is the same as the size of the carrier image, and the size of the secret image is also the same as the size of the carrier image, so each pixel in the target hidden feature map corresponds to each pixel in the secret image.

[0069] At the same time, the activation network may include an activation function and a pixel mapping function; then, the specific method of performing activation mapping processing on each pixel in the target hidden feature map through the activation network may be: normalizing each pixel in the target hidden feature map through the activation function to obtain the range mapping value of each pixel; mapping the range mapping value of each pixel in the target hidden feature map through the pixel mapping function to obtain the mapping pixel value of each pixel in the target hidden feature map; finally, based on the mapping pixel value of each pixel in the target hidden feature map, a encrypted image is obtained.

[0070] Specifically, the activation function can be any of a variety of functions such as the hyperbolic tangent function (i.e., Tanh function), the S-shaped growth curve function (i.e., sigmoid function), the linear rectification function (i.e., Relu function), the parameter-corrected linear function (i.e., PReLU function), etc., without limitation. At the same time, since the activation function is used to map each pixel to a corresponding numerical range, such as a numerical range of -1 to 1; therefore, the pixel mapping function is mainly to map the range mapping value of each pixel within the corresponding numerical range into a suitable pixel value to form a carrier image similar to the carrier image; then, the pixel mapping function can be determined when training the image hiding model, based on the relationship between the pixel value of each pixel in the carrier image used during training and the range mapping value of each pixel in the target hidden feature map obtained during training.

[0071] S303, performing image merging processing on the reference image area contained in the encrypted image and the background image to obtain a merged image area.

[0072] In the embodiment of the present application, the reference image area includes a pre-set image area in the background image and a randomly selected image area in the background image; the area size of the reference image area is the same as the image size of the confidential image.

[0073] In one embodiment, a specific method of performing image merging processing on the confidential image and the reference image area contained in the background image may be: overlaying the confidential image on the reference image area to obtain a merged image area.

[0074] Furthermore, the merged image area can also be processed for transparency change to obtain a processed merged image area. Specifically, the specific method of transparency change processing can be: using the layer of the secret image in the merged image area as the foreground image, and using the layer of the reference image area in the merged image area as the background image; obtaining the reference transparency, and according to the preset transparency, the values ​​of the three color channels of red, yellow, and blue of each pixel in the foreground image, and the values ​​of the three color channels of red, yellow, and blue of each pixel in the background image, changing the values ​​of the three color channels of red, yellow, and blue of each pixel in the merged image area to obtain the processed merged image area. The reference transparency can be pre-set or randomly selected. For example, the reference transparency can range from 0 to 1, where 1 represents complete transparency and 0 represents complete opacity.

[0075] In a specific implementation, the transparency change process of the merged image area can be performed by using an alpha synthesis algorithm, AlphaBlend (a function used to display an image with a specified transparency), and the like.

[0076] S304: Replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

[0077] In an embodiment of the present application, the specific method of replacing the reference image area of ​​the background image with the merged image area can be: extracting the reference image area from the background image; placing the merged image area in the area of ​​the original reference image area in the background image to obtain a steganographic image.

[0078] Optionally, since it is mentioned in step S303 that the transparency of the merged image area can also be changed to obtain a processed merged image area; therefore, the reference image area of ​​the background image is replaced with the merged image area, and the specific method of obtaining the steganographic image can also be: replacing the reference image area of ​​the background image with the processed merged image area to obtain the steganographic image.

[0079] Optionally, the server may also output the stego image. Optionally, the server may also obtain identification information of a terminal device to receive the stego image, and send the stego image to the terminal device indicated by the identification information.

[0080] In the embodiment of the present application, the feature extraction network in the image hiding model can extract the reduced-dimensional feature map of the secret image with the same size as the carrier image, so that the subsequent merging network can realize the preliminary merging of the secret image and the carrier image by merging the reduced-dimensional feature map and the carrier image. Finally, since the target hidden feature map is only a feature map, it is necessary to activate the network to perform activation mapping processing on it, and finally obtain the secret image that completely hides the secret image. Therefore, the embodiment of the present application can embed the secret image into a carrier image with a smaller size than the secret image through the image hiding model, thereby improving the embedding rate of image steganography.

[0081] At the same time, the embodiment of the present application makes the position of the secret image in the final stego image more flexible by merging and replacing the secret image with the reference image area of ​​the background image. The stego image may be affected by cropping and other interferences during transmission, resulting in the loss of part of the image area; but since the area containing the secret image in the stego image is not necessarily the lost image area, the possibility of complete transmission of the secret image in the embodiment of the present application is higher than that in which all image areas in the stego image contain the secret image. Therefore, the embodiment of the present application improves the flexibility of image steganography and the robustness to interferences such as cropping by flexibly embedding the secret image into the reference image area of ​​the background image to form the stego image.

[0082] In addition, the embodiment of the present application changes the transparency of the merged image area so that the confidential image and the background image are more harmonious and natural visually, which is beneficial to improving the concealment of the confidential image in the background image.

[0083] In one embodiment of the present application, another image processing method is provided, which can be applied to Figure 1 The image processing system shown in the figure, the method can be executed by the terminal device or the server, or can be executed by the terminal device and the server together. In the embodiment of the present application, the method is executed by the server as an example for explanation. Figure 4 FIG. 4 is a flow chart showing another image processing method. Figure 3 This paper expands on the method shown in .

[0084] Specifically, the image hiding model may include a feature extraction network, a merging network, and an activation network connected in sequence; and the merging network may specifically include a fusion network and a residual network connected in sequence.

[0085] Among them, S401 to S407 are described in detail as follows:

[0086] S401, obtaining a secret image and a carrier image.

[0087] In the embodiment of the present application, the specific implementation of step S401 can refer to the specific implementation of step S301 in the above embodiment, which will not be repeated here.

[0088] S402: extracting a dimensionality reduction feature map of the secret image through a feature extraction network.

[0089] In the embodiment of the present application, the specific implementation of step S401 can refer to the specific implementation of the feature extraction network in step S302 in the above embodiment, which will not be repeated here.

[0090] S403, performing feature fusion on the reduced dimension feature map and the carrier image through a fusion network to obtain an initial hidden feature map.

[0091] In the embodiment of the present application, feature fusion can be the fusion of the reduced dimension feature map and the carrier image at the feature channel level. Specifically, the reduced dimension feature map can include one or more reduced dimension feature sub-maps of feature channels, and the carrier image can include one or more carrier feature sub-maps of feature channels.

[0092] Then, the specific method of fusing the reduced dimension feature map and the carrier image through the fusion network to obtain the initial hidden feature map can be: merging multiple reduced dimension feature sub-maps included in the reduced dimension feature map with multiple carrier feature sub-maps included in the carrier image to obtain the initial hidden feature map; wherein the number of feature channels of the initial hidden feature map is the sum of the number of feature channels of the reduced dimension feature map and the carrier image. Specifically, the feature channels of the image can be color channels such as red, yellow, and blue, which are not limited to this.

[0093] In the specific implementation, since the dimensions of the reduced dimension feature map and the carrier image are the same, the length and width of the reduced dimension feature map and the carrier image are the same; from the pixel level, the shapes of the reduced dimension feature map and the carrier image can both be c*width*height, where c refers to the number of feature channels (here it can be 3), width refers to width, and height refers to length. Then, the feature fusion of the reduced dimension feature map and the carrier image is the merging of the reduced dimension feature map and the carrier image in the feature channel, and the shape of the final initial hidden feature map can be 6*width*height.

[0094] Optionally, since the reduced dimension feature map and the carrier image can be regarded as two feature vectors of the same dimension; therefore, the reduced dimension feature map and the carrier image are subjected to feature fusion through a fusion network, and the specific method of obtaining the initial hidden feature map can be feature fusion, and the specific method of obtaining the initial hidden feature map can also be: performing an outproduct (a linear operation method) on the reduced dimension feature map and the carrier image to obtain the initial hidden feature map. Optionally, the reduced dimension feature map and the carrier image can also be subjected to feature fusion through other methods for fusion between features, which are not limited to this.

[0095] S404: Perform residual feature extraction processing on the initial hidden feature map through a residual network to obtain a target hidden feature map.

[0096] In an embodiment of the present application, the residual network may include a downsampling subnetwork and an upsampling subnetwork connected in sequence, and the downsampling subnetwork includes a plurality of downsampling network layers connected in sequence.

[0097] Among them, the signal output end of each downsampling network layer is respectively connected to the signal input end of the upsampling sub-network, so that the upsampling sub-network performs upsampling feature extraction processing according to the downsampling feature maps output by each downsampling network layer to obtain the target hidden feature map.

[0098] In one embodiment, the upsampling subnetwork includes a plurality of upsampling network layers connected in sequence, and each downsampling network layer in the downsampling subnetwork corresponds to each upsampling network layer in the upsampling subnetwork except the first upsampling network layer, wherein: the first upsampling network layer is used to perform upsampling processing on the downsampling feature map output by the last downsampling network layer; the signal input end of each upsampling network layer except the first upsampling network layer in the plurality of upsampling network layers is also connected to the signal output end of the corresponding downsampling network layer in the downsampling subnetwork, so as to perform upsampling processing on the fused feature map after feature fusion of the received input signal.

[0099] Among them, the input signals received by each upsampling network layer except the first upsampling network layer among the multiple upsampling network layers include: the upsampling feature map output by the previous upsampling network layer of each upsampling network layer, and the downsampling network layer output by the downsampling network layer corresponding to each upsampling network layer.

[0100] Optionally, the correspondence between each downsampling network layer in the downsampling subnetwork and each upsampling network layer in the upsampling subnetwork except the first upsampling network layer can be: according to the connection order of each downsampling network layer in the downsampling subnetwork and the reverse order of the connection order of each upsampling network layer in the upsampling subnetwork except the first upsampling network layer, a one-to-one correspondence between each downsampling network layer in the downsampling subnetwork and each upsampling network layer in the upsampling subnetwork except the first upsampling network layer is established. For example, the downsampling subnetwork includes 3 downsampling network layers and the upsampling subnetwork includes 4 upsampling network layers; then, a correspondence between the first downsampling network layer and the last upsampling network layer, a correspondence between the second downsampling network layer and the last upsampling network layer, and a correspondence between the third downsampling network layer and the last upsampling network layer are established.

[0101] Optionally, the above correspondence may also be: according to the connection order of each downsampling network layer in the downsampling subnetwork and the connection order of each upsampling network layer except the first upsampling network layer in the upsampling subnetwork, a one-to-one correspondence between each downsampling network layer in the downsampling subnetwork and each upsampling network layer except the first upsampling network layer in the upsampling subnetwork is established. That is, the first downsampling network layer corresponds to the first upsampling network layer, the second downsampling network layer corresponds to the second upsampling network layer, and the third downsampling network layer corresponds to the third upsampling network layer.

[0102] Optionally, the correspondence between each downsampling network layer in the downsampling subnetwork and each upsampling network layer in the upsampling subnetwork except the first upsampling network layer may also be other one-to-one, one-to-many or many-to-many correspondences, which is not limited to this.

[0103] In one embodiment, the specific processing process of the first upsampling network layer may be: performing deconvolution activation processing on the downsampled feature map output by the last downsampled network layer through the first upsampling network layer to obtain the upsampled feature map output by the first upsampling network layer.

[0104] The specific processing process of each upsampling network layer except the first upsampling network layer in the multiple upsampling network layers may be: through the above-mentioned upsampling network layers, the upsampling feature map output by the previous upsampling network layer of each upsampling network layer and the downsampling network layer output by the downsampling network layer corresponding to each upsampling network layer are feature fused, and the fused feature map is deconvolutionally activated to obtain the upsampling feature map output by each upsampling network layer.

[0105] Specifically, the upsampling network layer may include a deconvolution layer, a first activation function layer, and a normalization layer connected in sequence, so the deconvolution activation process is also the process of the first activation function layer, the deconvolution layer, and the normalization layer. In a specific implementation, the activation function in the first activation function layer may be a ReLU function, or a LeakyReLU function (an activation function related to the Sigmoid function and the more common ReLU function), or other activation functions, without limitation. The normalization layer may use Instance Normalization, or may use normalization methods such as BatchNormalization and Layer Normalization, without limitation.

[0106] In one embodiment, the first downsampling network layer in the downsampling subnetwork is used to downsample the initial hidden feature map to obtain the downsampled feature map output by the first downsampling network layer; each downsampling network layer except the first downsampling feature map in the multiple downsampling network layers is used to downsample the downsampled feature map output by the previous downsampling network layer to obtain the downsampled feature map output by each downsampling network layer.

[0107] Optionally, the specific process of downsampling processing may be: through each downsampling network layer, convolution activation processing is performed on the downsampling feature map output by the previous downsampling network layer, and feature upsampling is performed on the processed feature map to obtain the downsampling feature map output by each downsampling network layer. It should be noted that the downsampling processing process of the first downsampling network layer can refer to the specific process of downsampling processing described above, which will not be repeated here.

[0108] Specifically, each downsampling network layer may include a convolution layer and a second activation function layer connected in sequence, so the convolution activation process is also the process of the convolution layer and the second activation function layer. In a specific implementation, the activation function in the second activation function layer may be a ReLU function, a LeakyReLU function (an activation function related to the Sigmoid function and the more common ReLU function), or other activation functions, which are not limited.

[0109] In one possible implementation, the residual network may include a residual convolution layer, a residual pooling layer, and a residual merging layer connected in sequence; then, the specific method of performing residual feature extraction processing on the initial hidden feature map through the residual network to obtain the target hidden feature map may be: inputting the initial hidden feature map into the residual network, performing residual convolution processing on the initial hidden feature map through the residual convolution layer and performing residual pooling processing through the residual pooling layer to obtain the processed hidden feature map; finally, merging the processed hidden feature map with the initial hidden feature map through the residual merging layer to obtain the target hidden feature map.

[0110] Optionally, the network structure of the residual network may also be other network structures that can utilize the feature map output by the previous network layer for subsequent feature processing, and there is no limitation on this.

[0111] S405, performing activation mapping processing on each pixel in the target hidden feature map through the activation network to obtain a encrypted image.

[0112] In the embodiment of the present application, the specific implementation of step S405 can refer to the specific implementation of activating the network in step S302 in the above embodiment, which will not be repeated here.

[0113] S406: Perform image merging processing on the reference image area contained in the encrypted image and the background image to obtain a merged image area.

[0114] In the embodiment of the present application, the specific implementation of step S406 can refer to the specific implementation of step S303 in the above embodiment, which will not be repeated here.

[0115] S407: Replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

[0116] In the embodiment of the present application, the specific implementation of step S407 can refer to the specific implementation of step S304 in the above embodiment, which will not be repeated here.

[0117] In practical applications, the feature extraction network may include two convolutional pooling layers (corresponding to the aforementioned dimension reduction convolutional layers) connected in sequence, each convolutional pooling layer includes a convolutional layer and a pooling layer; each downsampling network layer may include a convolutional layer and a LeakyReLU activation function layer, and each upsampling network layer may include a deconvolution layer, a ReLU activation function layer, and a normalization layer. The activation network may include a Tanh activation function and a built-in pixel mapping function.

[0118] Among them, the convolution kernels of the two convolutional pooling layers are 3x3 and the step size is 1; and the convolution kernels of the pooling layers in the two convolutional pooling layers are 2x2 and the step size is 2; the residual network includes four downsampling network layers, and the convolution kernels of the corresponding convolutional layers are 4x4 and the step size is 2, and the number of feature channels are 64, 128, 256, and 512 respectively. The residual network includes five upsampling network layers, and the convolution kernels of the corresponding deconvolutional layers are 4x4 and the step size is 2, and the number of feature channels are 512, 256, 128, 64, and 3 respectively.

[0119] Then, the image hiding model hides the secret image S with a size of 256x 256 into the carrier image C with a size of 64x64, and the process of obtaining the secret image C' can be:

[0120] 1) Input the secret image S into the image hiding model and pass it through the first convolutional pooling layer to obtain the feature map after convolutional pooling processing;

[0121] 2) The feature map after convolution and pooling is passed through the second convolution and pooling layer to obtain the reduced dimension feature map corresponding to the secret image S; at this time, the size of the reduced dimension feature map is 64x 64;

[0122] 3) Through the fusion network, the reduced dimension feature map corresponding to the secret image S is fused with the carrier image C at the level of feature channels to obtain the initial hidden feature map corresponding to the secret image S;

[0123] 4) Inputting the initial hidden feature map corresponding to the secret image S into the residual network, the residual network performs residual feature extraction processing on the initial hidden feature map corresponding to the secret image S to obtain the target hidden feature map corresponding to the secret image S;

[0124] 5) The target hidden feature map is processed in turn through the Tanh activation function and the built-in pixel mapping function to obtain the encrypted image C'.

[0125] After obtaining the confidential image C', the confidential image C' and the reference image area contained in the background image B can be merged to obtain a merged image area H; then, the reference transparency a can be obtained to modify the transparency of the merged image area H to a through the alpha synthesis algorithm to obtain the processed merged image area H.

[0126] Finally, the reference image region of the background image B is replaced with the processed merged image region H to obtain the stego-image C".

[0127] In the embodiment of the present application, the feature extraction network in the image hiding model can extract the reduced-dimensional feature map of the secret image with the same size as the carrier image, so that the subsequent merging network can realize the preliminary merging of the secret image and the carrier image by merging the reduced-dimensional feature map and the carrier image. Finally, since the target hidden feature map is only a feature map, it is necessary to activate the network to perform activation mapping processing on it, and finally obtain the secret image that completely hides the secret image. Therefore, the embodiment of the present application can embed the secret image into a carrier image with a smaller size than the secret image through the image hiding model, thereby improving the embedding rate of image steganography.

[0128] At the same time, the embodiment of the present application can complete the feature fusion between the reduced-dimensional feature map and the carrier image from the feature channel level through the fusion network, so that the feature fusion will not cause the loss of feature information of the reduced-dimensional feature map or the carrier image. In addition, the embodiment of the present application also performs multiple downsampling and upsampling through the residual network, and fuses the corresponding results, which is conducive to making full use of the high-dimensional features and low-dimensional features of the image, so that the feature fusion between the reduced-dimensional feature map of the secret image and the carrier image is more complete, that is, the secret information carried by the secret image can be more fully embedded in the carrier image, which is conducive to further improving the embedding rate of image steganography.

[0129] In one embodiment of the present application, another image processing method is provided, which can be applied to Figure 1 The image processing system shown in the figure, the method can be executed by the terminal device or the server, or can be executed by the terminal device and the server together. In the embodiment of the present application, the method is executed by the server as an example for explanation. Figure 5 FIG. 4 is a flow chart showing another image processing method. Figure 3 and Figure 4 This paper expands on the method shown in .

[0130] Among them, S501 to S508 are described in detail as follows:

[0131] S501, obtaining a secret image and a carrier image.

[0132] In the embodiment of the present application, the specific implementation of step S501 can refer to the specific implementation of step S301 in the above embodiment, which will not be repeated here.

[0133] S502: Call the image hiding model to hide the secret image in the carrier image to obtain the secret-carrying image.

[0134] In the embodiment of the present application, the specific implementation of step S502 can refer to the specific implementation of step S302 and steps S402 to S405 in the above embodiment, which will not be repeated here.

[0135] S503: Perform image merging processing on the reference image area contained in the encrypted image and the background image to obtain a merged image area.

[0136] In the embodiment of the present application, the specific implementation of step S503 can refer to the specific implementation of step S303 in the above embodiment, which will not be repeated here.

[0137] S504: Replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

[0138] In the embodiment of the present application, the specific implementation of step S504 can refer to the specific implementation of step S304 in the above embodiment, which will not be repeated here.

[0139] S505: Obtain a target image region containing a secret image from the stego-image.

[0140] In the embodiment of the present application, if the merged image area is a completely transparent image area after the transparency change processing, it is difficult to directly obtain the target image area from the stego image; therefore, the specific method of obtaining the target image area containing the secret image from the stego image can be: through a preset image tampering detection method, the stego image is detected, and the target image area containing the secret image in the stego image is determined.

[0141] Specifically, the preset image tampering detection method can be one or more of a variety of image tampering detection methods such as a feature point-based tampering detection method, a region generation network (Faster-RCNN network), and Mantra-Net (an end-to-end network that does not require preprocessing and post-processing, can perform positioning and detection, and is a fully convolutional neural network), without limitation.

[0142] S506: Calling the image restoration model to extract the scaling feature map of the target image area, and performing inverse feature extraction processing on the scaling feature map to obtain a restored image of the secret image.

[0143] In an embodiment of the present application, the image restoration model may include a plurality of convolutional layers connected in sequence, and the number of feature channels of the convolution kernels in the plurality of convolutional layers presents a quantity fluctuation. The quantity fluctuation may be a quantity fluctuation in which the number of feature channels of the convolution kernels in the plurality of convolutional layers first increases and then decreases, or a quantity fluctuation in which the number first decreases and then increases, or a quantity fluctuation in which the number first increases and then decreases, then increases and then decreases. Optionally, the quantity fluctuation may be in other ways, which are not limited thereto.

[0144] In one embodiment, a specific method of calling an image restoration model to extract a scaling feature map of a target image area may be: performing convolution processing on a feature map output by a previous convolution layer of each convolution layer through each of the multiple convolution layers in the image restoration model to obtain a feature map output by each convolution layer; and using a feature map output by the last convolution layer of the multiple convolution layers as a scaling feature map.

[0145] Specifically, in order to enlarge the feature map with high quality, the network structure of some convolutional layers in the multiple convolutional layers can be the same as or similar to the structure of super-resolution reconstruction networks such as SRCNN (Super Resolution Convolutional Network) and FSRCNN (Fast Super Resolution Convolutional Network).

[0146] In one embodiment, a plurality of convolutional layers in the image restoration model may be connected to a deconvolutional layer, so that the deconvolutional layer may be used to perform inverse feature extraction processing on the scaling feature map to obtain a restored image of the secret image.

[0147] In a specific implementation, the image restoration model may include thirteen convolutional layers and one deconvolution layer connected in sequence; wherein, the convolution kernel size of the first six convolutional layers among the thirteen convolutional layers is 3x3, and the step size is 1; the convolution kernel size of the seventh convolutional layer is 5x5, and the step size is 1; the convolution kernel size of the eighth to thirteenth convolutional layers is 3x3, and the step size is 1; the deconvolution kernel size of the last deconvolution layer is 9x9, and the step size is 4; the number of feature channels of the above fourteen convolution kernels are 80, 160, 320, 160, 80, 3, 56, 12, 12, 12, 12, 12, 56, and 3 respectively.

[0148] Then, referring to the example in step S407, after obtaining the stego-image C", the target image region Ccut containing the secret image can be obtained from the stego-image; then, the target image region Ccut is input into the image recovery model:

[0149] 1) The first convolutional layer in the image restoration model processes the target image region Ccut and obtains the first restored feature map S'(1) after passing the BatchNorm normalization function and the ReLU activation function;

[0150] 2) The first restored feature map S'(1) is input into the second convolutional layer in the image restoration model for processing, and after the BatchNorm normalization function and the ReLU activation function, a second restored feature map S'(2) is obtained;

[0151] 3) The second restored feature map S'(2) is input into the third convolutional layer in the image restoration model for processing, and after the BatchNorm normalization function and the ReLU activation function, a third restored feature map S'(3) is obtained;

[0152] 4) Similar to the previous step, repeat three more times to obtain the sixth restored feature map S'(6);

[0153] 5) Inputting the sixth restored feature map S'(6) into the seventh convolutional layer in the image restoration model for processing, and processing it through the PReLU activation function to obtain the seventh restored feature map S'(7);

[0154] 6) Inputting the seventh restored feature map S'(7) into the eighth convolutional layer in the image restoration model for processing, and processing it through the PReLU activation function to obtain the eighth restored feature map S'(8);

[0155] 7) Similar to the previous step, repeat five more times to obtain the thirteenth restored feature map S'(13);

[0156] 8) The thirteenth restored feature map S'(13) is input into the last deconvolution layer in the image restoration model for processing to obtain the restored secret image S'.

[0157] In practical applications, please refer to the attached Figure 6 , shows a schematic diagram of the image steganography process. Figure 6 As shown, the secret image 601 and the carrier image 602 are input into the image hiding model, and then the transparency of the secret image output by the image hiding model and the reference image area contained in the background image is changed by the alpha algorithm, so that the secret image 603 with a transparency of 70% can be obtained. Then, the reference image area in the background image is replaced with the secret image 603 to obtain the steganographic image 604.

[0158] After obtaining the stego image 604, Figure 6 As shown, the image region including the secret image 603 in the steganographic image 604 can be input into the image restoration model and processed by the image restoration model to obtain a restored image 605 of the secret image 601.

[0159] In the embodiment of the present application, after obtaining the stego image, the target image region containing the secret image in the stego image can be processed by the image restoration model to obtain the restored image of the secret image. It can be seen that the embodiment of the present application can accurately restore the restored image of the secret image from the stego image through the image restoration model, thereby accurately reading the secret information in the stego image and achieving the purpose of secure information transmission.

[0160] In addition, the image restoration model can make full use of the high-dimensional features and low-dimensional features of the target image area by extracting the telescopic feature map of the target image area, so that the information about the secret image can be fully read from the target image area, which is conducive to more accurate restoration of the secret image and obtaining a high-quality restored image.

[0161] In one embodiment of the present application, another image processing method is provided, which can be applied to Figure 1 The image processing system shown in the figure, the method can be executed by the terminal device or the server, or can be executed by the terminal device and the server together. In the embodiment of the present application, the method is executed by the server as an example for explanation. Figure 7 FIG. 4 is a flow chart showing another image processing method. Figure 3 , Figure 4 and Figure 5 This paper expands on the method shown in .

[0162] Among them, S701 to S712 are described in detail as follows:

[0163] S701, obtaining a secret training image, a carrier training image and a background training image.

[0164] In the embodiment of the present application, the number of secret training images, carrier training images and background training images may include multiple. Specifically, the secret training images, carrier training images and background training images can be selected from image datasets such as the COCO dataset (Microsoft Common Objects in Context) and the ImageNet dataset (a large visualization database for visual object recognition software research).

[0165] Optionally, after selecting multiple images from the image dataset, the images used as carrier training images can be adjusted to a smaller size (such as 64x 64); at the same time, the images used as secret training images and the images used as background training images can be adjusted to a larger size than the carrier training images, such as the sizes of the secret training images and the background training images can be adjusted to 256x 256.

[0166] S702: Call the image hiding model to be trained, hide the secret training image into the carrier training image, and obtain the carrier training image.

[0167] In an embodiment of the present application, the specific process of calling the image hiding model to be trained, hiding the secret training image into the carrier training image, and obtaining the carrier training image may be: extracting the dimension reduction training feature map of the secret training image through the feature extraction network in the image hiding model to be trained, wherein the size of the dimension reduction training feature map is the same as the size of the carrier training image; then, through the merging network in the image hiding model to be trained, the dimension reduction training feature map and the carrier training image are feature merged to obtain the target hidden training feature map; finally, through the activation network in the image hiding model to be trained, activation mapping processing is performed on each pixel in the target hidden training feature map to obtain the carrier training image;

[0168] It should be noted that the specific implementation of the feature extraction network, the merging network and the activation network in step S702 can also refer to the specific implementation of step S302 and steps S402 to S405 in the above embodiment, which will not be repeated here.

[0169] In a possible implementation, in order to improve the robustness of the image steganography model and enhance the ability of the image steganography model to resist interference, after obtaining the secret training image, the secret training image can also be subjected to pixel interference processing to obtain an updated secret training image.

[0170] Specifically, the specific method of pixel interference processing can be adding Gaussian noise to each pixel in the dense training image, randomly erasing or randomly cropping part of the image area in the dense training image, or any other method that can interfere with the pixels in the dense training image, which is not limited here.

[0171] S703: Perform image merging processing on the reference image area of ​​the density-carrying training image and the background training image to obtain a training image area.

[0172] In the embodiment of the present application, the specific implementation of the image merging process in step S703 can also refer to the specific implementation of step S303 in the above embodiment, which will not be repeated here.

[0173] Optionally, since it is mentioned in step S702 that pixel interference processing can also be performed on the dense training image to obtain an updated dense training image; therefore, the specific method of performing image merging processing on the reference image area of ​​the dense training image and the background training image to obtain the training image area can also be: performing image merging processing on the updated reference image area of ​​the dense training image and the background training image to obtain the training image area.

[0174] S704: Replace the reference image area of ​​the background image with the training image area to obtain a steganographic training image.

[0175] In the embodiment of the present application, the specific implementation of replacing the reference image area of ​​the background image with the training image area in step S704 can also refer to the specific implementation of step S304 in the above embodiment, which is not repeated here.

[0176] S705 , calling the image restoration model to be trained, performing image restoration processing on the steganographic training image, and obtaining a restored image of the secret training image.

[0177] In the embodiment of the present application, the specific implementation of the image restoration process in step S705 can refer to the specific implementation of step S506 in the above embodiment, which will not be repeated here.

[0178] S706: Perform joint training of the image steganography model based on the secret training image, the carrier training image, the secret carrier training image and the restored image of the secret training image.

[0179] In an embodiment of the present application, a specific method for jointly training an image steganography model based on a secret training image, a carrier training image, a secret-carrying training image, and a restored image of the secret training image may be: determining a first loss value based on the secret training image and the restored image of the secret training image; determining a second loss value based on the carrier training image and the secret-carrying training image; and training the image steganography model based on the first loss value and the second loss value.

[0180] Specifically, the first loss value between the secret training image and the restored image of the secret training image, and the second loss value between the carrier training image and the secret training image can be calculated by one or more of the loss functions such as the cross entropy loss function, the logarithmic loss function, and the square difference loss function.

[0181] In actual application, please refer to the attached Figure 8 , shows a schematic diagram of the training process of an image steganography model. Figure 8 As shown in step 801 in , a data set including multiple secret training images, a data set including multiple carrier training images, and a data set including multiple background training graphics can be first prepared, and the three data sets can be divided into a training set, a validation set, and a test set according to a preset ratio. Specifically, 20,000 images can be randomly selected from the COCO data set and the ImageNet data set; then the images used as secret training images and background training images are adjusted to the same size (such as 256x 256), and the images used as carrier training images are adjusted to a smaller size (such as 64x 64), and then the data set containing 20,000 images is divided into a training set, a validation set, and a test set in a non-overlapping manner according to a preset ratio. Among them, the preset ratio can be 8:1:1 or 7:2:1, etc., the ratio of the training set, the validation set, and the test set, which is not limited to this.

[0182] Optionally, since more data in the training set is needed to train a more accurate model, the proportion of the training set can be set to be greater than the proportion of the validation set and the proportion of the test set, or the proportion of the training set can be set to be greater than the sum of the proportions of the validation set and the test set.

[0183] Then, as shown in step 802, an image steganography model to be trained consisting of an image hiding model and an image restoration model may be constructed.

[0184] Afterwards, as shown in step 803, the image steganography model to be trained can be end-to-end trained and parameter tuned using the training set and the validation set, and an intermediate interference layer can be added to form adversarial training.

[0185] Specifically, end-to-end training refers to the stego image obtained based on the secret training image and the background training image output by the image hiding model, which will be processed by the image recovery model. At the same time, in the process of building the image steganography model, various relevant parameters need to be set, such as setting the size of the relevant image, setting the iteration cycle of training, setting the optimizer of the network, setting the learning rate, setting the weights of different networks, setting the scaling factor (i.e., the multiple of the secret training image reduction), setting the number of images processed in the same batch, setting the loss function, setting the training strategy, etc. Therefore, the image hiding model and the image recovery model in the image steganography model can be jointly trained through the first loss value and the second loss value obtained during the training process, so as to tune the relevant parameters in the image steganography model so that the image steganography model can hide the secret image more covertly and recover the secret image with high quality.

[0186] In addition, adding an intermediate interference layer means that after obtaining the secret training image, the secret training image can be processed using pixel interference methods such as Gaussian noise, random erasing, and random cropping to obtain an updated secret training image; and then the updated secret training image is input into a subsequent image recovery model to obtain a restored image of the secret training image.

[0187] Finally, as shown in step 804, the performance of the trained image steganography model can be tested on the test set, and the image steganography model can be evaluated on multiple evaluation indicators. Furthermore, the image steganography model that passes the evaluation can be used as the image steganography model to be finally used.

[0188] Specifically, the evaluation indicators may include visual information fidelity VIF, peak signal-to-noise ratio PSNR, and structural similarity SSIM value, etc.; the evaluation process may specifically be: calculating the visual information fidelity VIF, peak signal-to-noise ratio PSNR, and structural similarity SSIM value, etc. between the secret training image output based on the verification set and the corresponding carrier training image in the verification set, and calculating the visual information fidelity VIF, peak signal-to-noise ratio PSNR, and structural similarity SSIM value, etc. between the restored image of the secret training image output based on the verification set and the corresponding secret training image in the verification set; finally, the calculated values ​​may be weighted added to obtain the evaluation result of the trained image steganography model.

[0189] S707: Obtain the secret image and the carrier image.

[0190] In the embodiment of the present application, the specific implementation of step S707 can refer to the specific implementation of step S301 in the above embodiment, which will not be repeated here.

[0191] S708: Call the image hiding model to hide the secret image in the carrier image to obtain the secret image.

[0192] In the embodiment of the present application, the specific implementation of step S708 can refer to the specific implementation of step S302 and steps S402 to S405 in the above embodiment, which will not be repeated here.

[0193] S709: Perform image merging processing on the reference image area contained in the encrypted image and the background image to obtain a merged image area.

[0194] In the embodiment of the present application, the specific implementation of step S709 can refer to the specific implementation of step S303 in the above embodiment, which will not be repeated here.

[0195] S710: Replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

[0196] In the embodiment of the present application, the specific implementation of step S710 can refer to the specific implementation of step S304 in the above embodiment, which will not be repeated here.

[0197] S711. Obtain a target image region containing a steganographic image from the steganographic image.

[0198] In the embodiment of the present application, the specific implementation of step S711 can refer to the specific implementation of step S505 in the above embodiment, which will not be repeated here.

[0199] S712. Calling an image restoration model to extract a scaling feature map of a target image region, and performing inverse feature extraction processing on the scaling feature map to obtain a restored image of the secret image.

[0200] In the embodiment of the present application, the specific implementation of step S712 can refer to the specific implementation of step S506 in the above embodiment, which will not be repeated here.

[0201] In the embodiment of the present application, by jointly training the image hiding model and the image restoration model, the image hiding model can better hide the secret image, and the image restoration model can also restore the restored image of the secret image with high quality, which is conducive to improving the accuracy and practicality of the image steganography model. In addition, by adding pixel interference processing between the image hiding model and the image restoration model, adversarial interference training of the image steganography model can be achieved, thereby improving the robustness of the image steganography model, which is conducive to further improving the practicality of the image steganography model.

[0202] Here, an apparatus embodiment of the present application is introduced, which can be used to execute the image processing method in the above-mentioned embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the above-mentioned embodiment of the image processing method of the present application.

[0203] The present application embodiment provides an image processing device, such as Fig. 9 As shown, the device includes an acquisition unit 901 and an image hiding unit 902, wherein:

[0204] An acquisition unit 901 is used to acquire a secret image and a carrier image, wherein the size of the secret image is larger than the size of the carrier image;

[0205] The image hiding unit 902 is used to call the image hiding model to hide the secret image in the carrier image to obtain the secret image; wherein the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract a reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to merge the reduced dimension feature map with the carrier image to obtain a target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image;

[0206] The image hiding unit 902 is further used to perform image merging processing on the reference image area contained in the encrypted image and the background image to obtain a merged image area;

[0207] The image hiding unit 902 is further used to replace the reference image area of ​​the background image with the merged image area to obtain a stego-image.

[0208] In one embodiment of the present application, the merging network includes a fusion network and a residual network connected in sequence; based on the aforementioned scheme, when the image hiding unit 902 calls the image hiding model and merges the reduced dimension feature map with the carrier image to obtain a target hidden feature map, it can be specifically used for: performing feature fusion of the reduced dimension feature map and the carrier image through the fusion network to obtain an initial hidden feature map; the size of the initial hidden feature map is the same as the size of the carrier image; performing residual feature extraction processing on the initial hidden feature map through the residual network to obtain a target hidden feature map.

[0209] In one embodiment of the present application, the residual network includes a downsampling subnetwork and an upsampling subnetwork connected in sequence, and the downsampling subnetwork includes a plurality of downsampling network layers connected in sequence; wherein the signal output end of each downsampling network layer is respectively connected to the signal input end of the upsampling subnetwork, so that the upsampling subnetwork performs upsampling feature extraction processing according to the downsampling feature maps output by each downsampling network layer to obtain a target hidden feature map.

[0210] In one embodiment of the present application, the upsampling subnetwork includes a plurality of upsampling network layers connected in sequence, and each downsampling network layer in the downsampling subnetwork corresponds to each upsampling network layer in the upsampling subnetwork except the first upsampling network layer, wherein: the first upsampling network layer is used to perform upsampling processing on the downsampling feature map output by the last downsampling network layer; the signal input end of each upsampling network layer except the first upsampling network layer in the plurality of upsampling network layers is also connected to the signal output end of the corresponding downsampling network layer in the downsampling subnetwork, so as to perform upsampling processing on the fused feature map after feature fusion of the received input signal.

[0211] In one embodiment of the present application, the image processing device also includes an image restoration unit 903; based on the aforementioned scheme, the image restoration unit 903 can be used to: obtain a target image area containing a secret image from a stego image; call an image restoration model to extract a scaling feature map of the target image area, and perform inverse feature extraction processing on the scaling feature map to obtain a restored image of the secret image; wherein the image restoration model includes a plurality of convolutional layers connected in sequence, and the number of feature channels of the convolution kernels in the plurality of convolutional layers fluctuates in number.

[0212] In one embodiment of the present application, an image steganography model is composed of an image hiding model and an image restoration model connected in sequence; based on the aforementioned scheme, the image processing device also includes a training unit 904; based on the aforementioned scheme, the training unit 904 can be used to: obtain a secret training image, a carrier training image and a background training image; call the image hiding model to be trained, hide the secret training image into the carrier training image to obtain a secret training image, and perform image merging processing on the reference image area of ​​the secret training image and the background training image to obtain a training image area, and replace the reference image area of ​​the background image with the training image area to obtain a steganography training image; call the image restoration model to be trained, perform image restoration processing on the steganography training image to obtain a restored image of the secret training image; and perform joint training of the image steganography model based on the secret training image, the carrier training image, the secret training image and the restored image of the secret training image.

[0213] In one embodiment of the present application, based on the aforementioned scheme, the training unit 904 can be further used to: perform pixel interference processing on the steganographic training image to obtain an updated steganographic training image, and perform joint training of the image steganography model based on the updated steganographic training image.

[0214] In one embodiment of the present application, based on the aforementioned scheme, the image hiding unit 902 can be further used to: perform transparency change processing on the merged image area to obtain a processed merged image area; when the image hiding unit 902 replaces the reference image area of ​​the background image with the merged image area to obtain a steganographic image, it can be specifically used to: replace the reference image area of ​​the background image with the processed merged image area to obtain a steganographic image.

[0215] It should be noted that the device provided in the above embodiment and the method provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment and will not be repeated here.

[0216] The device provided in the above embodiment can be arranged in a terminal device or in a server. The device provided in the embodiment of the present application can extract a reduced-dimensional feature map of a secret image of the same size as the carrier image through the feature extraction network in the image hiding model, so that the subsequent merging network can realize the preliminary merging of the secret image and the carrier image by merging the reduced-dimensional feature map and the carrier image. Finally, since the target hidden feature map is only a feature map, it is necessary to activate the network to perform activation mapping processing on it, and finally obtain a secret image that completely hides the secret image. Therefore, the embodiment of the present application can embed the secret image into a carrier image smaller in size than the secret image through the image hiding model, thereby improving the embedding rate of image steganography. In addition, by flexibly embedding the secret image into the reference image area in the background image to form a stego image, the flexibility of image steganography and the robustness to interference such as cropping are improved.

[0217] An embodiment of the present application also provides an electronic device, comprising one or more processors and a storage device, wherein the storage device is used to store one or more computer programs, and when the one or more computer programs are executed by one or more processors, the electronic device implements the above image processing method.

[0218] Fig.10 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present application is shown.

[0219] It should be noted that Fig.10 The computer system 1000 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0220] like Fig.10 As shown, the computer system 1000 includes a processor (Central Processing Unit, CPU) 1001, which can perform various appropriate actions and processes according to the program stored in the read-only memory (Read-Only Memory, ROM) 1002 or the program loaded from the storage part 1008 to the random access memory (Random Access Memory, RAM) 1003, such as executing the method in the above embodiment. In RAM 1003, various programs and data required for system operation are also stored. CPU 1001, ROM 1002 and RAM 1003 are connected to each other through bus 1004. Input / output (Input / Output, I / O) interface 1005 is also connected to bus 1004.

[0221] In some embodiments, the following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0222] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1009, and / or installed from a removable medium 1011. When the computer program is executed by a processor (CPU) 1001, various functions defined in the system of the present application are executed.

[0223] It should be noted that the computer-readable medium shown in the embodiment of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (Erasable Programmable Read Only Memory), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. A computer program contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0224] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the devices, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and a computer program.

[0225] The units or modules involved in the embodiments described in this application may be implemented by software or hardware, and the units or modules described may also be set in a processor. The names of these units or modules do not, in some cases, constitute limitations on the units or modules themselves.

[0226] Another aspect of the present application further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the image processing method as described above is implemented. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device.

[0227] Another aspect of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of an electronic device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the electronic device performs the image processing method as described above in the above embodiments.

[0228] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0229] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0230] The above content is only a preferred exemplary embodiment of the present application and is not intended to limit the implementation scheme of the present application. A person of ordinary skill in the art can easily make corresponding changes or modifications based on the main concept and spirit of the present application. Therefore, the scope of protection of the present application shall be based on the scope of protection required by the claims.

Claims

1. An image processing method, characterized in that: include: Acquire a secret image and a carrier image, wherein the size of the secret image is larger than the size of the carrier image; Calling an image hiding model to hide the secret image in the carrier image to obtain a secret image; wherein the image hiding model comprises a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract a reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to perform feature merging of the reduced dimension feature map with the carrier image to obtain a target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image; Performing image merging processing on the reference image area contained in the secret image and the background image to obtain a merged image area; The reference image area of ​​the background image is replaced by the merged image area to obtain a steganographic image.

2. The method according to claim 1, characterized in that The merging network includes a fusion network and a residual network connected in sequence; the calling of the image hidden model to merge the reduced dimension feature map with the carrier image to obtain a target hidden feature map includes: The reduced-dimensional feature map and the carrier image are subjected to feature fusion through the fusion network to obtain an initial hidden feature map; the size of the initial hidden feature map is the same as the size of the carrier image; The residual feature extraction process is performed on the initial hidden feature map through the residual network to obtain the target hidden feature map.

3. The method according to claim 2, characterized in that The residual network includes a downsampling subnetwork and an upsampling subnetwork connected in sequence, and the downsampling subnetwork includes a plurality of downsampling network layers connected in sequence, wherein: The signal output end of each downsampling network layer is respectively connected to the signal input end of the upsampling sub-network, so that the upsampling sub-network performs upsampling feature extraction processing according to the downsampling feature maps output by each downsampling network layer to obtain the target hidden feature map.

4. The method according to claim 3, characterized in that The upsampling subnetwork includes a plurality of upsampling network layers connected in sequence, and each downsampling network layer in the downsampling subnetwork corresponds to each upsampling network layer in the upsampling subnetwork except the first upsampling network layer, wherein: The first upsampling network layer is used to upsample the downsampled feature map output by the last downsampling network layer; The signal input end of each upsampling network layer in the multiple upsampling network layers except the first upsampling network layer is also connected to the signal output end of the corresponding downsampling network layer in the downsampling sub-network, so as to perform upsampling processing on the fused feature map after performing feature fusion on the received input signal.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire a target image region containing the steganographic image from the steganographic image; An image restoration model is called to extract a scaling feature map of the target image area, and an inverse feature extraction process is performed on the scaling feature map to obtain a restored image of the secret image; wherein the image restoration model includes a plurality of convolutional layers connected in sequence, and the number of feature channels of the convolutional kernels in the plurality of convolutional layers fluctuates in number.

6. The method according to claim 5, characterized in that The image hiding model and the image restoration model connected in sequence form an image steganography model; the method further comprises: obtaining a secret training image, a carrier training image, and a background training image; Calling the image hiding model to be trained, hiding the secret training image into the carrier training image to obtain a secret training image, performing image merging processing on the secret training image and the reference image area of ​​the background training image to obtain a training image area, and replacing the reference image area of ​​the background image with the training image area to obtain a steganographic training image; Calling the image restoration model to be trained, performing image restoration processing on the steganographic training image, and obtaining a restored image of the secret training image; Based on the secret training image, the carrier training image, the carrier training image and the restored image of the secret training image, a joint training of an image steganography model is performed.

7. The method according to claim 6, characterized in that The method further comprises: Pixel interference processing is performed on the secret training image to obtain an updated secret training image, and joint training of the image steganography model is performed based on the updated secret training image.

8. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Performing transparency change processing on the merged image area to obtain a processed merged image area; The step of replacing the reference image area of ​​the background image with the merged image area to obtain the steganographic image comprises: The reference image area of ​​the background image is replaced by the processed merged image area to obtain the steganographic image.

9. An image processing device, characterized in that: It includes an acquisition unit and an image hiding unit, wherein: The acquisition unit is used to acquire a secret image and a carrier image, wherein the size of the secret image is larger than the size of the carrier image; The image hiding unit is used to call the image hiding model to hide the secret image in the carrier image to obtain the secret image; wherein the image hiding model includes a feature extraction network, a merging network and an activation network connected in sequence, the feature extraction network is used to extract a reduced dimension feature map of the secret image, the size of the reduced dimension feature map is the same as the size of the carrier image, the merging network is used to perform feature merging of the reduced dimension feature map with the carrier image to obtain a target hidden feature map, and the activation network is used to perform activation mapping processing on each pixel in the target hidden feature map to obtain the secret image; The image hiding unit is further used to perform image merging processing on the reference image area contained in the secret image and the background image to obtain a merged image area; The image hiding unit is further used to replace the reference image area of ​​the background image with the merged image area to obtain a steganographic image.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the image processing method according to any one of claims 1 to 8 is implemented.

11. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the image processing method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product comprises a computer program, and the computer program is suitable for being loaded by a processor and executing the image processing method according to any one of claims 1 to 8.

Images