Image processing method and device, electronic equipment and storage medium

Abstract

The invention provides an image processing method and device, electronic equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring a to-be-processed source domain image, wherein the source domain image is an image containing a human face; converting the source domain image into a first confrontation domain image through the trained generator; converting the first confrontation domain image into a non-confrontation domain image through a trained regularization module; and covering the source domain image through the non-adversarial domain image to obtain a processed image. The source domain image can be converted to the first adversarial domain through the trained generator, so that the adversarial example corresponding to the to-be-processed source domain image is obtained, the face image is protected, then the visual effect can be ensured while the image protection effect is ensured through the trained regularization module, and the user experience is improved. Therefore, according to the embodiment of the invention, the reliability of face protection by image processing can be ensured, and the user experience is improved.

Description

Background Art

[0002] Face recognition is a biometric technology that identifies people based on their facial features. It is achieved by extracting facial features through deep neural networks. With the development of computer technology, the application of face recognition technology has become more and more extensive. Since face recognition has the risk of privacy leakage, once the user's facial image information is leaked, it may be used maliciously and endanger the user's interests and safety. Therefore, it is necessary to process the face image to ensure user safety.

[0003] In related technologies, some studies use adversarial examples to protect photos from being recognized by unauthorized face recognition systems. However, existing methods for adversarial face images have many limitations and cannot ensure the visual effect while ensuring image protection. Therefore, there is an urgent need for an image processing technology that improves the reliability of face protection and improves the visual effect of processed images.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides an image processing method, device, electronic device and storage medium, which at least to some extent overcome the problem in the related art that it is impossible to ensure the visual effect while ensuring the image protection effect.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, there is provided an image processing method, comprising: obtaining a source domain image to be processed, the source domain image being an image containing a face; converting the source domain image into a first adversarial domain image by a trained generator; converting the first adversarial domain image into a non-adversarial domain image by a trained regularization module; and overlaying the source domain image with the non-adversarial domain image to obtain a processed image.

[0008] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: training a generator based on a first discriminator and a second discriminator to obtain the trained generator; wherein the first discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the source domain sample, and the second discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the reference domain sample, and the reference domain is a preset image processing target domain.

[0009] In some exemplary embodiments, training a generator based on a first discriminator and a second discriminator to obtain the trained generator includes: iteratively executing the following steps until the generator meets a first preset end condition to obtain the trained generator: freezing the parameters of the generator, and updating the parameters of the first discriminator and the second discriminator so that the discriminator loss functions corresponding to the first discriminator and the second discriminator meet the maximization condition; freezing the parameters of the first discriminator and the second discriminator, and updating the parameters of the generator so that the generator loss function corresponding to the generator meets the minimization condition.

[0010] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: training a regularization module by using the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module.

[0011] In some exemplary embodiments, the regularization module is trained by the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module, including: obtaining a cycle-consistent model according to the combination of the trained generator, the regularization module, the first autoencoder, and the second autoencoder, wherein the first autoencoder is used to perform image conversion from the non-adversarial domain to the second adversarial domain, and the second autoencoder is used to perform image conversion from the second adversarial domain to the source domain; by calculating at least one of the cycle-consistent loss and the visual consistency loss of the cycle-consistent model, the regularization module is iteratively trained until a second preset end condition is met to obtain the trained regularization module.

[0012] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: forming a mobility enhancement module based on multiple trained face recognition models; obtaining a diversity sample set through the mobility enhancement module; jointly training the trained generator and the trained regularization module based on the diversity sample set to obtain a target generator and a target regularization module; wherein, the image processing method provided by the embodiments of the present disclosure includes: converting the source domain image into the first adversarial domain image through the target generator; converting the first adversarial domain image into the non-adversarial domain image through the target regularization module.

[0013] According to another aspect of the present disclosure, an image processing device is also provided, including: a source domain image acquisition module, used to acquire a source domain image to be processed, wherein the source domain image is an image containing a human face; a first adversarial domain image conversion module, used to convert the source domain image into a first adversarial domain image through a trained generator; a non-adversarial domain image conversion module, used to convert the first adversarial domain image into a non-adversarial domain image through a trained regularization module; and an image covering module, used to cover the source domain image with the non-adversarial domain image to obtain a processed image.

[0014] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure also includes: a generator training module, used to train the generator based on a first discriminator and a second discriminator to obtain the trained generator; wherein the first discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the source domain sample, and the second discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the reference domain sample, and the reference domain is a preset image processing target domain.

[0015] In some exemplary embodiments, the generator training module is used to iteratively execute the following steps until the generator meets a first preset end condition to obtain the trained generator: freezing the parameters of the generator, and updating the parameters of the first discriminator and the second discriminator so that the discriminator loss function corresponding to the first discriminator and the second discriminator meets the maximization condition; freezing the parameters of the first discriminator and the second discriminator, and updating the parameters of the generator so that the generator loss function corresponding to the generator meets the minimization condition.

[0016] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure further includes: a regularization module training module, which is used to train the regularization module through the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module.

[0017] In some exemplary embodiments, a regularization module training module is used to obtain a cycle-consistent model based on the trained generator, regularization module, first autoencoder, and second autoencoder combination, wherein the first autoencoder is used to perform image conversion from the non-adversarial domain to the second adversarial domain, and the second autoencoder is used to perform image conversion from the second adversarial domain to the source domain; by calculating at least one of the cycle consistency loss and the visual consistency loss of the cycle-consistent model, the regularization module is iteratively trained until a second preset end condition is met to obtain the trained regularization module.

[0018] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure also includes: a joint training module, used to form a transferability enhancement module based on multiple trained face recognition models; obtain a diversity sample set through the transferability enhancement module; jointly train the trained generator and the trained regularization module according to the diversity sample set to obtain a target generator and a target regularization module; wherein, a first adversarial domain image conversion module is used to convert the source domain image into the first adversarial domain image through the target generator; and a non-adversarial domain image conversion module is used to convert the first adversarial domain image into the non-adversarial domain image through the target regularization module.

[0019] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned image processing methods by executing the executable instructions.

[0020] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the image processing method described in any one of the above is implemented.

[0021] According to another aspect of the present disclosure, a computer program product is provided, including: a computer program or instructions, wherein when the computer program or instructions are executed by a processor, any one of the above-mentioned image processing methods is implemented.

[0022] The technical solution provided in the embodiment of the present disclosure can convert the source domain image to the first adversarial domain through the trained generator, thereby obtaining the adversarial example corresponding to the source domain image to be processed, thereby realizing the protection of the face image, and then the trained regularization module can ensure the visual effect while ensuring the image protection effect. Therefore, the embodiment of the present disclosure can ensure the reliability of image processing for face protection and improve the user experience.

[0023] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0025] Figure 1A schematic diagram showing a system structure in an embodiment of the present disclosure;

[0026] Figure 2 A flow chart of an image processing method in an embodiment of the present disclosure is shown;

[0027] Figure 3 A schematic diagram showing an overall architecture of image processing in an embodiment of the present disclosure is shown;

[0028] Figure 4 A flow chart of a method for model training in an embodiment of the present disclosure is shown;

[0029] Figure 5 A flow chart of a method for protecting the privacy of a facial image in an embodiment of the present disclosure is shown;

[0030] Figure 6 A schematic diagram of an image processing device in an embodiment of the present disclosure is shown;

[0031] Figure 7 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0033] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0034] The specific implementation of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0035] Figure 1 FIG. 2 shows an exemplary application system architecture diagram to which the image processing method in the embodiment of the present disclosure can be applied. Figure 1 As shown, the system architecture may include a terminal device 101 , a network 102 and a server 103 .

[0036] Among them, the terminal device 101 can obtain the source domain image to be processed and send the source domain image to the server 103. The source domain image is an image containing a face. Then, the server 103 can convert the source domain image into a first adversarial domain image through a trained generator. The first adversarial domain image is converted into a non-adversarial domain image through a trained regularization module. Finally, the terminal device 101 can obtain the non-adversarial domain image and cover the source domain image with the non-adversarial domain image to obtain a processed image.

[0037] The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103, and can be a wired network or a wireless network.

[0038] Alternatively, the terminal device 101 may obtain a source domain image to be processed, where the source domain image is an image containing a face. Then, the terminal device 101 may convert the source domain image into a first adversarial domain image through a trained generator. The first adversarial domain image is converted into a non-adversarial domain image through a trained regularization module. Finally, the terminal device 101 may obtain the non-adversarial domain image, and overlay the source domain image with the non-adversarial domain image to obtain a processed image.

[0039] Alternatively, the server 103 may obtain a source domain image to be processed, where the source domain image is an image containing a face. Then, the server 103 may convert the source domain image into a first adversarial domain image through a trained generator. The first adversarial domain image is converted into a non-adversarial domain image through a trained regularization module. Finally, the server 103 obtains the non-adversarial domain image, and overlays the source domain image with the non-adversarial domain image to obtain a processed image.

[0040] Optionally, the wireless network or wired network described above uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPSec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above data communication technologies.

[0041] The terminal device 101 can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, smart speakers, smart watches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0042] The server 103 may be a server that provides various services, such as a background management server that provides support for the device operated by the user using the terminal device 101. The background management server may analyze and process the received request and other data, and feed back the processing results to the terminal device.

[0043] Optionally, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud 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, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0044] Those skilled in the art will know that Figure 1The number of terminal devices, networks and servers in the embodiment is only for illustration, and any number of terminal devices, networks and servers may be provided according to actual needs, and the embodiments of the present disclosure do not limit this.

[0045] Under the above system architecture, an image processing method is provided in an embodiment of the present disclosure. The method can be executed by any electronic device with computing and processing capabilities.

[0046] In some embodiments, the image processing method provided in the embodiments of the present disclosure can be executed by a terminal device of the above-mentioned system architecture; in other embodiments, the image processing method provided in the embodiments of the present disclosure can be executed by a server in the above-mentioned system architecture; in other embodiments, the image processing method provided in the embodiments of the present disclosure can be implemented by the terminal device and the server in the above-mentioned system architecture through interaction.

[0047] Figure 2 A flow chart of an image processing method in an embodiment of the present disclosure is shown as follows: Figure 2 As shown, the image processing method provided in the embodiment of the present disclosure includes the following steps S202 to S208.

[0048] S202, obtaining a source domain image to be processed, where the source domain image is an image containing a face.

[0049] The embodiments of the present disclosure do not limit the source domain image to be processed, and the source domain image to be processed can be determined based on the application scenario. For example, in the field of short videos, the source domain image to be processed can be any frame image in a certain video, or in the field of instant messaging applications, the source domain image to be processed can be a selfie containing a face, etc.

[0050] In some exemplary embodiments, face detection may be performed on a video or a picture based on a terminal, and when a face is detected, a source domain image containing the face may be acquired.

[0051] In a possible implementation, the source domain image to be processed may be a complete image containing a human face, or the source domain image to be processed may be a facial image obtained by extracting a facial part from an image containing a human face.

[0052] S204, converting the source domain image into a first adversarial domain image through the trained generator.

[0053] In an exemplary embodiment, the trained generator is used to achieve conversion from a source domain to a first adversarial domain. The first adversarial domain can be based on the source domain and can protect the face image by generating an adversarial makeup.

[0054] It should be noted that adversarial makeup refers to changing facial features through specific makeup techniques, so that facial recognition systems cannot accurately recognize or classify faces. This technology uses the concept of adversarial samples to make machine learning models produce incorrect predictions by adding tiny perturbations to the input data.

[0055] Since the generator can generate adversarial makeup in various styles, the generator can be pre-trained so that the generator can output a first adversarial domain image similar to a reference domain, where the reference domain can be the adversarial makeup style that the generator is currently required to output. Therefore, by training the generator, it can be ensured that the generated first adversarial domain image can play an adversarial role, and it can also be ensured that the first adversarial domain maintains a target adversarial makeup style similar to the reference domain.

[0056] S206, converting the first adversarial domain image into a non-adversarial domain image through the trained regularization module.

[0057] In some exemplary embodiments, the generator may be represented as G and the regularization module may be represented as H. Thus, the first adversarial domain image output by the trained generator may be represented as G(x, y), and the trained regularization module may perform image conversion on the first adversarial domain image to obtain a non-adversarial domain image H(G(x, y)).

[0058] It should be noted that the non-adversarial domain image H(G(x, y)) has the same content, style and dimension as the first adversarial domain image G(x, y), but is not adversarial. Therefore, the trained regularization module can make the image more natural while ensuring that the image processing can achieve the adversarial function, thereby ensuring the visual effect while ensuring the image protection effect.

[0059] S208, overlaying the source domain image with the non-adversarial domain image to obtain a processed image.

[0060] In an exemplary embodiment, the non-adversarial domain image may be an image of a human face, so the human face in the source domain image may be covered. Alternatively, the non-adversarial domain image may be a complete image containing a human face, in which case the source domain image may be directly replaced based on the non-adversarial domain image.

[0061] Exemplarily, it is possible to determine whether it is currently in a protection state. When it is in a protection state, the corresponding protection pattern can be obtained according to the source domain image of the face part according to the image processing method provided by the present disclosure to cover it, thereby realizing information protection, and misleading the normal recognition model output by the image based on the adversarial makeup. When it is not in a protection state, the makeup protection generation is turned off, the key area of ​​the face image is completely unobstructed, and normal face recognition interactive verification can be completed. Therefore, the embodiment of the present disclosure can achieve a good balance between face privacy protection and subjective recognition. The privacy image obtained by the method provided by the embodiment of the present disclosure can remove some biometric features of the source domain image to be processed, shielding the possibility of the terminal machine obtaining face information, and the regularization module can be used to prevent the processed image from being seriously distorted, thereby retaining good subjective recognition of the human eye.

[0062] In an exemplary embodiment, after obtaining the processed image, it can be determined whether the current user has authorized it. If the user has authorized it, face-related operations can be performed based on the processed image. If the user has not authorized it, the data can be destroyed. The face-related operations can include, for example, face recognition, face interaction traces, etc.

[0063] It should be noted that in order to make the generator and regularization module have better performance, the generator and regularization module can be trained.

[0064] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: training the generator based on the first discriminator and the second discriminator to obtain a trained generator; wherein the first discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the source domain sample, and the second discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the reference domain sample, and the reference domain is a preset image processing target domain.

[0065] In a possible implementation, the input of the first discriminator can be a color image with a resolution of 112×112 and three color channels of RGB (Red, Green, Blue). After the color image is input, it can be sequentially output through a 32-channel output 5×5 convolution layer with a stride of 2, a LeakyReLU (Leaky Rectified Linear Unit) activation function, a 64-channel output 5×5 convolution layer with a stride of 2, a LeakyReLU activation function, a BN (Batch Normalization) layer, a 128-channel output 5×5 convolution layer with a stride of 2, a LeakyReLU activation function, a BN layer, a 64-channel output 5×5 convolution layer with a stride of 2, a Leaky ReLU activation function, a BN layer, a 1-unit fully connected layer, and a sigmoid activation function, and output a binary classification result of whether the color image is a true source domain image. Therefore, the first discriminator can be used to distinguish the difference between the first adversarial domain sample generated by the generator during the training process and the true source domain image.

[0066] Exemplarily, the structure of the second discriminator can be the same as that of the first discriminator, and the second discriminator can be used to input a color image and output a binary classification result of whether the color image is a reference domain image. Therefore, the second discriminator can be used to distinguish the difference between the first adversarial domain sample generated by the generator during the training process and the reference domain image.

[0067] In a possible implementation, the input of the generator can be a 100-dimensional Gaussian white noise sequence. After the Gaussian white noise sequence is input, the 14592-dimensional feature can be converted into a feature map with a resolution of 6×19 and a number of channels of 128 through a fully connected layer of 14592 units and a BN layer. The feature map then passes through a 5×5 deconvolution layer with a 128-channel output with a stride of 2, a BN layer, a 5×5 deconvolution layer with a 64-channel output with a stride of 2, a BN layer, a 5×5 deconvolution layer with a 32-channel output with a stride of 2, a 5×5 deconvolution layer with a 3-channel output with a stride of 2 in the BN layer, and a tanh activation function, and finally outputs the first adversarial domain sample.

[0068] In an exemplary embodiment, the generator, the first discriminator, and the second discriminator can form a GAN (Generative Adversarial Networks). Therefore, by training the generator through the first discriminator and the second discriminator, the adversarial nature of the first adversarial domain image output by the trained generator can be ensured, and the similarity of the adversarial makeup style between the first adversarial domain and the reference domain can be improved.

[0069] In some exemplary embodiments, a generator is trained based on a first discriminator and a second discriminator to obtain a trained generator, including: iteratively executing the following steps until the generator meets a first preset end condition to obtain the trained generator: freezing the parameters of the generator, updating the parameters of the first discriminator and the second discriminator so that the discriminator loss functions corresponding to the first discriminator and the second discriminator meet the maximization condition; freezing the parameters of the first discriminator and the second discriminator, and updating the parameters of the generator so that the generator loss function corresponding to the generator meets the minimization condition.

[0070] In a possible implementation, the original sample set and the corresponding labels can be input into a GAN composed of a generator, a first discriminator, and a second discriminator. The original sample set may include source domain samples, first adversarial domain samples, reference domain samples, Gaussian white noise sequence samples, and corresponding labels. Any sample corresponds to a label, and the label is used to indicate that the corresponding sample belongs to a source domain sample, a first adversarial domain sample, or a reference domain sample.

[0071] Exemplarily, after the original sample set and the corresponding labels are input into GAN, the parameter initialization method can be used to initialize the GAN network parameters, and the corresponding model optimization algorithm can be set. The Adam parameter optimization algorithm is used for parameter optimization, and the learning rate, the first-order smoothing parameter β1, the second-order smoothing parameter β2 and the constant e that avoids the denominator being 0 during the calculation process are set respectively.

[0072] Then, the parameters of the generator can be frozen, and the parameters of the first discriminator and the second discriminator can be unfrozen. At this time, a certain number of source domain samples and the same number of Gaussian white noise sequence samples can be obtained, and the generator can be used to output generated samples based on the source domain samples and Gaussian white noise sequence samples. Then, the source domain samples and the generated samples are sent to the two discriminators for training, and the parameters of the first discriminator are adjusted with the goal of maximizing the loss function of the first discriminator.

[0073] Then, the parameters of the first and second discriminators can be frozen, and the parameters of the generator can be unfrozen. A certain number of source domain samples and the same number of Gaussian white noise sequence samples can be obtained. The generator can be used to output generated samples based on the source domain samples and Gaussian white noise sequence samples. The source domain samples and generated samples are sent to the two discriminators for training, and the generator parameters are adjusted with the goal of minimizing the generator loss function L. Repeat the above operation until the network parameters are stable, and the trained generator can be obtained.

[0074] In an exemplary embodiment, the adversarial loss function of the generator may be as shown in formula (1).

[0075]

[0076] In formula (1), It can represent the adversarial loss function of the generator. G(y,x) is used to represent the sample output by the autoencoder based on the first adversarial domain sample for source domain conversion. G(x,y) can represent the generated sample output by the generator based on the source domain sample for the first adversarial domain conversion. D X (G(y,x)) represents the discrimination result of the first discriminator on the sample generated by the self-encoder. Y (G(x,y)) represents the discrimination result of the second discriminator on the generated sample output by the generator.

[0077] In an exemplary embodiment, the adversarial loss function of the discriminator may be as shown in formula (2).

[0078]

[0079] In formula (2), It can represent the adversarial loss function of the discriminator. G(y,x) is used to represent the sample output by the autoencoder based on the first adversarial domain sample for source domain conversion. G(x,y) can represent the generated sample output by the generator based on the source domain sample for the first adversarial domain conversion. D X (G(y,x)) represents the discrimination result of the first discriminator on the sample generated by the self-encoder. Y (G(x,y)) represents the discrimination result of the second discriminator on the generated sample output by the generator. X (x) can represent the discrimination result of the first discriminator on the source domain sample. Y (y) may represent the discrimination result of the second discriminator on the generated sample output by the generator.

[0080] In some exemplary embodiments, GAN can also be iteratively trained based on cycle consistency loss. It should be noted that after the source domain sample x is sent into G, a G(x, y) sample is obtained. Then G(x, y) can be converted to the source domain through the autoencoder to obtain the reconstructed G'(G(x, y)). Ideally, G'(G(x, y)) should be almost the same as the original source domain sample x, thus constituting a cycle consistency loss. Since the generator performs adversarial perturbations on the samples in the input source domain image x, the autoencoder cannot extract the features of G(x, y), and therefore cannot convert G(x, y) from the first adversarial domain back to the source domain. Therefore, the obtained G'(G(x, y)) may be significantly different from the source domain sample x, that is, the cycle consistency loss fails.

[0081] For example, the loss function of the cycle consistency loss of GAN can be shown as formula (3).

[0082] L cycle=||G(G(x,y),x)-x||1 (3)

[0083] In formula (3), L cycle may represent the cycle consistency loss. G(G(x,y),x) may represent the sample generated by the autoencoder according to the first adversarial domain sample generated by the generator based on the source domain sample. ||·||1 may represent the L1 norm.

[0084] It should be noted that by setting the first discriminator and the second discriminator, the generator can be iteratively trained based on the generator adversarial loss, discriminator adversarial loss, and cycle consistency loss respectively shown in the above formulas (1) to (3), thereby obtaining a trained generator. Therefore, the trained generator can ensure the quality and effectiveness of the output first adversarial domain image.

[0085] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: training a regularization module through the first discriminator, the second discriminator, and the trained generator to obtain a trained regularization module.

[0086] In an exemplary embodiment, the generator can be used to perform image conversion from a source domain to a first adversarial domain, and the regularization module can perform image conversion from the first adversarial domain to a non-adversarial domain. Therefore, the regularization module can be trained based on the process of training the generator.

[0087] In some possible implementations, the combination of the generator and the regularization module can achieve the conversion of the source domain image x to the first adversarial domain image G(x, y), and then convert G(x, y) to the non-adversarial domain H(G(x, y)). Therefore, the regularization module can be trained in a similar process to the above generator, that is, based on adversarial loss and cycle consistency loss.

[0088] In some exemplary embodiments, a regularization module is trained by a first discriminator, a second discriminator, and a trained generator to obtain a trained regularization module, including: a cycle-consistent model is obtained according to a combination of the trained generator, the regularization module, the first autoencoder, and the second autoencoder, wherein the first autoencoder is used for image conversion from a non-adversarial domain to a second adversarial domain, and the second autoencoder is used for image conversion from the second adversarial domain to a source domain; the regularization module is iteratively trained by calculating at least one of a cycle-consistent loss and a visual consistency loss of the cycle-consistent model until a second preset end condition is met to obtain the trained regularization module.

[0089] In a possible implementation, the image conversion from the source domain to the first adversarial domain can be realized by the generator, and then the image conversion from the first adversarial domain to the non-adversarial domain can be realized by the regularization module, and then the image conversion from the non-adversarial domain to the second adversarial domain can be realized by the first autoencoder, and finally the image conversion from the second adversarial domain to the source domain can be realized by the second autoencoder. Therefore, the present disclosure can realize the cycle from source domain X→adversarial domain Ya→non-adversarial domain Y→second adversarial domain Xa→source domain X, thereby ensuring that during the cyclic reconstruction training process, the adversarial loss will not attack the generator and cause the cycle consistency loss to fail.

[0090] Exemplarily, the cycle-consistent loss function of the regularization module can be shown as follows:

[0091]

[0092] In formula (4), Used to represent cycle-consistent loss. G(x,y) can represent the sample output by the generator based on the source domain sample to perform the conversion to the first adversarial domain. H(G(x,y)) can represent the sample output by the regularization module to perform the conversion from the first adversarial domain to the non-adversarial domain. G(H(G(x,y)),x) can represent the sample output by the first autoencoder to perform the conversion from the non-adversarial domain to the second adversarial domain. H)G(H(G(x,y)),x)) can represent the sample output by the second autoencoder to perform the conversion from the second adversarial domain to the source domain. Similarly, H(G(H(G(y,x)),y)) can represent the inverse process of the above process, thereby realizing the cycle.

[0093] It should be noted that the regularization module should reduce the adversarial effect of the first adversarial domain image output by the generator and keep the non-adversarial domain image and the first adversarial domain image visually unchanged. Therefore, the regularization module can be made to accept the reward penalty from the first discriminator and the second discriminator, so as to further train the regularization module, thereby enhancing the real performance of the regularization module reconstruction generator in the style transfer of adversarial makeup.

[0094] Therefore, training can be performed based on the visual consistency loss function of the regularization module as shown in the following formula (5).

[0095]

[0096] In formula (5), It can represent the visual consistency loss of the regularization module. x (H(G(y,x))) is used to represent the discrimination result of the first discriminator on the samples generated by the regularization module. Y(H(G(x,y))) is used to represent the discrimination result of the second discriminator on the samples generated by the regularization module.

[0097] In some exemplary embodiments, the image processing method provided by the embodiments of the present disclosure further includes: forming a mobility enhancement module based on multiple trained face recognition models; obtaining a diversity sample set through the mobility enhancement module; and jointly training the trained generator and the trained regularization module based on the diversity sample set to obtain a target generator and a target regularization module.

[0098] In this case, the image processing method provided by the embodiment of the present disclosure includes: converting the source domain image into a first adversarial domain image through a target generator; and converting the first adversarial domain image into a non-adversarial domain image through a target regularization module.

[0099] It should be noted that the transferability enhancement module can be composed of K pre-trained face recognition models with higher accuracy, where K is a positive integer. These face recognition models have high accuracy on public face image datasets. When the decision boundaries of inaccessible potential target models are imitated during generator training, these face recognition models can act as white-box models, and an integrated training strategy with input diversity enhancement can be used to generate adversarial examples with high transferability and black-box attack success rate. The disclosed embodiment does not limit the integrated training strategy, which can be set based on experience or application scenarios.

[0100] In an exemplary embodiment, the generator can be jointly trained with a regularization module and input diversity processing can be applied to maximize the decision boundary, thereby encouraging the generator to generate adversarial examples with high transferability and black-box attack success rate through an integrated training strategy with input diversity enhancement.

[0101] In an exemplary embodiment, the adversarial attack loss of the generator can be expressed as follows: Formula (6) The adversarial attack loss of the regularization module can be expressed as follows: Formula (7)

[0102]

[0103] in, It can represent the adversarial attack loss of the generator in joint training. M can represent the adversarial attack loss of the regularization module in joint training. krepresents the feature extractor of the k-th local face recognition model, using cosine similarity as the distance function. T may represent a transformation function, and p may represent a predefined probability of whether a transformation will be performed on G(x, y). Exemplarily, image resizing and Gaussian denoising may be selected as the transformation function. z may represent a target image, for example, the target image may be any real facial image.

[0104] The method provided by the embodiment of the present disclosure can convert the source domain image to the first adversarial domain through the trained generator, thereby obtaining the adversarial example corresponding to the source domain image to be processed, thereby protecting the face image, and then the trained regularization module can ensure the visual effect while ensuring the image protection effect. Therefore, the embodiment of the present disclosure can ensure the reliability of image processing for face protection and improve the user experience.

[0105] For example, an overall architecture of image processing provided by the embodiment of the present disclosure can be as follows: Figure 3 As shown. Among them, the first adversarial domain sample can be constructed based on the generator, and then the corresponding non-adversarial domain sample can be output based on the regularization module and the transfer enhancement module. Finally, the generative adversarial network and the regularization module can be jointly trained based on the joint training module.

[0106] For example, a flowchart of a model training method provided by an embodiment of the present disclosure can be as follows: Figure 4 As shown. Figure 4 As shown, the model training method provided in the embodiment of the present disclosure includes the following steps S402 to S406.

[0107] S402, training the generator.

[0108] Exemplarily, a generator may be used to generate a first adversarial domain sample corresponding to an input source domain image, and then a first discriminator and a second discriminator may be trained to distinguish between true and false images generated by the generator.

[0109] S404, training regularization module.

[0110] Exemplarily, the regularization module can be trained based on the cycle-consistent loss, so as to achieve the normal operation of the cycle-consistent loss and ensure the visual effect of the image output by the regularization module.

[0111] S406, performing joint training based on the mobility enhancement module.

[0112] Exemplarily, the mobility enhancement module may be composed of K pre-trained face recognition models with higher accuracy. Based on multiple pre-trained face recognition models, the decision boundary may be maximized, and model joint training and input diversity processing may be achieved.

[0113] It should be noted that the initial training sample set can be a dataset of tens of thousands of makeup styles that are the same as the reference. Through training, the generator and regularization module can generate different makeup styles that fit the skin, texture and other elements of the face, ensuring the image protection effect while ensuring the visual effect, protecting the face area information while sufficiently shielding information, and gaining higher user acceptance.

[0114] For example, a flowchart of a facial image privacy protection method provided by an embodiment of the present disclosure can be as follows: Figure 5 As shown. Figure 5 As shown, the model training method provided in the embodiment of the present disclosure includes the following steps S502 to S514.

[0115] S502: The terminal detects a face image.

[0116] S504, converting the face image into a corresponding non-adversarial domain image through a generator and a regularization module.

[0117] S506, protecting the face privacy information through non-adversarial domain images.

[0118] For example, the privacy information of a person's face can be protected by overlaying a non-adversarial domain image on the face image.

[0119] S508, determining the facial image after information protection.

[0120] S510, determine whether the user authorizes. If yes, execute S512. If no, execute S514.

[0121] S512, performing face-related operations according to the face image.

[0122] S514, destroying the facial image.

[0123] The disclosed embodiment can flexibly switch the protection or normal exposure state of the facial image during actual application by determining whether the user has authorized it.

[0124] It should be noted that the acquisition, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations, and various types of data such as personal identity data, operation data, behavioral data, etc. related to individuals, customers, and groups obtained in the embodiments of this disclosure have all been authorized.

[0125] Based on the same inventive concept, an image processing device is also provided in the embodiment of the present disclosure, as described in the following embodiment. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0126] Figure 6 A schematic diagram of an image processing device in an embodiment of the present disclosure is shown. Figure 6 As shown, the device comprises:

[0127] A source domain image acquisition module 601 is used to acquire a source domain image to be processed, where the source domain image is an image containing a human face;

[0128] A first adversarial domain image conversion module 602, configured to convert a source domain image into a first adversarial domain image through a trained generator;

[0129] A non-adversarial domain image conversion module 603, configured to convert the first adversarial domain image into a non-adversarial domain image through a trained regularization module;

[0130] The image covering module 604 is used to cover the source domain image with the non-adversarial domain image to obtain a processed image.

[0131] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure further includes:

[0132] A generator training module is used to train the generator based on a first discriminator and a second discriminator to obtain a trained generator; wherein the first discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the source domain sample, and the second discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the reference domain sample, and the reference domain is a preset image processing target domain.

[0133] In some exemplary embodiments, the generator training module is used to iteratively execute the following steps until the generator meets the first preset end condition to obtain a trained generator: freeze the parameters of the generator, update the parameters of the first discriminator and the second discriminator so that the discriminator loss function corresponding to the first discriminator and the second discriminator meets the maximization condition; freeze the parameters of the first discriminator and the second discriminator, and update the parameters of the generator so that the generator loss function corresponding to the generator meets the minimization condition.

[0134] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure further includes:

[0135] The regularization module training module is used to train the regularization module through the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module.

[0136] In some exemplary embodiments, a regularization module training module is used to obtain a cycle-consistent model based on a combination of a trained generator, a regularization module, a first autoencoder, and a second autoencoder, wherein the first autoencoder is used to perform image conversion from a non-adversarial domain to a second adversarial domain, and the second autoencoder is used to perform image conversion from the second adversarial domain to a source domain; by calculating at least one of the cycle-consistent loss and the visual consistency loss of the cycle-consistent model, the regularization module is iteratively trained until a second preset end condition is met to obtain a trained regularization module.

[0137] In some exemplary embodiments, the image processing device provided by the embodiments of the present disclosure further includes:

[0138] A joint training module is used to form a transfer enhancement module based on multiple trained face recognition models; obtain a diverse sample set through the transfer enhancement module; and jointly train the trained generator and the trained regularization module based on the diverse sample set to obtain a target generator and a target regularization module;

[0139] The first adversarial domain image conversion module 602 is used to convert the source domain image into the first adversarial domain image through the target generator;

[0140] The non-adversarial domain image conversion module 603 is used to convert the first adversarial domain image into a non-adversarial domain image through a target regularization module.

[0141] The device provided by the embodiment of the present disclosure can convert the source domain image to the first adversarial domain through a trained generator, thereby obtaining an adversarial example corresponding to the source domain image to be processed, thereby protecting the facial image, and then the trained regularization module can ensure the visual effect while ensuring the image protection effect. Therefore, the embodiment of the present disclosure can ensure the reliability of image processing for face protection and improve user experience.

[0142] It should be noted that the examples and application scenarios implemented by the modules in the above-mentioned device embodiment are the same as those of the corresponding steps in the method embodiment, but are not limited to the contents disclosed in the above-mentioned method embodiment. It should be noted that the above-mentioned modules as part of the device can be executed in a computer system such as a set of computer executable instructions.

[0143] Those skilled in the art will appreciate that various aspects of the present disclosure may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation that combines hardware and software aspects, which may be collectively referred to herein as a "circuit," "module," or "system."

[0144] Based on the same inventive concept, an electronic device is also provided in an embodiment of the present disclosure, the electronic device comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned image processing methods by executing the executable instructions. Since the principle of solving the problem in the electronic device embodiment is similar to that in the above-mentioned method embodiment, the implementation of the electronic device embodiment can refer to the implementation of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0145] Refer to the following Figure 7 An electronic device 700 according to this embodiment of the present disclosure is described. Figure 7 The electronic device 700 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0146] like Figure 7 As shown, the electronic device 700 is in the form of a general computing device. The components of the electronic device 700 may include but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including the storage unit 720 and the processing unit 710).

[0147] The storage unit stores a program code, which can be executed by the processing unit 710, so that the processing unit 710 performs the steps of various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 710 can perform the following steps of the above method embodiment:

[0148] A source domain image to be processed is obtained, where the source domain image is an image containing a face; the source domain image is converted into a first adversarial domain image through a trained generator; the first adversarial domain image is converted into a non-adversarial domain image through a trained regularization module; the source domain image is covered with the non-adversarial domain image to obtain a processed image.

[0149] The storage unit 720 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 7201 and / or a cache storage unit 7202 , and may further include a read-only storage unit (ROM) 7203 .

[0150] The storage unit 720 may also include a program / utility 7204 having a set (at least one) of program modules 7205, such program modules 7205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0151] Bus 730 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0152] The electronic device 700 may also communicate with one or more external devices 740 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 700, and / or communicate with any device that enables the electronic device 700 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 750. Furthermore, the electronic device 700 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 760. As shown, the network adapter 760 communicates with other modules of the electronic device 700 via a bus 730. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0153] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0154] Based on the same inventive concept, the embodiment of the present disclosure also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, any of the above-mentioned image processing methods is implemented. Since the principle of solving the problem in the embodiment of the computer-readable storage medium is similar to that in the above-mentioned method embodiment, the implementation of the embodiment of the computer-readable storage medium can refer to the implementation of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0155] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a 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.

[0157] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0158] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0159] Based on the same inventive concept, a computer program product is also provided in an embodiment of the present disclosure, including a computer program product, including: a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, the image processing method of any one of the above method embodiments is implemented. Since the principle of solving the problem in the computer program product embodiment is similar to that in the above method embodiment, the implementation of the computer program product embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0160] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, 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 multiple modules or units to be embodied.

[0161] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0162] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0163] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be regarded as exemplary only, and the true scope of the present disclosure is indicated by the appended claims.

Claims

1. An image processing method, characterized in that: include: Acquire a source domain image to be processed, wherein the source domain image is an image containing a human face; Converting the source domain image into a first adversarial domain image through the trained generator; Converting the first adversarial domain image into a non-adversarial domain image through the trained regularization module; The source domain image is overlaid with the non-adversarial domain image to obtain a processed image.

2. The image processing method according to claim 1, characterized in that: The method further comprises: Training the generator based on the first discriminator and the second discriminator to obtain the trained generator; The first discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the source domain sample, and the second discriminator is used to describe the difference between the first adversarial domain sample generated by the generator and the reference domain sample, and the reference domain is a preset image processing target domain.

3. The image processing method according to claim 2, characterized in that: The step of training the generator based on the first discriminator and the second discriminator to obtain the trained generator includes: Iteratively perform the following steps until the generator meets the first preset end condition to obtain the trained generator: Freezing the parameters of the generator, and updating the parameters of the first discriminator and the second discriminator so that the discriminator loss functions corresponding to the first discriminator and the second discriminator meet the maximization condition; The parameters of the first discriminator and the second discriminator are frozen, and the parameters of the generator are updated so that the generator loss function corresponding to the generator satisfies a minimization condition.

4. The image processing method according to claim 2 or 3, characterized in that: The method further comprises: The regularization module is trained by using the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module.

5. The image processing method according to claim 4, characterized in that: The regularization module is trained by the first discriminator, the second discriminator, and the trained generator to obtain the trained regularization module, comprising: A cycle-consistent model is obtained according to the trained generator, the regularization module, the first autoencoder, and the second autoencoder, wherein the first autoencoder is used to perform image conversion from the non-adversarial domain to the second adversarial domain, and the second autoencoder is used to perform image conversion from the second adversarial domain to the source domain; By calculating at least one of the cycle consistency loss and the visual consistency loss of the cycle consistency model, the regularization module is iteratively trained until a second preset end condition is met to obtain the trained regularization module.

6. The image processing method according to claim 1, characterized in that: The method further comprises: A transfer enhancement module is formed based on multiple trained face recognition models; Obtaining a diverse sample set through the mobility enhancement module; Jointly training the trained generator and the trained regularization module according to the diversity sample set to obtain a target generator and a target regularization module; Wherein, the method comprises: Converting the source domain image into the first adversarial domain image through the target generator; The first adversarial domain image is converted into the non-adversarial domain image through the target regularization module.

7. An image processing device, characterized in that: include: A source domain image acquisition module, used to acquire a source domain image to be processed, wherein the source domain image is an image containing a human face; A first adversarial domain image conversion module, configured to convert the source domain image into a first adversarial domain image through a trained generator; a non-adversarial domain image conversion module, configured to convert the first adversarial domain image into a non-adversarial domain image through a trained regularization module; The image covering module is used to cover the source domain image with the non-adversarial domain image to obtain a processed image.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the image processing method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage 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 6 is implemented.

10. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs the image processing method as described in any one of claims 1 to 6.

Images