Data Processing Method, Apparatus and Equipment

Abstract

An embodiment of this specification provides a data processing method, apparatus, and device. The method includes: receiving a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtaining a first image corresponding to the target user; determining a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service; sending the target desensitized image to a server so that the server executes the target service based on the target desensitized image.

Description

Technical Field

[0001] This document relates to the field of data processing technologies, and in particular, to a data processing method, apparatus, and device. Background Art

[0002] With the rapid development of computer technologies, it is possible to verify the true identity of a user through private data such as face images, fingerprints, and irises, and provide corresponding services to the user after successful verification.

[0003] To protect the security of the user's private data, after obtaining the user's private data, the terminal device can encrypt the user's private data with a key and send the encrypted private data to the server. The server decrypts the data with the key and authenticates the user based on the decrypted private data.

[0004] However, the encrypted private data may be stolen by a malicious third party during data transmission, and the security of encrypting the user's private data with a key is poor, resulting in the user's private data being easily stolen maliciously. Therefore, a solution to improve the security of the user's private data is needed. Summary of the Invention

[0005] The objective of the embodiments of this specification is to provide a data processing method, apparatus, and device to provide a solution that can improve the security of the user's private data.

[0006] To implement the above technical solution, the embodiments of this specification are implemented as follows:

[0007] In a first aspect, the embodiments of this specification provide a data processing method, including: receiving a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtaining a first image corresponding to the target user, where the first image includes biometric data of the target user; determining a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service; sending the target desensitized image to the server so that the server executes the target service based on the target desensitized image.

[0008] Second aspect, an embodiment of this specification provides a data processing method, including: obtaining a training data set for a target service, where the training data set includes multiple first images, and the first images include biometric data of a user; training an image desensitization model constructed by a preset deep learning algorithm based on the first images, a first loss function, and a second loss function to obtain a trained image desensitization model, where the first loss function is used to make the desensitized images output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized images output by the image desensitization model meet a preset image usage requirement of the target service; sending the trained image desensitization model to a client so that the client processes a target image including biometric data of a target user based on the trained image desensitization model.

[0009] Third aspect, an embodiment of this specification provides a data processing apparatus, where the apparatus includes: an instruction receiving module, configured to receive a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image includes biometric data of the target user; an image determination module, configured to determine a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function is used to make the desensitized images output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized images output by the image desensitization model meet a preset image usage requirement of the target service; an image sending module, configured to send the target desensitized image to a server so that the server executes the target service based on the target desensitized image.

[0010] Fourth aspect, an embodiment of this specification provides a data processing apparatus, including: a data acquisition module, configured to obtain a training data set for a target service, where the training data set includes multiple first images, and the first images include biometric data of a user; a model training module, configured to train an image desensitization model constructed by a preset deep learning algorithm based on the first images, a first loss function, and a second loss function to obtain a trained image desensitization model, where the first loss function is used to make the desensitized images output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized images output by the image desensitization model meet a preset image usage requirement of the target service; a data sending module, configured to send the trained image desensitization model to a client so that the client processes a target image including biometric data of a target user based on the trained image desensitization model.

[0011] Fifth aspect, an embodiment of this specification provides a data processing device, the data processing device includes: a processor; and a memory arranged to store computer-executable instructions, the executable instructions, when executed, cause the processor to: receive a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, the first image including biometric data of the target user; based on a pre-trained image desensitization model, determine a target desensitized image corresponding to the first image, the image desensitization model being obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function being used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function being used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service; send the target desensitized image to a server so that the server executes the target service based on the target desensitized image.

[0012] Sixth aspect, an embodiment of this specification provides a data processing device, the data processing device includes: a processor; and a memory arranged to store computer-executable instructions, the executable instructions, when executed, cause the processor to: obtain a training data set for a target service, the training data set including a plurality of first images, the first images including biometric data of users; based on the first images, a first loss function, and a second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model, the first loss function being used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function being used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service; send the trained image desensitization model to a client so that the client processes a target image including biometric data of a target user based on the trained image desensitization model.

[0013] In a seventh aspect, an embodiment of the present specification provides a storage medium for storing computer-executable instructions, and when the executable instructions are executed, the following process is implemented: receiving a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtaining a first image corresponding to the target user, where the first image includes biometric data of the target user; determining a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function is used to make the desensitized image output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized image output by the image desensitization model meet a preset image usage requirement of the target service; sending the target desensitized image to a server so that the server executes the target service based on the target desensitized image.

[0014] In an eighth aspect, an embodiment of the present specification provides a storage medium for storing computer-executable instructions, and when the executable instructions are executed by a processor, the following process is implemented: obtaining a training data set for a target service, where the training data set includes a plurality of first images, and the first images include biometric data of users; training an image desensitization model constructed by a preset deep learning algorithm based on the first images, a first loss function, and a second loss function to obtain a trained image desensitization model, where the first loss function is used to make the desensitized image output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized image output by the image desensitization model meet a preset image usage requirement of the target service; sending the trained image desensitization model to a client so that the client processes a target image including biometric data of a target user based on the trained image desensitization model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1A It is a flowchart of an embodiment of a data processing method in the present specification;

[0017] Figure 1B It is a schematic diagram of the processing process of an embodiment of a data processing method in the present specification;

[0018] Figure 2 Schematic diagram of the processing procedure for another embodiment of the data processing method in this specification;

[0019] Figure 3 Schematic diagram of a data processing in this specification;

[0020] Figure 4 Schematic diagram of another data processing in this specification;

[0021] Figure 5A Flowchart of an embodiment of the data processing method in this specification;

[0022] Figure 5B Schematic diagram of the processing procedure for another embodiment of the data processing method in this specification;

[0023] Figure 6 Schematic diagram of the processing procedure for another embodiment of the data processing method in this specification;

[0024] Figure 7 Schematic diagram of the processing procedure for another embodiment of the data processing method in this specification;

[0025] Figure 8 Schematic diagram of a data processing in this specification;

[0026] Figure 9 Schematic diagram of the structure of an embodiment of a data processing device in this specification;

[0027] Figure 10 Schematic diagram of the structure of another embodiment of a data processing device in this specification;

[0028] Figure 11 Schematic diagram of the structure of a data processing device in this specification. Detailed implementation manners

[0029] The embodiments of this specification provide a data processing method, device and equipment.

[0030] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this specification without creative efforts shall fall within the scope of protection of this specification.

[0031] Embodiment 1

[0032] As Figure 1A and Figure 1BAs shown in the figure, an embodiment of this specification provides a data processing method. The execution subject of this method can be a terminal device, which can be a mobile terminal device such as a mobile phone or a tablet computer, or a terminal device such as a personal computer, or a smart wearable device such as a smart watch. The method can specifically include the following steps:

[0033] In S102, receive a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtain a first image corresponding to the target user.

[0034] Among them, the target service can be any service that needs to authenticate the target user based on the biometric data of the target user. For example, the target service can be a resource transfer service, a privacy information change service, etc. The first image can contain the biometric data of the target user. For example, the first image can contain the face image, fingerprint information, or iris information of the target user.

[0035] In practice, with the rapid development of computer technology, the real identity of a user can be verified through privacy data such as face images, fingerprints, and irises, and corresponding services can be provided to the user after the verification passes. To protect the security of the user's privacy data, after the terminal device obtains the user's privacy data, it can encrypt the user's privacy data with a key and send the encrypted privacy data to the server. The server decrypts it with the key and authenticates the user based on the decrypted privacy data. However, the encrypted privacy data may be stolen by a malicious third party during the data transmission process, and the security of encrypting the user's privacy data with a key is poor, resulting in the user's privacy data being easily stolen by a malicious party. Therefore, a solution to improve the security of the user's privacy data is needed. For this reason, an embodiment of this specification provides a technical solution that can solve the above problems. For specific details, please refer to the following content.

[0036] Taking the resource transfer service as an example of the target service, the target user can trigger the start of a resource management application program in the terminal device and trigger the execution of the resource transfer service in the resource management application program. When the terminal device detects a trigger execution instruction for the resource transfer service from the target user, it can start the camera and obtain a first image including the face image of the target user through the camera. Alternatively, when the terminal device detects a trigger execution instruction for the resource transfer service from the target user, it can also generate a first image based on the fingerprint information of the target user.

[0037] The above method for obtaining the first image is an optional and implementable method. In actual application scenarios, there can be various different acquisition methods, which can vary according to different actual application scenarios. This specification does not make specific limitations on this.

[0038] In S104, based on a pre-trained image desensitization model, a target desensitized image corresponding to the first image is determined.

[0039] Among them, the image desensitization model can be obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service.

[0040] In implementation, the terminal device can receive the pre-trained image desensitization model sent by the server. That is, the server can train a model constructed by a preset deep learning algorithm based on the first loss function, the second loss function, and historical first images to obtain a trained image desensitization model.

[0041] Alternatively, the terminal device can also train a model constructed by a preset deep learning algorithm based on the first loss function, the second loss function, and historical first images to obtain a trained image desensitization model.

[0042] For example, the terminal device can send the service identifier of the target service to the server and receive the historical first images determined by the server based on the service identifier of the target service. Among them, the historical first images can be images containing user biometric data corresponding to the target service obtained by the server within a preset model training period. The terminal device can train a model constructed by a neural network algorithm based on the obtained historical first images, the first loss function, and the second loss function to obtain a trained image desensitization model.

[0043] Among them, the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements. For example, the first loss function can be used to determine whether the desensitized image (i.e., the historical first desensitized image) output by the image desensitization model meets the preset image desensitization requirements according to the image similarity between the historical first image and the historical first desensitized image (i.e., the desensitized image corresponding to the historical first image determined by the image desensitization model based on the historical first image). Specifically, if the image similarity between the historical first image and the historical first desensitized image is less than the preset similarity threshold, it can be determined that the desensitized image output by the image desensitization model meets the preset image desensitization requirements.

[0044] The second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service. For example, the second loss function can be used to determine whether the desensitized image output by the image desensitization model meets the preset image usage requirements of the target service according to the recognizability rate of the historical first desensitized image. Specifically, if the recognizability rate of the historical first desensitized image is not less than the preset recognizability rate threshold, it can be determined that the desensitized image output by the image desensitization model meets the preset image usage requirements of the target service. Among them, there are various methods for determining the recognizability rate of the historical first desensitized image, which may vary according to different actual application scenarios, and this specification does not make specific limitations on this.

[0045] The terminal device can input the first image into a pre-trained image desensitization model to obtain a target desensitized image corresponding to the first image. In this way, the obtained target desensitized image meets both the preset image desensitization requirements and the preset image usage requirements of the target service.

[0046] In addition, for different target services, different first loss functions, second loss functions, and deep learning algorithms used to construct the image desensitization model can be set to meet the image usage requirements and image desensitization requirements of different target services.

[0047] For example, by setting different similarity thresholds, the image desensitization model can meet different image desensitization requirements, and by setting different recognizability rate thresholds, the image desensitization model can meet different image usage requirements.

[0048] In S106, the target desensitized image is sent to the server so that the server can execute the target service based on the target desensitized image.

[0049] In implementation, the terminal device can send the target desensitized image and the service identifier of the target service to the server. The server can perform identity verification processing on the target user based on the target desensitized image and execute the target service after the identity verification is passed.

[0050] An embodiment of this specification provides a data processing method, which receives a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtains a first image corresponding to the target user. The first image includes biometric data of the target user. Based on a pre-trained image desensitization model, a target desensitized image corresponding to the first image is determined. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service. The target desensitized image is sent to the server so that the server can execute the target service based on the target desensitized image. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the malicious third party stealing the target desensitized image during the data transmission process. On the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the privacy data of the target user is improved.

[0051] Embodiment 2

[0052] As Figure 2 shown, an embodiment of this specification provides a data processing method. The execution subject of this method can be a terminal device, which can be a mobile terminal device such as a mobile phone or a tablet computer, or a terminal device such as a personal computer, or a smart wearable device such as a smart watch. The method can specifically include the following steps:

[0053] In S202, a trigger execution instruction of a target user for a target service is received, and in response to the trigger execution instruction, a first image corresponding to the target user is obtained.

[0054] Among them, the first image may include biometric data of the target user.

[0055] For the specific processing process of the above S202, reference can be made to the relevant content of S102 in the first embodiment above, which will not be elaborated here.

[0056] In S204, historical first images are obtained.

[0057] In implementation, the historical first image can be an image determined by the server based on the service identifier of the target service, or the historical first image can also be one or more images in a pre-stored training dataset in the terminal device, and the images in the training dataset can be images sent by the server to the terminal device within a preset model training period.

[0058] In S206, the historical first image is input into the image desensitization model to obtain the historical first desensitized image.

[0059] In implementation, in practical applications, the processing method of the above S206 can be various. The following provides an optional implementation method, which can specifically refer to the following processing steps 1 to 2:

[0060] Step 1, based on a preset image convolution algorithm, perform feature extraction processing on the historical first image to obtain a second image.

[0061] Step 2, based on a preset image recombination algorithm, perform image recombination processing on the second image to obtain the historical first desensitized image.

[0062] In implementation, as Figure 3 shown, taking the historical first desensitized image as an example containing the user's face image, the terminal device can extract the effective features related to face recognition in the historical first image based on the image convolution algorithm (Conv), that is, a second image can be obtained, and then through the image recombination algorithm (such as DeConv), perform inverse processing on the second image to obtain the historical first desensitized image.

[0063] In addition, in practical applications, the processing method of generating the second image in the above step 1 can be various. The following provides an optional implementation method, which can specifically refer to the following processing:

[0064] Based on a preset normalization algorithm and a preset signal convolution algorithm, perform compression processing on the historical first image to obtain a second image. For example, as Figure 4 shown, the historical first image can be compressed through the signal convolution algorithm (such as SignalConv2D) and the normalization algorithm (such as GDN) to obtain a second image. Among them, the signal convolution algorithm can ensure strict alignment of pixel positions, and GDN can ensure the quality of the second image with less introduced noise. Correspondingly, the pixel shuffle image recombination algorithm can be used to perform image recombination processing on the second image to obtain the historical first desensitized image. Specifically, for example, the second image of 28*28*48 can be restored to the historical first desensitized image of 112*112*3 through pixel shuffle.

[0065] In addition, by introducing uniform noise through the GDN to perturb the first historical image, the quantization loss during image storage can also be simulated, improving the accuracy of data processing.

[0066] In S208, based on the first historical image, the first desensitized historical image, the first loss function, and the second loss function, it is determined whether to retrain the image desensitization model to obtain a trained image desensitization model.

[0067] Among them, the first loss function can determine whether the image desensitization model meets the preset image desensitization requirements through the first historical image and the first desensitized historical image, and the second loss function can determine whether the image desensitization model meets the preset image usage requirements of the target service through the desensitized historical image.

[0068] In implementation, for example, the first loss function can determine whether the image desensitization model meets the preset image desensitization requirements through one or more of the first loss score, the second loss score, and the third loss score. The first loss score can be used to determine the desensitization effect between the first historical image and the first desensitized historical image at the pixel level, the second loss score can be used to determine the desensitization effect between the first historical image and the first desensitized historical image at the feature vector level, and the third loss score can be used to determine the desensitization effect between the first historical image and the first desensitized historical image at the recognition effect level. Among them, the desensitization effect can represent the distance between the first historical image and the first desensitized historical image.

[0069] In practical applications, the processing method of S208 above can be various. The following provides an optional implementation method, which can be specifically referred to the processing of the following steps 1 to 5:

[0070] Step 1, based on a preset image dimensionality reduction algorithm, obtain a first vector corresponding to the first historical image and a second vector corresponding to the first desensitized historical image, and determine a first loss score based on the first vector and the second vector.

[0071] Among them, the preset image dimensionality reduction algorithm can be any algorithm that can be used for image dimensionality reduction processing. For example, the image dimensionality reduction algorithm can be a Discrete Cosine Transform (DCT) algorithm, a Principal Component Analysis (PCA) algorithm, etc.

[0072] In implementation, based on a preset image dimensionality reduction algorithm, the historical first image and the historical first desensitized image can be processed respectively to obtain a first vector corresponding to the historical first image and a second vector corresponding to the historical first desensitized image. The mean square error of the first vector and the second vector can be used as the error value of the historical first desensitized image, and based on the error value of the historical first desensitized image, a first loss score can be determined.

[0073] For example, assuming there are 3 historical first images, corresponding to 3 historical first desensitized images, then based on the preset image dimensionality reduction algorithm, a first vector corresponding to each historical first image and a second vector corresponding to each historical first desensitized image can be obtained. The mean square error 1 between the historical first image 1 and the historical first desensitized image 1, the mean square error 2 between the historical first image 2 and the historical first desensitized image 2, and the mean square error 3 between the historical first image 3 and the historical first desensitized image 3 are obtained, and the mean of the mean square error 1, the mean square error 2, and the mean square error 3 is determined as the first loss score. In this way, the smaller the first loss score, the greater the difference between the historical first image and the historical first desensitized image at the pixel level, that is, the better the desensitization effect.

[0074] Step 2: Based on a pre-trained feature extraction model, obtain a third vector corresponding to the historical first image and a fourth vector corresponding to the historical first desensitized image, and based on the third vector and the fourth vector, determine a second loss score.

[0075] Among them, the feature extraction model can be obtained by training a model constructed based on a preset feature extraction algorithm with respect to the historical second image, and the historical second image can be an image containing user biometric data.

[0076] In implementation, assuming there are 3 historical first images, corresponding to 3 historical first desensitized images, the cosine similarity between each historical first image and the corresponding historical first desensitized image can be obtained based on the third vector and the fourth vector, and based on the determined cosine similarity, a second loss score can be determined. Specifically, for example, the mean of the cosine similarity 1 corresponding to the historical first image 1 and the historical first desensitized image 1, the cosine similarity 2 corresponding to the historical first image 2 and the historical first desensitized image 2, and the cosine similarity 3 corresponding to the historical first image 3 and the historical first desensitized image 3 can be determined as the second loss score. In this way, the smaller the second loss score, the greater the difference between the historical first image and the historical first desensitized image at the feature vector level, that is, the better the desensitization effect.

[0077] Step 3: Based on a pre-trained first recognition model, obtain a first recognition rate corresponding to the historical first image and a second recognition rate corresponding to the historical first desensitized image, and based on the first recognition rate and the second recognition rate, determine a third loss score.

[0078] Among them, the first recognition model can be obtained by training a model constructed with a preset image recognition algorithm based on historical second images.

[0079] In implementation, based on the pre-trained first recognition model, the recognition rates between the historical first image and each historical first image in the training dataset can be obtained, and the historical first images in the training dataset can be sorted based on the recognition rates. Based on the sorted historical first images and a preset selection quantity, the corresponding historical first images are selected, and based on the recognition rates corresponding to the selected historical first images, the first recognition rate is determined.

[0080] For example, assuming that the training dataset contains 10 historical first images, based on the pre-trained first recognition model, the recognition rates between historical first image 1 and the other 9 historical first images can be obtained, and these 9 historical first images can be sorted according to the recognition rates. The historical first images with the top 5 recognition rates can be selected, and the average value of the recognition rates of these 5 selected historical first images is determined as the first recognition rate corresponding to the historical first image.

[0081] Based on the above method for determining the first recognition rate, the second recognition rate corresponding to the historical first desensitized image can be determined. The ratio of the second recognition rate to the first recognition rate can be determined as the third loss score. In this way, the smaller the third loss score, the more difficult it is to recognize the image after desensitization, that is, the better the desensitization effect.

[0082] Step 4: Based on the pre-trained second recognition model, obtain the third recognition rate corresponding to the historical first desensitized image, and determine the fourth loss score based on the third recognition rate.

[0083] Among them, the second recognition model can be obtained by training a model constructed with a preset second recognition algorithm based on historical second desensitized images, and the historical second desensitized images are obtained by desensitizing historical second images.

[0084] In implementation, the algorithms used to construct the first recognition model, the second recognition model, and the feature extraction model can be the same or different machine learning algorithms. The machine learning algorithms used to construct the first recognition model, the second recognition model, and the feature extraction model can vary according to different actual application scenarios, and this specification does not make specific limitations on this.

[0085] Based on the pre-trained second recognition model, the third recognition rate corresponding to the historical first desensitized image can be obtained, and the third recognition rate is determined as the fourth loss score. In this way, the higher the fourth loss score, the better the image distinguishability of the image after desensitization, and the smaller loss of information can be ensured through the fourth loss score.

[0086] The determination methods of the above first loss score, second loss score, third loss score, and fourth loss score are optional and implementable determination methods. In actual application scenarios, there can be multiple different determination methods, which may vary according to different actual application scenarios. The embodiments of this specification do not make specific limitations on this.

[0087] Step Five, based on one or more of the first loss score, second loss score, and third loss score, and the fourth loss score, determine whether to retrain the image desensitization model.

[0088] In implementation, as Figure 4 shown, in actual application scenarios, a second image can also be generated through a signal convolution algorithm and a normalization algorithm. Correspondingly, the above Step Five can also be processed through the following A1 to A3:

[0089] In A1, perform decompression processing on the second image to obtain a third image.

[0090] In implementation, the second image can be decompressed through a signal convolution algorithm (such as SignalConv2D) and an inverse normalization algorithm (such as IGDN) to obtain a third image.

[0091] In A2, based on the third image and the first image, determine a fifth loss value.

[0092] Among them, the fifth loss value can be used to determine the degree of difference between the first image and the third image.

[0093] In implementation, the mean square error between the first image and the third image can be determined as the fifth loss value to ensure that the historical first desensitized image can contain the main information in the historical first image through the fifth loss value.

[0094] In A3, based on one or more of the first loss score, second loss score, and third loss score, and the fourth loss score and the fifth loss value, determine whether to retrain the image desensitization model.

[0095] In S210, based on the pre-trained image desensitization model, determine the target desensitized image corresponding to the first image.

[0096] In S212, send the target desensitized image to the server so that the server can execute the target service based on the target desensitized image.

[0097] For the specific processing procedures of the above S210 to S212, refer to the relevant content of S104 to S106 in the first embodiment above, and details are not described herein again.

[0098] An embodiment of this specification provides a data processing method, which receives a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtains a first image corresponding to the target user, where the first image includes biometric data of the target user. Based on a pre-trained image desensitization model, a target desensitized image corresponding to the first image is determined. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service. The target desensitized image is sent to the server so that the server can execute the target service based on the target desensitized image. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, the problem of information leakage caused by the target desensitized image being stolen by a malicious third party during data transmission can be avoided, and on the other hand, the server can execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the private data of the target user is improved.

[0099] Embodiment III

[0100] As Figure 5A and Figure 5B shown, an embodiment of this specification provides a data processing method. The execution subject of this method can be a server, which can be an independent server or a server cluster composed of multiple servers. This method can specifically include the following steps:

[0101] In S502, a training data set for the target service is obtained.

[0102] Among them, the training data set can include multiple first images, and the first images can include biometric data of users.

[0103] In implementation, the server can obtain the first images sent by the client (i.e., the terminal device) based on a preset model training period. Among them, the server can store the first images in the corresponding training data sets according to the service identifiers of the target services sent by the client.

[0104] In S504, based on the first images, the first loss function, and the second loss function, the image desensitization model constructed by the preset deep learning algorithm is trained to obtain a trained image desensitization model.

[0105] Among them, the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service.

[0106] In implementation, the training process of the image desensitization model can refer to the relevant content of S208 in the second embodiment, which will not be elaborated here.

[0107] In S506, the trained image desensitization model is sent to the client, so that the client processes the target image containing the biometric data of the target user based on the trained image desensitization model.

[0108] In implementation, the server can retrain the image desensitization model based on the preset model training period and send the trained image desensitization model to the client.

[0109] An embodiment of this specification provides a data processing method, which includes obtaining a training data set for a target service, where the training data set includes multiple first images, and the first images include the biometric data of users; training an image desensitization model constructed by a preset deep learning algorithm based on the first images, a first loss function, and a second loss function to obtain a trained image desensitization model, where the first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service; sending the trained image desensitization model to the client, so that the client processes the target image containing the biometric data of the target user based on the trained image desensitization model. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the malicious third party stealing the target desensitized image during data transmission, and on the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the private data of the target user is improved.

[0110] Embodiment 4

[0111] As Figure 6As shown in the figure, an embodiment of this specification provides a data processing method. The execution subject of this method can be a server, which can be an independent server or a server cluster composed of multiple servers. The method can specifically include the following steps:

[0112] In S602, obtain a training data set for a target service.

[0113] Among them, the training data set contains multiple first images, and the first images contain biometric data of users.

[0114] In S604, based on the first images, the first loss function, and the second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model.

[0115] Among them, the first loss function can be used to make the desensitized images output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized images output by the image desensitization model meet the preset image usage requirements of the target service.

[0116] In S606, send the trained image desensitization model to the client.

[0117] For the specific processing procedures of the above S602 to S604, please refer to the relevant content of S502 to S504 in the first embodiment above, which will not be elaborated here.

[0118] In S608, receive the target desensitized image sent by the client.

[0119] Among them, the target desensitized image can be the desensitized image corresponding to the target image determined by the client based on the trained image desensitization model.

[0120] In S610, process the target service based on the target desensitized image.

[0121] In implementation, the server can perform identity verification processing on the target user based on the target desensitized image, and after the identity verification is passed, process the target service.

[0122] In addition, after the identity verification is passed, the server can also store the target desensitized image as the first image in the training data set of the target service to retrain the image desensitization model during the model training cycle.

[0123] An embodiment of this specification provides a data processing method, which includes obtaining a training data set for a target service. The training data set includes multiple first images, and the first images include biometric data of users. Based on the first images, a first loss function, and a second loss function, an image desensitization model constructed by a preset deep learning algorithm is trained to obtain a trained image desensitization model. The first loss function is used to make the desensitized images output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized images output by the image desensitization model meet the preset image usage requirements of the target service. The trained image desensitization model is sent to the client so that the client can process a target image containing biometric data of a target user based on the trained image desensitization model. Since the first loss function can be used to make the desensitized images output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized images output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, the problem of information leakage caused by malicious third parties stealing the target desensitized images during data transmission can be avoided. On the other hand, the server can execute the target service based on the target desensitized images, that is, the security of the privacy data of the target user is improved while ensuring the normal execution of the target service.

[0124] Embodiment 5

[0125] As Figure 7 shown, an embodiment of this specification provides a data processing method. The execution subject of this method can be a terminal device (i.e., a client) or a server. The terminal device can be a mobile terminal device such as a mobile phone or a tablet computer, or a terminal device such as a personal computer, or a smart wearable device such as a smart watch. The server can be an independent server or a server cluster composed of multiple servers. The method can specifically include the following steps:

[0126] In S702, the server obtains a training data set for a target service.

[0127] Among them, the training data set includes multiple historical first images, and the historical first images include biometric data of users.

[0128] In implementation, as Figure 8 shown, the server can obtain historical first data containing biometric data of users sent by the client within a preset model training period.

[0129] In S704, the server trains an image desensitization model constructed by a preset deep learning algorithm based on historical first images, a first loss function, and a second loss function, and obtains a trained image desensitization model.

[0130] Among them, the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service.

[0131] In S706, the server sends the trained image desensitization model to the client.

[0132] In S708, the client receives a trigger execution instruction for the target service from the target user, and in response to the trigger execution instruction, obtains a first image corresponding to the target user.

[0133] In S710, the client determines a target desensitized image corresponding to the first image based on the pre-trained image desensitization model.

[0134] In S712, the client sends the target desensitized image to the server.

[0135] In S714, the server processes the target service based on the target desensitized image.

[0136] The embodiments of this specification provide a data processing method. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, therefore, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the malicious third party stealing the target desensitized image during data transmission, and on the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the private data of the target user is improved.

[0137] Embodiment Six

[0138] The above is the data processing method provided by the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a data processing device, as Figure 9 shown.

[0139] The data processing device includes: an instruction receiving module 901, an image determination module 902, and an image sending model 903, where:

[0140] An instruction receiving module 901, configured to receive a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image includes biometric data of the target user;

[0141] An image determination module 902, configured to determine a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images, the first loss function is used to make the desensitized image output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized image output by the image desensitization model meet a preset image usage requirement of the target service;

[0142] An image sending module 903, configured to send the target desensitized image to a server, so that the server executes the target service based on the target desensitized image.

[0143] In an embodiment of this specification, the apparatus further includes:

[0144] A first obtaining module, configured to obtain the historical first images;

[0145] A second obtaining module, configured to input the historical first images into the image desensitization model to obtain historical first desensitized images;

[0146] A model training module, configured to determine whether to retrain the image desensitization model according to the historical first images, the historical first desensitized images, the first loss function, and the second loss function, so as to obtain the trained image desensitization model.

[0147] In an embodiment of this specification, the first loss function determines whether the image desensitization model meets the preset image desensitization requirement through the historical first images and the historical first desensitized images, and the second loss function determines whether the image desensitization model meets the preset image usage requirement of the target service through the historical desensitized images.

[0148] In an embodiment of this specification, the first loss function determines whether the image desensitization model meets the preset image desensitization requirement through one or more of a first loss score, a second loss score, and a third loss score. The first loss score is used to determine the desensitization effect of the historical first images and the historical first desensitized images at the pixel level, the second loss score is used to determine the desensitization effect of the historical first images and the historical first desensitized images at the feature vector level, and the third loss score is used to determine the desensitization effect of the historical first images and the historical first desensitized images at the recognition effect level.

[0149] In the embodiments of this specification, the model training module is used for:

[0150] Based on a preset image dimensionality reduction algorithm, obtain a first vector corresponding to the historical first image and a second vector corresponding to the historical first desensitized image, and determine the first loss score based on the first vector and the second vector;

[0151] Based on a pre-trained feature extraction model, obtain a third vector corresponding to the historical first image and a fourth vector corresponding to the historical first desensitized image, and determine the second loss score based on the third vector and the fourth vector. The feature extraction model is obtained by training a model constructed based on a preset feature extraction algorithm using a historical second image;

[0152] Based on a pre-trained first recognition model, obtain a first recognition rate corresponding to the historical first image and a second recognition rate corresponding to the historical first desensitized image, and determine the third loss score based on the first recognition rate and the second recognition rate. The first recognition model is obtained by training a model constructed based on a preset image recognition algorithm using the historical second image;

[0153] Based on a pre-trained second recognition model, obtain a third recognition rate corresponding to the historical first desensitized image, and determine the fourth loss score based on the third recognition rate. The second recognition model is obtained by training a model constructed based on a preset second recognition algorithm using the historical second desensitized image, and the historical second desensitized image is obtained by performing desensitization processing on the historical second image;

[0154] Based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score, determine whether to retrain the image desensitization model.

[0155] In the embodiments of this specification, the second image acquisition is used for:

[0156] Based on a preset image convolution algorithm, perform feature extraction processing on the historical first image to obtain a second image;

[0157] Based on a preset image recombination algorithm, perform image recombination processing on the second image to obtain the historical first desensitized image.

[0158] In the embodiments of this specification, the second acquisition is used for:

[0159] Based on a preset normalization algorithm and a preset signal convolution algorithm, perform compression processing on the historical first image to obtain the second image;

[0160] The model training module is used for:

[0161] Decompress the second image to obtain a third image;

[0162] Based on the third image and the first image, determine a fifth loss value, which is used to determine the degree of difference between the first image and the third image;

[0163] Based on one or more of the first loss score, the second loss score, and the third loss score, as well as the fourth loss score and the fifth loss value, determine whether to retrain the image desensitization model.

[0164] An embodiment of this specification provides a data processing device, which receives a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtains a first image corresponding to the target user. The first image includes biometric data of the target user. Based on a pre-trained image desensitization model, a target desensitized image corresponding to the first image is determined. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service. The target desensitized image is sent to the server so that the server can execute the target service based on the target desensitized image. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, the problem of information leakage caused by the target desensitized image being stolen by a malicious third party during data transmission can be avoided, and on the other hand, the server can execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the privacy data of the target user is improved.

[0165] Embodiment Seven

[0166] The above is the data processing method provided by the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a data processing device, as Figure 10 shown.

[0167] The data processing device includes: a data acquisition module 1001, a model training module 1002, and a data sending module 1003, where:

[0168] A data acquisition module 1001, configured to acquire a training data set for a target service, where the training data set includes a plurality of first images, and the first images include biometric data of a user;

[0169] A model training module 1002, configured to train an image desensitization model constructed by a preset deep learning algorithm based on the first images, a first loss function, and a second loss function, to obtain a trained image desensitization model, where the first loss function is used to make the desensitized images output by the image desensitization model meet a preset image desensitization requirement, and the second loss function is used to make the desensitized images output by the image desensitization model meet a preset image usage requirement of the target service;

[0170] A data sending module 1003, configured to send the trained image desensitization model to a client, so that the client processes a target image including biometric data of a target user based on the trained image desensitization model.

[0171] In an embodiment of this specification, the device further includes:

[0172] A data receiving module, configured to receive a target desensitized image sent by the client, where the target desensitized image is a desensitized image corresponding to the target image determined by the client based on the trained image desensitization model;

[0173] A service processing module, configured to process the target service based on the target desensitized image.

[0174] An embodiment of this specification provides a data processing device that receives a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtains a first image corresponding to the target user. The first image includes biometric data of the target user. Based on a pre-trained image desensitization model, a target desensitized image corresponding to the first image is determined. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service. The target desensitized image is sent to the server so that the server can execute the target service based on the target desensitized image. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitized model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the malicious third party stealing the target desensitized image during data transmission, and on the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the privacy data of the target user is improved.

[0175] Embodiment VIII

[0176] Based on the same idea, an embodiment of this specification also provides a data processing device, as Figure 11 shown.

[0177] The data processing device may vary greatly due to configuration or performance, and may include one or more processors 1101 and a memory 1102. One or more application programs or data may be stored in the memory 1102. Among them, the memory 1102 may be short-term storage or persistent storage. The application programs stored in the memory 1102 may include one or more modules (not shown in the figure), and each module may include a series of computer-executable instructions in the data processing device. Further, the processor 1101 may be set to communicate with the memory 1102 and execute a series of computer-executable instructions in the memory 1102 on the data processing device. The data processing device may also include one or more power supplies 1103, one or more wired or wireless network interfaces 1104, one or more input / output interfaces 1105, and one or more keyboards 1106.

[0178] Specifically, in this embodiment, the data processing device includes a memory and one or more programs, where one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of computer-executable instructions in the data processing device, and is configured to execute the one or more programs by one or more processors, and the one or more programs include computer-executable instructions for performing the following:

[0179] Receive a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image includes biometric data of the target user;

[0180] Based on a pre-trained image desensitization model, determine a target desensitized image corresponding to the first image. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service;

[0181] Send the target desensitized image to the server so that the server executes the target service based on the target desensitized image.

[0182] Optionally, before determining the target desensitized image corresponding to the first image based on the pre-trained image desensitization model, it further includes:

[0183] Obtain the historical first image;

[0184] Input the historical first image into the image desensitization model to obtain a historical first desensitized image;

[0185] Determine whether to retrain the image desensitization model according to the historical first image, the historical first desensitized image, the first loss function, and the second loss function to obtain the trained image desensitization model.

[0186] Optionally, the first loss function determines whether the image desensitization model meets the preset image desensitization requirements through the historical first image and the historical first desensitized image, and the second loss function determines whether the image desensitization model meets the preset image usage requirements of the target service through the historical desensitized image.

[0187] Optionally, the first loss function determines whether the image desensitization model meets the preset image desensitization requirements through one or more of a first loss score, a second loss score, and a third loss score. The first loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the pixel level. The second loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the feature vector level. The third loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level.

[0188] Optionally, determining whether to retrain the image desensitization model according to the historical first image, the historical first desensitized image, the first loss function, and the second loss function includes:

[0189] Based on a preset image dimensionality reduction algorithm, obtain a first vector corresponding to the historical first image and a second vector corresponding to the historical first desensitized image, and determine the first loss score based on the first vector and the second vector;

[0190] Based on a pre-trained feature extraction model, obtain a third vector corresponding to the historical first image and a fourth vector corresponding to the historical first desensitized image, and determine the second loss score based on the third vector and the fourth vector. The feature extraction model is obtained by training a model constructed based on a preset feature extraction algorithm using the historical second image;

[0191] Based on a pre-trained first recognition model, obtain a first recognition rate corresponding to the historical first image and a second recognition rate corresponding to the historical first desensitized image, and determine the third loss score based on the first recognition rate and the second recognition rate. The first recognition model is obtained by training a model constructed based on a preset image recognition algorithm using the historical second image;

[0192] Based on a pre-trained second recognition model, obtain a third recognition rate corresponding to the historical first desensitized image, and determine a fourth loss score based on the third recognition rate. The second recognition model is obtained by training a model constructed based on a preset second recognition algorithm using the historical second desensitized image, and the historical second desensitized image is obtained by desensitizing the historical second image;

[0193] Determine whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score.

[0194] Optionally, inputting the historical first image into the image desensitization model to obtain a historical first desensitized image includes:

[0195] Performing feature extraction processing on the historical first image based on a preset image convolution algorithm to obtain a second image;

[0196] Performing image recombination processing on the second image based on a preset image recombination algorithm to obtain the historical first desensitized image.

[0197] Optionally, performing feature extraction processing on the historical first image based on a preset image convolution algorithm to obtain a second image includes:

[0198] Performing compression processing on the historical first image based on a preset normalization algorithm and a preset signal convolution algorithm to obtain the second image;

[0199] Determining whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score includes:

[0200] Performing decompression processing on the second image to obtain a third image;

[0201] Determining a fifth loss value based on the third image and the first image, where the fifth loss value is used to determine the degree of difference between the first image and the third image;

[0202] Determining whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score and the fifth loss value.

[0203] In addition, specifically in this embodiment, the data processing device includes a memory and one or more programs, where one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of computer-executable instructions in the data processing device, and is configured to be executed by one or more processors. The one or more programs include the following computer-executable instructions:

[0204] Obtaining a training data set for a target service, where the training data set includes multiple first images, and the first images include biometric data of users;

[0205] Based on the first image, the first loss function, and the second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service;

[0206] Send the trained image desensitization model to the client so that the client can process a target image containing biometric data of a target user based on the trained image desensitization model.

[0207] Optionally, the method further includes:

[0208] Receive the target desensitized image sent by the client, where the target desensitized image is the desensitized image corresponding to the target image determined by the client based on the trained image desensitization model;

[0209] Process the target service based on the target desensitized image.

[0210] An embodiment of this specification provides a data processing device. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, the image desensitization model trained based on the first loss function and the second loss function can make the target desensitization model take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the target desensitized image being stolen by a malicious third party during data transmission, and on the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, it improves the security of the privacy data of the target user.

[0211] Embodiment Nine

[0212] An embodiment of this specification also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, it implements each process of the above data processing method embodiment and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0213] An embodiment of this specification provides a computer-readable storage medium. Since the first loss function can be used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements, and the second loss function can be used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service, therefore, the image desensitization model trained based on the first loss function and the second loss function can enable the target desensitization model to take into account both the image desensitization requirements and the image usage requirements of the target service. In this way, on the one hand, it can avoid the information leakage problem caused by the malicious third party stealing the target desensitized image during data transmission. On the other hand, it can enable the server to execute the target service based on the target desensitized image, that is, while ensuring the normal execution of the target service, the security of the target user's private data is improved.

[0214] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0215] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by the user programming the device. Designers can program themselves to "integrate" a digital system onto a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called Hardware Description Language (HDL), and there is not just one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply making a little logical programming of the method flow using the above-mentioned several hardware description languages and programming it into the integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0216] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same functions. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0217] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0218] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0219] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0220] Embodiments of this specification are described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0221] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0222] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0223] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0224] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0225] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0226] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0227] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0228] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0229] The various embodiments in this specification are described in a progressive manner. For the identical or similar parts among the various embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the related content.

[0230] The above is only the embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A data processing method, comprising: Receiving a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtaining a first image corresponding to the target user, where the first image includes biometric data of the target user; Based on a pre-trained image desensitization model, determining a target desensitized image corresponding to the first image, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognizability of the historical first desensitized image. The historical first desensitized image is a desensitized image obtained by the image desensitization model performing desensitization processing on the historical first image; the image desensitization model is trained through the third loss score and the fourth loss score, where the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate, and the first recognition rate and the second recognition rate are recognition rates obtained by processing the historical first image and the historical first desensitized image respectively based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate, and the third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Sending the target desensitized image to a server so that the server executes the target service based on the target desensitized image.

2. The method according to claim 1, before determining the target desensitized image corresponding to the first image based on the pre-trained image desensitization model, further comprising: Obtaining the historical first image; Inputting the historical first image into the image desensitization model to obtain the historical first desensitized image; Determining whether to retrain the image desensitization model according to the historical first image, the historical first desensitized image, the first loss function, and the second loss function to obtain the trained image desensitization model.

3. The method according to claim 2, where the first loss function determines whether the image desensitization model meets the preset image desensitization requirements through the historical first image and the historical first desensitized image, and the second loss function determines whether the image desensitization model meets the preset image usage requirements of the target service through the historical first desensitized image.

4. According to the method described in claim 3, the first loss function determines whether the image desensitization model meets the preset image desensitization requirements through one or more of a first loss score, a second loss score, and a third loss score. The first loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the pixel level. The second loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the feature vector level. The third loss score is used to determine the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level.

5. According to the method described in claim 4, determining whether to retrain the image desensitization model based on the historical first image, the historical first desensitized image, the first loss function, and the second loss function includes: Based on a preset image dimensionality reduction algorithm, obtaining a first vector corresponding to the historical first image and a second vector corresponding to the historical first desensitized image, and determining the first loss score based on the first vector and the second vector; Based on a pre-trained feature extraction model, obtaining a third vector corresponding to the historical first image and a fourth vector corresponding to the historical first desensitized image, and determining the second loss score based on the third vector and the fourth vector. The feature extraction model is obtained by training a model constructed based on a preset feature extraction algorithm using a historical second image; Based on the pre-trained first recognition model, obtaining a first recognition rate corresponding to the historical first image and a second recognition rate corresponding to the historical first desensitized image, and determining the third loss score based on the first recognition rate and the second recognition rate. The first recognition model is obtained by training a model constructed based on a preset image recognition algorithm using the historical second image; Based on the pre-trained second recognition model, obtaining a third recognition rate corresponding to the historical first desensitized image, and determining the fourth loss score based on the third recognition rate. The second recognition model is obtained by training a model constructed based on a preset second recognition algorithm using the historical second desensitized image, and the historical second desensitized image is obtained by performing desensitization processing on the historical second image; Determining whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score.

6. According to the method described in claim 5, inputting the historical first image into the image desensitization model to obtain the historical first desensitized image includes: Performing feature extraction processing on the historical first image based on a preset image convolution algorithm to obtain a second image; Performing image recombination processing on the second image based on a preset image recombination algorithm to obtain the historical first desensitized image.

7. According to the method described in claim 6, performing feature extraction processing on the historical first image based on a preset image convolution algorithm to obtain a second image includes: Compress the historical first image based on a preset normalization algorithm and a preset signal convolution algorithm to obtain the second image; Determining whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score, includes: Decompress the second image to obtain a third image; Based on the third image and the first image, determine a fifth loss value, where the fifth loss value is used to determine the degree of difference between the first image and the third image; Determine whether to retrain the image desensitization model based on one or more of the first loss score, the second loss score, and the third loss score, and the fourth loss score and the fifth loss value.

8. A data processing method, including: Obtain a training data set for a target service, where the training data set includes multiple first images, and the first images include biometric data of users; Based on the first image, a first loss function, and a second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirement through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirement of the target service through a fourth loss score according to the recognizability of the historical first desensitized image. The historical first desensitized image is the desensitized image obtained by the image desensitization model performing desensitization processing on the historical first image; the image desensitization model is trained through the third loss score and the fourth loss score, where the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively the recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the distinguishability and information loss of the image after image desensitization according to the third recognition rate. The third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Send the trained image desensitization model to the client so that the client processes a target image including biometric data of a target user based on the trained image desensitization model.

9. The method according to claim 8, the method further includes: Receive the target desensitized image sent by the client, where the target desensitized image is the desensitized image corresponding to the target image determined by the client based on the trained image desensitization model; Process the target service based on the target desensitized image.

10. A data processing device, including: An instruction receiving module, configured to receive a trigger execution instruction for a target service from a target user, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image includes biometric data of the target user; An image determination module, configured to determine a target desensitized image corresponding to the first image based on a pre-trained image desensitization model, where the image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognition rate of the historical first desensitized image. The historical first desensitized image is a desensitized image obtained by the image desensitization model performing desensitization processing on the historical first image. The image desensitization model is trained through the third loss score and the fourth loss score. Among them, the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are recognition rates obtained by processing the historical first image and the historical first desensitized image respectively based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate. The third recognition rate is a recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; An image sending module, configured to send the target desensitized image to a server, so that the server executes the target service based on the target desensitized image.

11. A data processing device, comprising: A data acquisition module, configured to acquire a training data set for a target service, where the training data set includes a plurality of first images, and the first images include biometric data of users; A model training module, configured to train an image desensitization model constructed by a preset deep learning algorithm based on the first image, the first loss function, and the second loss function, to obtain a trained image desensitization model. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognizability of the historical first desensitized image. The historical first desensitized image is the desensitized image obtained by the image desensitization model through desensitization processing of the historical first image. The image desensitization model is trained through the third loss score and the fourth loss score. Among them, the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively the recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the distinguishability and the loss of information amount of the image after image desensitization according to the third recognition rate. The third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; A data sending module, configured to send the trained image desensitization model to a client, so that the client processes a target image containing biometric data of a target user based on the trained image desensitization model.

12. A data processing device, the data processing device includes: A processor; And A memory arranged to store computer-executable instructions, and when the executable instructions are executed, the processor is caused to: Receive a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image includes biometric data of the target user; Based on a pre-trained image desensitization model, determine a target desensitized image corresponding to the first image. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and historical first images. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image at the recognition effect level. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognizability of the historical first desensitized image. The historical first desensitized image is a desensitized image obtained by the image desensitization model performing desensitization processing on the historical first image. The image desensitization model is trained through the third loss score and the fourth loss score. Among them, the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate. The third recognition rate is a recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Send the target desensitized image to the server so that the server executes the target service based on the target desensitized image.

13. A data processing device, the data processing device includes: A processor; And A memory arranged to store computer-executable instructions, and the executable instructions, when executed, cause the processor to: Obtain a training data set for a target service, the training data set includes a plurality of first images, and the first images include biometric data of a user; Based on the first image, the first loss function, and the second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image in terms of recognition effect. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognition rate of the historical first desensitized image. The historical first desensitized image is the desensitized image obtained by the image desensitization model through desensitization processing of the historical first image. The image desensitization model is trained through the third loss score and the fourth loss score. Among them, the third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively the recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate. The third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Send the trained image desensitization model to the client so that the client processes a target image containing biometric data of a target user based on the trained image desensitization model.

14. A storage medium, which is used to store computer-executable instructions, and the executable instructions, when executed by a processor, implement the following process: Receive a trigger execution instruction of a target user for a target service, and in response to the trigger execution instruction, obtain a first image corresponding to the target user, where the first image contains biometric data of the target user; Based on a pre-trained image desensitization model, determine a target desensitized image corresponding to the first image. The image desensitization model is obtained by training a model constructed by a preset deep learning algorithm based on a first loss function, a second loss function, and a historical first image. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image in terms of recognition effect. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognition rate of the historical first desensitized image. The historical first desensitized image is the desensitized image obtained by the image desensitization model through desensitization processing of the historical first image. The image desensitization model is trained through the third loss score and the fourth loss score. Among them, The third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively the recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate. The third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Send the target desensitized image to the server so that the server executes the target service based on the target desensitized image.

15. A storage medium for storing computer-executable instructions, which when executed by a processor implement the following process: Obtain a training data set for a target service, the training data set includes a plurality of first images, and the first images include biometric data of users; Based on the first images, a first loss function, and a second loss function, train an image desensitization model constructed by a preset deep learning algorithm to obtain a trained image desensitization model. The first loss function is used to make the desensitized image output by the image desensitization model meet the preset image desensitization requirements through a third loss score according to the desensitization effect of the historical first image and the historical first desensitized image in terms of recognition effect. The second loss function is used to make the desensitized image output by the image desensitization model meet the preset image usage requirements of the target service through a fourth loss score according to the recognition rate of the historical first desensitized image. The historical first desensitized image is the desensitized image obtained by the image desensitization model performing desensitization processing on the historical first image; the image desensitization model is trained through the third loss score and the fourth loss score, where The third loss score is used to control the recognizability after image desensitization according to the ratio between the second recognition rate and the first recognition rate. The first recognition rate and the second recognition rate are respectively the recognition rates obtained by processing the historical first image and the historical first desensitized image based on a pre-trained first recognition model. The fourth loss score is used to control the loss of image distinguishability and information volume after image desensitization according to the third recognition rate. The third recognition rate is the recognition rate obtained by processing the historical first desensitized image based on a pre-trained second recognition model; Send the trained image desensitization model to the client so that the client processes a target image containing biometric data of a target user based on the trained image desensitization model.

Images