Probabilistic-diffusion-based method for protecting and recovering facial data by means of differential privacy and gradient
Through the differential privacy gradient method of probability diffusion, the face data is protected by neural networks and Gaussian tensors, and the pictures are restored by simulated gradient descent, solving the problems of face data protection and recovery, realizing data security and privacy protection.
Patent Information
- Application Number
- PCT/CN2023/097008
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-30
- Publication Date
- 2024-12-05
AI Technical Summary
The prior art is difficult to effectively protect and restore facial data, especially due to the direct recognition and uniqueness of facial information, which is difficult to reverse once it is leaked, and the existing methods cannot effectively protect the privacy and security of data.
Using a differential privacy gradient method based on probability diffusion, the neural network uses time tensors and Gaussian tensors to protect face data, and uses simulated gradient descent and Gaussian random noise during the recovery process to achieve the protection of face pictures and reduction.
It realizes effective protection and recovery of facial data, ensures one-to-one correspondence of data in high-dimensional space, and ensures data privacy and security.
Smart Images

Figure CN2023097008_05122024_PF_FP_ABST
Abstract
Description
Differential privacy gradient face data protection and recovery method based on probability diffusion Technical Field
[0001] The present invention relates to computer network technology, artificial intelligence technology, and privacy protection technology, and in particular to a differential privacy gradient face data protection and recovery method based on probability diffusion. Background Art
[0002] As AI empowers ever-increasing depth and breadth, facial recognition technology is maturing, accelerating its commercialization, and gaining widespread adoption across various sectors. As a universal identifier for mutual recognition and verification in human society, the face is instantly recognizable, unique, and readily accessible.
[0003] Facial information is sensitive personal information and must be protected according to special rules for sensitive personal information. At the same time, due to the direct recognition, uniqueness, and specificity of facial information, once an individual's facial information is leaked, the loss of privacy is difficult to modify or reverse.
[0004] The security and privacy protection of facial information data are increasingly valued by data users and data providers. Therefore, there is an urgent need to propose a facial data protection and recovery method.
[0005] Summary of the Invention
[0006] The purpose of the present invention is to address the deficiencies of the existing technology and provide a differential privacy gradient face data protection and recovery method based on probability diffusion.
[0007] According to a first aspect of an embodiment of the present invention, a differential privacy gradient face data protection and recovery method based on probability diffusion is provided, the method comprising:
[0008] Step S1: Obtain a face image dataset and perform random sampling to obtain input data; obtain a time tensor; randomly sample the input data from a standard Gaussian distribution according to the size of the input data to obtain a first Gaussian tensor; input the time tensor and the first Gaussian tensor into a neural network for training to obtain an approximate Gaussian tensor and a hidden layer tensor;
[0009] Step S2: randomly sample from the standard Gaussian distribution according to the size of the hidden layer tensor to obtain a second Gaussian tensor; generate an estimated image, input the estimated image into the trained neural network to obtain an approximate Gaussian tensor function, and perform simulated gradient descent on the estimated image based on the second Gaussian tensor and the approximate Gaussian tensor function to obtain a restored face image.
[0010] Furthermore, the step S1 includes the following sub-steps:
[0011] Step S101, obtain a face image dataset and perform random sampling to obtain input data x0;
[0012] Step S102, randomly sampling time from the time scale 1 to T with uniform probability to obtain the time tensor t;
[0013] Step S103, randomly sampling from a standard Gaussian distribution to obtain a first Gaussian tensor e with the same size as the input data x0;
[0014] Step S104: input the first Gaussian tensor e and the time tensor t into a neural network, train the neural network, and obtain the approximate Gaussian tensor e output by the neural network after the training converges. s and the hidden layer tensor x T .
[0015] Furthermore, the face image dataset adopts the celebA dataset.
[0016] Furthermore, the structure of the neural network adopts CNN or VGG.
[0017] Furthermore, training the neural network includes:
[0018] The loss function loss is optimized using the gradient descent method. After the loss function loss converges, the training is completed and the approximate Gaussian tensor e output by the neural network is obtained. s ; The expression of loss function loss is as follows:
[0019] Where, α t is the initialization parameter of the neural network, and e is the first Gaussian tensor.
[0020] Furthermore, training the neural network also includes:
[0021] The Xavier method is used to initialize the parameters of the neural network to obtain the initialization parameters α of the neural network. t .
[0022] Furthermore, training the neural network also includes:
[0023] During the neural network training process, the hidden layer tensor x is obtained by publishing with differential privacy protection. T .
[0024] Furthermore, step S2 includes the following sub-steps:
[0025] Step S201, randomly sampling from the standard Gaussian distribution to obtain the hidden layer tensor x T Gaussian tensor z of the same size;
[0026] Step S202: Generate an estimated image The estimated image x t Input into the trained neural network to calculate the approximate Gaussian tensor function e s (x T ,t), based on the Gaussian tensor z and the approximate Gaussian tensor function e s (x T ,t) in the estimated image Perform simulated gradient descent on the image, multiply the estimated image by the sparse step size during the simulated gradient descent process, and add Gaussian random noise to obtain the restored face image
[0027] Furthermore, based on the Gaussian tensor z and the approximate Gaussian tensor function e s (x T ,t) in the estimated image The expression for simulated gradient descent is as follows:
[0028] Where, β t is a hyperparameter, σ t is Gaussian random noise.
[0029] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned differential privacy gradient face data protection method based on the probability diffusion equation.
[0030] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the above-mentioned differential privacy gradient face data protection method based on the probability diffusion equation is implemented.
[0031] The beneficial effects of the present invention are as follows: the present invention provides a differential privacy gradient face data protection and recovery method based on the probability diffusion equation. In the process of face data protection, the present invention uses the time tensor and the first Gaussian tensor to train the neural network to obtain the approximate Gaussian tensor and the hidden layer tensor, wherein the hidden layer tensor is obtained through differential privacy protection and release during the neural network training process, thereby realizing the protection of the original face data set. In the process of face data recovery, the present invention simulates gradient descent on the estimated image based on the hidden layer tensor and the second Gaussian tensor and adds Gaussian random noise to restore the real face image. Moreover, the first Gaussian tensor and the second Gaussian tensor constructed by the present invention are one-to-one corresponding to the original face image in high-dimensional space, thereby realizing the protection of the face image and the restoration from the hidden layer tensor to the face image. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0033] FIG1 is a flowchart of a differential privacy gradient face data protection and recovery method based on probability diffusion provided by an embodiment of the present invention;
[0034] FIG2 is a schematic diagram of a differential privacy gradient face data protection and recovery method based on probability diffusion provided by an embodiment of the present invention;
[0035] FIG3 is a schematic diagram of the face data protection stage provided by an embodiment of the present invention;
[0036] FIG4 is a schematic diagram of the face data recovery stage provided by an embodiment of the present invention;
[0037] FIG5 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0039] It should be noted that, unless there is any conflict, the features in the following embodiments and implementations may be combined with each other.
[0040] As shown in Figures 1 and 2, the present invention provides a differential privacy gradient face data protection and recovery method based on probability diffusion, which specifically includes the following steps:
[0041] Step S1: Obtain a face image dataset and perform random sampling to obtain input data; obtain a time tensor; perform random sampling from a standard Gaussian distribution according to the size of the input data to obtain a first Gaussian tensor; input the time tensor and Gaussian tensor into a neural network for training to obtain an approximate Gaussian tensor;
[0042] Specifically, as shown in FIG3 , step S1 specifically includes the following sub-steps:
[0043] Step S101: Obtain a face image dataset and perform random sampling to obtain input data x0.
[0044] In this example, the face image dataset uses the celebA dataset, and the celebA dataset is randomly sampled to obtain the input data x0.
[0045] Step S102 : randomly perform time sampling from the time scale 1 to T with uniform probability to obtain a time tensor t.
[0046] Step S103 : Randomly sample from a standard Gaussian distribution to obtain a first Gaussian tensor e with the same size as the input data x0.
[0047] Step S104: input the Gaussian tensor e and the time tensor t into a neural network, train the neural network, and obtain the approximate Gaussian tensor e output by the neural network after the training converges. s and the hidden layer tensor x T .
[0048] In this example, the structure of the neural network adopts CNN (Convolutional Neural Networks) or VGG (Visual Geometry Group, convolutional neural network).
[0049] Training the neural network includes: optimizing the loss function loss using the gradient descent method, completing the training after the loss function loss converges, and obtaining the approximate Gaussian tensor e output by the neural network s Among them, the expression of loss function loss is as follows:
[0050] Where, α t is the initialization parameter of the neural network, and e is the first Gaussian tensor. Among them, the initialization parameter α of the neural network t The parameters of the neural network are initialized using the Xavier initialization method.
[0051] It should be noted that the hidden layer tensor x T It is obtained through differential privacy protection during the neural network training process.
[0052] Step S2: randomly sample from the standard Gaussian distribution according to the size of the hidden layer tensor to obtain a second Gaussian tensor; generate an estimated image, input the estimated image into the trained neural network to obtain an approximate Gaussian tensor function, and perform simulated gradient descent on the estimated image based on the second Gaussian tensor and the approximate Gaussian tensor function to obtain a restored face image.
[0053] Specifically, as shown in FIG4 , step S2 specifically includes the following sub-steps:
[0054] Step S201, randomly sampling from the standard Gaussian distribution to obtain the hidden layer tensor x T The first Gaussian tensor z of the same size;
[0055] Step S202: Generate an estimated image The estimated image x t Input into the trained neural network to calculate the approximate Gaussian tensor function e s (x T ,t), based on the second Gaussian tensor z and the approximate Gaussian tensor function e s (x T ,t) in the estimated image Perform simulated gradient descent on the image to get the restored face image
[0056] Specifically, based on the Gaussian tensor e s (x,t) in the estimated image The simulated gradient descent includes: multiplying the estimated image by the sparse step size And add Gaussian random noise during the gradient descent process; the expression is as follows:
[0057] Where, β t is a hyperparameter, σ t is Gaussian random noise.
[0058] In summary, the present invention provides a differential privacy gradient face data protection and recovery method based on the probability diffusion equation. In the process of face data protection, the present invention uses the time tensor and the first Gaussian tensor to train the neural network to obtain the approximate Gaussian tensor and the hidden layer tensor, wherein the hidden layer tensor is obtained through differential privacy protection and release during the neural network training process, thereby realizing the protection of the original face data set. In the process of face data recovery, the present invention simulates gradient descent on the estimated image based on the hidden layer tensor and the second Gaussian tensor and adds Gaussian random noise to restore the real face image. Moreover, the first Gaussian tensor and the second Gaussian tensor constructed by the present invention are one-to-one corresponding to the original face image in high-dimensional space, thereby realizing the protection of the face image and the restoration from the hidden layer tensor to the face image.
[0059] As shown in Figure 5, an embodiment of the present application provides an electronic device, which includes a memory 101 for storing one or more programs and a processor 102. When the one or more programs are executed by the processor 102, the method of any one of the above-mentioned first aspects is implemented.
[0060] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes the software programs and modules stored in the memory 101 to perform various functional applications and data processing. The communication interface 103 can be used to communicate signaling or data with other node devices.
[0061] Among them, the memory 101 can be, but is not limited to, a random access memory 101 (Random Access Memory, RAM), a read-only memory 101 (Read Only Memory, ROM), a programmable read-only memory 101 (Programmable Read-Only Memory, PROM), an erasable programmable read-only memory 101 (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable read-only memory 101 (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.
[0062] The processor 102 may be an integrated circuit chip having signal processing capabilities. The processor 102 may be a general-purpose processor 102, including a central processing unit (CPU) 102, a network processor (NP) 102, etc.; it may also be a digital signal processing (DSP) 102, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0063] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the methods and systems, methods, and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions.
[0064] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0065] On the other hand, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the method of any one of the first aspects described above when executed by the processor 102. If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory 101 (ROM, Read-Only Memory), a random access memory 101 (RAM, Random Access Memory), a magnetic disk or an optical disk.
[0066] The above embodiments are intended only to illustrate the design concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The scope of protection of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design concepts disclosed in the present invention are within the scope of protection of the present invention.
Claims
1. A differential privacy gradient face data protection and recovery method based on probability diffusion, characterized in that: The method comprises: Step S1, obtaining a face image dataset and performing random sampling to obtain input data; obtaining a time tensor; performing random sampling from a standard Gaussian distribution according to the size of the input data to obtain a first Gaussian tensor; inputting the time tensor and the first Gaussian tensor into a neural network for training to obtain an approximate Gaussian tensor and a hidden layer tensor; Step S2, randomly sampling from the standard Gaussian distribution according to the size of the hidden layer tensor to obtain a second Gaussian tensor; generating an estimated image, inputting the estimated image into the trained neural network to obtain an approximate Gaussian tensor function, and performing simulated gradient descent on the estimated image based on the second Gaussian tensor and the approximate Gaussian tensor function to obtain a restored face image.
2. According to the probability diffusion-based differential privacy gradient face data protection and recovery method of claim 1, it is characterized in that: The step S1 comprises the following sub-steps: Step S101, obtaining a face image dataset and performing random sampling to obtain input data x0; Step S102, randomly sampling time from the time scale from 1 to T with uniform probability to obtain a time tensor t; Step S103, randomly sampling from a standard Gaussian distribution to obtain a first Gaussian tensor e with the same size as the input data x0; Step S104, inputting the first Gaussian tensor e and the time tensor t into a neural network, training the neural network, and obtaining an approximate Gaussian tensor e output by the neural network after the training converges. s and the hidden layer tensor x T .
3. According to claim 2, the differential privacy gradient face data protection and recovery method based on probability diffusion is characterized in that: The face image dataset adopts the celebA dataset; the structure of the neural network adopts CNN or VGG.
4. According to claim 2, the differential privacy gradient face data protection and recovery method based on probability diffusion is characterized in that: Training a neural network involves: The loss function loss is optimized using the gradient descent method. After the loss function loss converges, the training is completed and the approximate Gaussian tensor e output by the neural network is obtained. s ; The expression of loss function loss is as follows: In the formula, α t is the initialization parameter of the neural network, and e is the first Gaussian tensor.
5. According to claim 4, the differential privacy gradient face data protection and recovery method based on probability diffusion is characterized in that: Training a neural network also involves: The Xavier method is used to initialize the parameters of the neural network and the initialization parameters α of the neural network are obtained. t .
6. The differential privacy gradient face data protection and recovery method based on probability diffusion according to claim 1 or 2, characterized in that: Training a neural network also involves: During the neural network training process, the hidden layer tensor x is obtained by publishing with differential privacy protection. T .
7. The differential privacy gradient face data protection and recovery method based on probability diffusion according to claim 1 is characterized in that: The step S2 comprises the following sub-steps: Step S201, randomly sampling from a standard Gaussian distribution to obtain the hidden layer tensor x T A second Gaussian tensor z of the same size; Step S202: Generate an estimated image The estimated image x t Input into the trained neural network to calculate the approximate Gaussian tensor function e s (x T ,t), based on the second Gaussian tensor z and the approximate Gaussian tensor function e s (x T ,t) in the estimated image Perform simulated gradient descent on the image, multiply the estimated image by the sparse step size during the simulated gradient descent process, and add Gaussian random noise to obtain the restored face image 8. The differential privacy gradient face data protection and recovery method based on probability diffusion according to claim 7 is characterized in that: Based on the Gaussian tensor z and the approximate Gaussian tensor function e s (x T ,t) in the estimated image The expression for simulated gradient descent is as follows: In the formula, β t is a hyperparameter, σ t is Gaussian random noise.
9. An electronic device, comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the differential privacy gradient face data protection method based on the probability diffusion equation as described in any one of claims 1-8 above.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, the differential privacy gradient face data protection method based on the probability diffusion equation as described in any one of claims 1-8 is implemented.
Citation Information
Patent Citations
Differential privacy deep learning classification method based on analytic Gaussian mechanism
CN111144500A
Face recognition method and equipment
CN114283483A
Image steganography method based on diffusion model
CN116091288A
Training neural networks using learned optimizers
US20220092429A1
Cited By
Traffic infrastructure monitoring data probability abnormal value diagnosis method based on conditional diffusion model
CN119862510A
Detection method and system based on diffusion manifold perception and state space alignment
CN121617158A
