Identity verification method for generating face image based on electroencephalogram signal

Abstract

The invention relates to the technical field of identity verification, and provides an identity verification method for generating a face image based on an electroencephalogram signal, which is characterized by comprising the following steps: S10, a registration stage: collecting the electroencephalogram signal and the face image of a user as registration information; s20, a model training stage: jointly training a face generation model based on the electroencephalogram signal and the face image in the registration information; and S30, a verification stage: collecting an electroencephalogram signal of the user as verification information, inputting the electroencephalogram signal into the trained face generation model to generate a face image, and comparing the face image with the user himself / herself to realize identity verification. The face generation model is jointly trained by collecting the electroencephalogram and face data of the user, the electroencephalogram signals of the user are collected in the application stage, and the face image can be directly generated through the face generation model, so that the identity authentication effect is achieved.

Description

Technical Field

[0001] The invention belongs to the field of identity authentication, and in particular relates to an identity authentication method for generating a face image based on an electroencephalogram signal. Background Art

[0002] With the rapid development of information technology and the popularization of network applications, the increasing security needs of all walks of life have prompted the widespread application of biometric technology. Biometric technology can greatly improve the accuracy and security of identity authentication by identifying and verifying the inherent physiological characteristics of the human body, such as fingerprints, irises, faces and voiceprints, and has become an indispensable part of modern security systems.

[0003] Among them, EEG signals are a unique biometric feature that is highly personal and difficult to replicate, providing an innovative and secure solution for identity recognition. EEG-based brainprint recognition technology measures brain activity through EEG and extracts EEG signal features of individuals under specific cognition. These features present unique patterns in each person's EEG, making brainprint recognition a highly secure and difficult-to-forge biometric method, thus providing higher security. With the advancement of technology, brainprint recognition is receiving more and more attention, especially in areas with high security requirements such as financial services, healthcare, and personal electronic devices. This technology can not only improve the convenience and security of identity authentication, but may also change the way we interact with technology in the future.

[0004] However, the current brainprint recognition algorithm is mainly based on the discriminant model. Its general process is to collect EEG signals through hardware equipment and input the EEG signals into the classification model to obtain the classification prediction results. This method has certain defects, that is, the recognition results cannot be visualized in real time during the recognition process, and it is not easy for people to understand the characteristics and data in the recognition process. In addition, the brainprint signal itself is complex. Brainprint recognition generally identifies the user's ID information through the collected EEG signals. It cannot directly show the differences between individuals and it is difficult to clearly distinguish different individuals. Summary of the invention

[0005] In view of the above-mentioned defects of the prior art, the present invention proposes an identity authentication method based on EEG signal to generate a face image. The method collects the user's EEG and face data to jointly train a face generation model. In the application stage, the user's EEG signal is collected and passed through the face generation model to directly generate a face image to achieve the effect of identity authentication, including the following steps:

[0006] S10: In the registration phase, the user's EEG signal and face image are collected as registration information;

[0007] S20: Model training phase, the face generation model is jointly trained based on the EEG signals and face images in the registration information;

[0008] S30: Verification stage: The user's EEG signal is collected as verification information, input into the trained face generation model to generate a face image, and compared with the user himself to achieve identity authentication.

[0009] Preferably, the S10 includes:

[0010] In the registration stage, the EEG registration signal is collected through an EEG cap based on an EEG signal induction paradigm, wherein the EEG signal induction paradigm includes visual evoked potential, event-related potential and motor imagery;

[0011] During the collection process, the user performs tasks under different paradigms according to the prompts to induce EEG signals, and collects facial images through the camera. A frontal face photo of the user is taken during the process of collecting EEG. The EEG registration signal and the face registration image are bound and submitted to the model for training.

[0012] Preferably, the S20 includes the following steps:

[0013] S2001: pre-training feature encoder E using EEG registration signal;

[0014] S2002: Encode the EEG registration signal e into a feature vector c using a pre-trained feature encoder E;

[0015] S2003: training a face generation model using feature vectors and face registration images;

[0016] The S2001 includes:

[0017] The EEG registration signal is passed through the feature encoder E to obtain the feature vector, the feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result, the encoder parameters are updated based on the loss calculated according to the true category label, and the encoder E is obtained after iterative repeated training. The specific formula for encoding the feature encoder E is as follows:

[0018] c=E(e)

[0019] In the formula, e is the EEG registration signal, c is the encoded feature vector;

[0020] The feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result. The formula is as follows:

[0021] z=Wc+b

[0022]

[0023] In the formula, z is the classification prediction result, W is the weight matrix, b is the bias, Softmax(·) is the Softmax function, and z i is the i-th component in z, z j is the jth component in z;

[0024] The loss function used in the training process is the cross entropy loss, and the formula is as follows:

[0025]

[0026] Where n is the number of categories, y is the true category label, and p is the predicted category probability.

[0027] Preferably, the S2002 includes:

[0028] During the registration process, all EEG registration signals e are encoded into feature vectors c through the pre-trained feature encoder E. All feature vectors of the same user are averaged and bound to the corresponding face registration images for training the face generation model.

[0029] Preferably, the S2003 includes:

[0030] In the registration phase, the feature vector is used to guide the diffusion model to generate face images from random noise data. The generated face images are compared with the real face images to calculate the loss. The loss is back-propagated to calculate the gradient to update the model parameters. The trained face generation model can be obtained by iteratively repeating the update. The formula is as follows:

[0031] The forward diffusion process formula is as follows:

[0032]

[0033] The reverse generation process formula is as follows:

[0034] p(x t-1 |x t )=N(x t-1 ;u θ (x t ,t,c),σ θ (x t ,t,c))

[0035] In the formula, q(x t |x0) is x under the condition that the distribution of x0 is known t The probability distribution function of x t is a Gaussian distributed random variable, is the forward noise mean, (1-α t )I is the forward noise variance, p(x t-1 |x t ) is a known xt Under the condition of distribution x t-1 The probability distribution function, u θ (x t ,t,c) is the reverse denoising mean, σ θ (x t ,t,c) is the reverse denoising variance. The prediction of the reverse denoising mean and variance can be converted into the prediction of noise;

[0036] A neural network is used as a noise prediction model, and the noise prediction model is used to t Predict x t-1 The distribution of , through continuous reverse prediction, the face image can be generated from the noise image. The reverse prediction formula is as follows:

[0037]

[0038] In the formula, β t is the moment hyperparameter, ε is a standard normal distribution random variable;

[0039] The training goal of iterative training is to optimize the variational upper bound of the negative log-likelihood. The loss function formula during iterative training is as follows:

[0040]

[0041] In the formula, Find the numerical expectation under the premise of knowing t,x0,ε,ε θ (x t ,t,c) is the estimated noise value, ε is the standard normal distribution random noise, x represents the image in the denoising step, x has different forms at different time steps, t represents the time step used to provide step information for the model, and the noise prediction model ε is obtained by repeatedly iterative training and updating the model parameters. θ , using the noise prediction model ε θ Generate face images.

[0042] Preferably, the S30 includes the following steps:

[0043] S3001: Collecting the user's EEG verification signal;

[0044] S3002: Encode the EEG registration signal into a feature vector using a feature encoder;

[0045] S3003: using the feature vector to guide the face generation model to generate a face image;

[0046] S3004: Compare the generated face images to achieve identity authentication.

[0047] Preferably, the S3001 includes:

[0048] In the verification stage, a paradigm is used to induce EEG signals, and the induced EEG signals are collected as EEG verification signals. The EEG verification signals and the EEG signals in the registration stage undergo the same preprocessing process for subsequent generation of facial images.

[0049] Preferably, the S3002 includes:

[0050] The collected EEG verification signal is encoded into a feature vector using the pre-trained feature encoder E. The specific encoding formula is as follows:

[0051] c=E(e)

[0052] Where e is the EEG verification signal and c is the encoded feature vector.

[0053] Preferably, the S3003 includes:

[0054] Randomly sampled Gaussian noise distribution x T ~N(0,I) is used as the initial noise image, and the noise prediction model ∈ θ The noise in the image is reversely predicted and removed under the guidance of the feature vector c. The face image x0 can be generated by multiple iterations of reverse prediction and noise removal. The formula for reverse noise removal is as follows:

[0055]

[0056] In the formula, β t is the moment hyperparameter, and ε is a standard normal distribution random variable.

[0057] Preferably, the S3004 includes:

[0058] According to the face recognition model, the generated face image is converted into a feature vector through the face recognition model, and the feature similarity is compared with the feature vector of the image in the face registration library to determine the identity. The method of feature similarity comparison is vector inner product, and the formula is as follows:

[0059] a.b i =|a||b i |cosθ

[0060] Where a is the feature vector for generating face images, b i is the feature vector of the face image in the face registration database, which b i The larger the inner product value is, the more likely it is that the generated face belongs to which user.

[0061] Beneficial effects:

[0062] 1. The EEG signals of the present invention are generated by individual brain activities, which are highly individual specific and complex and difficult to be forged by simple copying or imitation. This feature enables the EEG-based identity authentication system to effectively prevent forgery and fraud, and has high security;

[0063] 2. The EEG signal of the present invention is not affected by external environmental factors such as ambient light, temperature, humidity, etc., and can maintain stable performance under different environmental conditions. The EEG signal acquisition device can usually effectively filter out background noise to ensure the quality of the collected signal and has strong anti-interference performance;

[0064] 3. The EEG signal of the present invention has unique brain wave characteristics, including frequency, amplitude and phase, etc. These characteristics are determined by the individual's brain structure and function, so it has a high ability to distinguish individuals. The identity authentication system based on EEG signals can accurately identify individuals and maintain high accuracy even in large-scale crowds;

[0065] 4. The EEG signal of the present invention can be combined with the face image to form an identity authentication system. This system comprehensively utilizes multiple biometric features and can display the recognition results in real time to facilitate subsequent processing, greatly improving the accuracy and robustness of identity authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flow chart of a preferred embodiment of the present invention;

[0067] Figure 2 It is a schematic diagram of a flow chart of training a face generation model according to a preferred embodiment of the present invention;

[0068] Figure 3 Schematic diagram of the network structure of an encoder E according to a preferred embodiment of the present invention;

[0069] Figure 4 It is a schematic diagram of the network structure of a face generation model of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0070] The embodiments of the present invention are described in detail below. The following embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.

[0071] The present invention designs an identity authentication method based on the generation of face images by EEG signals. The technical solution comprises the following steps: Figure 1-4 As shown, specifically including:

[0072] S10: In the registration phase, the user's EEG signal and face image are collected as registration information;

[0073] S20: Model training phase, the face generation model is jointly trained based on the EEG signals and face images in the registration information;

[0074] S30: Verification stage: The user's EEG signal is collected as verification information, input into the trained face generation model to generate a face image, and compared with the user himself to achieve identity authentication.

[0075] Preferably, the S10 includes:

[0076] In the registration stage, the EEG registration signal is collected through an EEG cap based on an EEG signal induction paradigm, wherein the EEG signal induction paradigm includes visual evoked potential, event-related potential and motor imagery;

[0077] During the collection process, the user performs tasks under different paradigms according to the prompts to induce EEG signals, and collects facial images through the camera. A frontal face photo of the user is taken during the process of collecting EEG. The EEG registration signal and the face registration image are bound and submitted to the model for training.

[0078] Specifically, the number of sampling channels collected by the camera is 64, the sampling frequency is 256HZ, the EEG signal is cut into a sample every two seconds, and each EEG sample is bound to a corresponding face image.

[0079] Preferably, the S20 includes the following steps:

[0080] S2001: pre-training feature encoder E using EEG registration signal;

[0081] S2002: Encode the EEG registration signal e into a feature vector c using a pre-trained feature encoder E;

[0082] S2003: training a face generation model using feature vectors and face registration images;

[0083] The S2001 includes:

[0084] The EEG registration signal is passed through the feature encoder E to obtain the feature vector, the feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result, the encoder parameters are updated based on the loss calculated according to the true category label, and the encoder E is obtained after iterative repeated training. The specific formula for encoding the feature encoder E is as follows:

[0085] c=E(e)

[0086] In the formula, e is the EEG registration signal, c is the encoded feature vector;

[0087] The feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result. The formula is as follows:

[0088] z=Wc+b

[0089]

[0090] In the formula, z is the classification prediction result, W is the weight matrix, b is the bias, Softmax(·) is the Softmax function, and z i is the i-th component in z, z j is the jth component in z;

[0091] The loss function used in the training process is the cross entropy loss, and the formula is as follows:

[0092]

[0093] Where n is the number of categories, y is the true category label, and p is the predicted category probability.

[0094] Specifically, the EEG registration signal is passed through the feature encoder E to obtain a 512-dimensional feature vector, the feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result, the encoder parameters are updated by calculating the loss backpropagation based on the true category label, and the encoder E is obtained by iterative repeated training. The specific model used by the feature encoder is a convolutional neural network combined with an attention mechanism architecture, which is currently the best model for feature extraction in identity recognition.

[0095] Preferably, the S2002 includes:

[0096] During the registration process, all EEG registration signals e are encoded into feature vectors c through the pre-trained feature encoder E. All feature vectors of the same user are averaged and bound to the corresponding face registration images for training the face generation model.

[0097] Preferably, the S2003 includes:

[0098] In the registration phase, the feature vector is used to guide the diffusion model to generate face images from random noise data. The generated face images are compared with the real face images to calculate the loss. The loss is back-propagated to calculate the gradient to update the model parameters. The trained face generation model can be obtained by iteratively repeating the update. The formula is as follows:

[0099] The forward diffusion process formula is as follows:

[0100]

[0101] The reverse generation process formula is as follows:

[0102] p(x t-1 |x t )=N(x t-1 ;uθ (x t ,t,c),σ θ (x t ,t,c))

[0103] In the formula, q(x t |x0) is x under the condition that the distribution of x0 is known t The probability distribution function of x t is a Gaussian distributed random variable, is the forward noise mean, (1-α t )I is the forward noise variance, p(x t-1 |x t ) is a known x t Under the condition of distribution x t-1 The probability distribution function, u θ (x t ,t,c) is the reverse denoising mean, σ θ (x t ,t,c) is the reverse denoising variance. The prediction of the reverse denoising mean and variance can be converted into the prediction of noise;

[0104] A neural network is used as a noise prediction model, and the noise prediction model is used to t Predict x t-1 The distribution of , through continuous reverse prediction, the face image can be generated from the noise image. The reverse prediction formula is as follows:

[0105]

[0106] In the formula, β t is the moment hyperparameter, ε is a standard normal distribution random variable;

[0107] The training goal of iterative training is to optimize the variational upper bound of the negative log-likelihood. The loss function formula during iterative training is as follows:

[0108]

[0109] In the formula, Find the numerical expectation under the premise of knowing t,x0,ε,ε θ (x t ,t,c) is the estimated noise value, ε is the standard normal distribution random noise, x represents the image in the denoising step, x has different forms at different time steps, t represents the time step used to provide step information for the model, and the noise prediction model ε is obtained by repeatedly iterative training and updating the model parameters. θ , using the noise prediction model ε θ Generate face images.

[0110] Specifically, the process of training the face generation model to generate faces includes the forward diffusion process and the reverse generation process. The forward diffusion process is to gradually add noise from the real data to make the data closer and closer to a simple distribution. This process can be understood as gradually destroying the data structure to make it disordered. The reverse generation process is to gradually remove the noise from the simple noise distribution, restore the data structure and make the data orderly. Since the value of the reverse generation process cannot be directly calculated, it can only be estimated using a neural network, so the mean u of the reverse process is θ (x t , t, c) and variance σ θ (x t , t, c) are two neural networks, input model data x t , step length t, conditional vector c to estimate x t-1 The mean and variance of .

[0111] The diffusion model theory is derived through a series of optimizations to obtain x t-1 The variance is a fixed value, and we only need to use a neural network to estimate the mean, which can be expressed as Indicates that the variance is σ t , we can use the mean and variance to get from x t Indicates x t-1 The distribution of the reverse denoising mean u θ and variance σ θ The prediction can be transformed into the prediction of noise ∈ θ Prediction

[0112] Preferably, the S30 includes the following steps:

[0113] S3001: Collecting the user's EEG verification signal;

[0114] S3002: Encode the EEG registration signal into a feature vector using a feature encoder;

[0115] S3003: using the feature vector to guide the face generation model to generate a face image;

[0116] S3004: Compare the generated face images to achieve identity authentication.

[0117] Preferably, the S3001 includes:

[0118] In the verification stage, a paradigm is used to induce EEG signals, and the induced EEG signals are collected as EEG verification signals. The EEG verification signals and the EEG signals in the registration stage undergo the same preprocessing process for subsequent generation of facial images.

[0119] Preferably, the S3002 includes:

[0120] The collected EEG verification signal is encoded into a feature vector using the pre-trained feature encoder E. The specific encoding formula is as follows:

[0121] c=E(e)

[0122] Where e is the EEG verification signal and c is the encoded feature vector.

[0123] Preferably, the S3003 includes:

[0124] Randomly sampled Gaussian noise distribution x T ~N(0,I) is used as the initial noise image, and the noise prediction model ∈ θ The noise in the image is reversely predicted and removed under the guidance of the feature vector c. The face image x0 can be generated by multiple iterations of reverse prediction and noise removal. The formula for reverse noise removal is as follows:

[0125]

[0126] In the formula, β t is the moment hyperparameter, and ε is a standard normal distribution random variable.

[0127] Preferably, the S3004 includes:

[0128] According to the face recognition model, the generated face image is converted into a feature vector through the face recognition model, and the feature similarity is compared with the feature vector of the image in the face registration library to determine the identity. The method of feature similarity comparison is vector inner product, and the formula is as follows:

[0129] a.b i =|a||b i |cosθ

[0130] Where a is the feature vector for generating face images, b i is the feature vector of the face image in the face registration database, which b i The larger the inner product value is, the more likely it is that the generated face belongs to which user.

[0131] Specifically, after comparing with the face recognition model, it also includes counting the accuracy of generating face images. The image generated by EEG has a label corresponding to the original EEG. The face recognition model is used to extract features to determine the label of the generated image. The performance of the generated model can be evaluated by counting the number of correct labels of the image generated by the model. The evaluation indicator we use is TOP accuracy, which is a common indicator for evaluating model performance. InsightFace is a deep learning tool library for face analysis. Based on advanced deep learning algorithms, it can provide efficient and accurate face detection, alignment, recognition and feature extraction functions. The process of extracting features first performs face detection to find the face position, and then extracts features through the face recognition model. The face detection model used by the Insightface tool library is RetinaFace and the face recognition model is ArcFace. We use the default model for feature extraction and compare feature similarity. The final statistical generation accuracy is as follows:

[0132] ACC TOP-1 TOP-3 TOP-5 EEG 97.7％ 99.23％ 99.23％

[0133] In addition, the S3004 also includes subjective comparison. The comparison method for generating a facial image mainly includes subjective comparison and objective comparison. The above-mentioned comparison based on the face recognition model is an objective comparison, while the subjective comparison is generally completed based on the subjective judgment of the staff. The user's EEG signal is collected, and a facial image is generated through a face generation model. The staff compares the generated facial image with the person himself to determine whether the identity meets the authentication requirements.

[0134] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. An identity authentication method for generating a face image based on an electroencephalogram signal, characterized in that: The following steps are involved: S10: In the registration phase, the user's EEG signal and face image are collected as registration information; S20: Model training phase, the face generation model is jointly trained based on the EEG signals and face images in the registration information; S30: Verification stage: The user's EEG signal is collected as verification information, input into the trained face generation model to generate a face image, and compared with the user himself to achieve identity authentication.

2. The identity authentication method for generating a face image based on an EEG signal according to claim 1, characterized in that: The S10 includes: In the registration stage, the EEG registration signal is collected through an EEG cap based on an EEG signal induction paradigm, wherein the EEG signal induction paradigm includes visual evoked potential, event-related potential and motor imagery; During the collection process, the user performs tasks under different paradigms according to the prompts to induce EEG signals, and the face image is collected through the camera. A frontal face photo of the user is taken during the process of collecting EEG. The EEG registration signal and the face registration image are bound and submitted to the model for training.

3. The identity authentication method for generating a face image based on an EEG signal according to claim 1, characterized in that: The S20 includes the following steps: S2001: pre-training feature encoder E using EEG registration signal; S2002: Encode the EEG registration signal e into a feature vector c using a pre-trained feature encoder E; S2003: training a face generation model using feature vectors and face registration images; The S2001 includes: The EEG registration signal is passed through the feature encoder E to obtain the feature vector, the feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result, the encoder parameters are updated by backpropagation based on the loss calculated according to the true category label, and the encoder E is obtained after iterative repeated training. The specific formula for encoding the feature encoder E is as follows: c=E(e) In the formula, e is the EEG registration signal, c is the encoded feature vector; The feature vector is reduced in dimension and passed through the Softmax function to obtain the classification prediction result. The formula is as follows: z=Wc+b In the formula, z is the classification prediction result, W is the weight matrix, b is the bias, Softmax(·) is the Softmax function, and z i is the i-th component in z, z j is the jth component in z; The loss function used in the training process is the cross entropy loss, and the formula is as follows: Where n is the number of categories, y is the true category label, and p is the predicted category probability.

4. The identity authentication method for generating a face image based on an EEG signal according to claim 3, characterized in that: The S2002 includes: During the registration process, all EEG registration signals e are encoded into feature vectors c through the pre-trained feature encoder E. All feature vectors of the same user are averaged and bound to the corresponding face registration images for training the face generation model.

5. The identity authentication method for generating a face image based on an EEG signal according to claim 3, characterized in that: The S2003 includes: In the registration phase, the feature vector is used to guide the diffusion model to generate a face image from random noise data. The generated face image is compared with the real face image to calculate the loss. The loss is back-propagated to calculate the gradient to update the model parameters. The training face generation model is obtained by iteratively repeating the update. The formula is as follows: The forward diffusion process formula is as follows: The reverse generation process formula is as follows: p(x t-1 |x t )=N(x t-1 ;u θ (x t ,t,c),σ θ (x t ,t,c)) In the formula, q(x t |x0) is x under the condition that the distribution of x0 is known t The probability distribution function of x t is a Gaussian distributed random variable, is the forward noise mean, (1-α t )I is the forward noise variance, p(x t-1 |x t ) is a known x t Under the condition of distribution x t-1 The probability distribution function, u θ (x t , t, c) is the reverse noise reduction mean, σ θ (x t , t, c) is the reverse denoising variance, and the prediction of the reverse denoising mean and variance can be converted into the prediction of the noise; A neural network is used as a noise prediction model, and the noise prediction model is used to t Predict x t-1 The distribution of , through continuous reverse prediction, the face image can be generated from the noise image. The reverse prediction formula is as follows: In the formula, β t is the moment hyperparameter, ε is a standard normal distribution random variable; The training goal of iterative training is to optimize the variational upper bound of the negative log-likelihood. The loss function formula during iterative training is as follows: In the formula, Find the numerical expectation given t, x0, ε, θ (x t , t, c) is the estimated noise value, ε is the standard normal distribution random noise, x represents the image in the denoising step, x has different forms at different time steps, t represents the time step used to provide step information for the model, and the noise prediction model ε is obtained by repeatedly iterative training and updating the model parameters. θ , using the noise prediction model ε θ Generate face images.

6. The identity authentication method for generating a face image based on an EEG signal according to claim 1, characterized in that: The S30 includes the following steps: S3001: Collecting the user's EEG verification signal; S3002: Encode the EEG registration signal into a feature vector using a feature encoder; S3003: using the feature vector to guide the face generation model to generate a face image; S3004: Compare the generated face images to achieve identity authentication.

7. The identity authentication method for generating a face image based on an EEG signal according to claim 6, characterized in that: The S3001 includes: In the verification stage, a paradigm is used to induce EEG signals, and the induced EEG signals are collected as EEG verification signals. The EEG verification signals and the EEG signals in the registration stage undergo the same preprocessing process for subsequent generation of facial images.

8. The identity authentication method for generating a face image based on an EEG signal according to claim 6, characterized in that: The S3002 includes: The collected EEG verification signal is encoded into a feature vector using the pre-trained feature encoder E. The specific encoding formula is as follows: c=E(e) Where e is the EEG verification signal and c is the encoded feature vector.

9. The identity authentication method for generating a face image based on an EEG signal according to claim 6, characterized in that: The S3003 includes: Randomly sampled Gaussian noise distribution x T ~N(0,I) is used as the initial noise image, and the noise prediction model ∈ θ The noise in the image is reversely predicted and removed under the guidance of the feature vector c. The face image x0 can be generated by multiple iterations of reverse prediction and noise removal. The formula for reverse noise removal is as follows: In the formula, β t is the moment hyperparameter, and ε is a standard normal distribution random variable.

10. The identity authentication method for generating a face image based on an EEG signal according to claim 6, characterized in that: The S3004 includes: According to the face recognition model, the generated face image is converted into a feature vector through the face recognition model, and the feature similarity is compared with the feature vector of the image in the face registration library to determine the identity. The method of feature similarity comparison is the vector inner product, and the formula is as follows: a·b i =|a|||b i |cosθ Where a is the feature vector for generating face images, b i is the feature vector of the face image in the face registration database, which b i The larger the inner product value is, the more likely it is that the generated face belongs to which user.

Images