An identity verification method based on edge information feature comparison of short-interval face images

Abstract

The present invention belongs to the field of face recognition technology, and specifically relates to an identity authentication method based on short-interval face image edge information feature comparison, comprising: obtaining two images to be identified within a time interval; performing background segmentation processing on the two images to be identified; removing facial information of the two face images respectively to obtain a first face edge image and a second face edge image; using a face edge information feature extraction model to extract edge features of the first face edge image and the second face edge image to obtain a first face edge information feature and a second face edge information feature; authenticating a user based on the first face edge information feature and the second face edge information feature; the auxiliary identity authentication method proposed in the present invention improves the identity discrimination degree of the identity authentication system for persons to be identified with similar faces, and solves the difficult problem of difficulty in identifying similar users in existing face recognition technical methods.

Description

Technical Field

[0001] The present invention belongs to the technical field of face recognition, and in particular relates to an identity authentication method based on short-interval face image edge information feature comparison. Background Art

[0002] Facial recognition is one of the most widely used biometric technologies, and improving its accuracy has long been a focus of research in the field of computer vision. However, despite extensive research and technological innovation, facial recognition accuracy remains elusive. This is primarily due to the numerous factors involved in practical applications, such as lighting, posture, occlusion, expression, deformation, photos, and videos, all of which can affect the accuracy of facial recognition results. Furthermore, facial recognition technology faces several challenges, such as acquiring and processing facial videos, recognizing facial expressions and postures, and the non-uniqueness of facial features. These difficulties and challenges increase the difficulty of facial recognition and motivate researchers to continuously explore and innovate. Despite this, facial recognition technology still holds broad application prospects in areas such as identity authentication, secure payment, and human-computer interaction.

[0003] Taking airports, train stations, and bus stations as examples, facial recognition technology has been widely used for identity verification at entry points and security checkpoint verification. The typical passenger security checkpoint verification process involves first completing identity verification at entry points and comparing their ID with their ticket, then entering the waiting area to queue for security check, then passing through the security checkpoint verification channel to complete security check, and finally entering the security checkpoint area to wait for transportation. During the identity verification process, all passengers entering the waiting area undergo identity verification. Facial recognition technology primarily uses a 1:1 comparison of facial features between their ID photo and the entry-level photo to verify their identity. During the security checkpoint verification process, all passengers passing through the security checkpoint undergo a secondary verification, primarily by comparing their ID photo with the entry-level photo. The application of facial recognition technology in identity verification at entry points and security checkpoint verification can effectively improve the security and accuracy of passenger flow management during security checks, preventing identity forgery and theft.

[0004] However, in actual application, due to the large passenger flow, it is sometimes inevitable that two (or more) passengers with similar facial features but completely different identities will enter the waiting area through identity verification of ID card and ticket. However, when facial recognition is performed at the security check channel, they will be confused with each other, resulting in ambiguous identities that are difficult to distinguish. Summary of the Invention

[0005] To solve the above problems in the prior art, the present invention proposes an identity authentication method based on short-interval facial image edge information feature comparison, the method comprising:

[0006] S1: Set the time interval; use the image acquisition device to capture the face of the person to be authenticated according to the time interval to obtain two images to be identified;

[0007] S2: Use the trained semantic segmentation model to perform background segmentation on the two images to be recognized, and obtain two face images;

[0008] S3: removing facial information of the two face images respectively to obtain a first face edge image and a second face edge image;

[0009] S4: Using the pre-trained face edge information feature extraction model to extract edge features from the first face edge image and the second face edge image, to obtain a first face edge information feature and a second face edge information feature;

[0010] S5: Authenticate the user based on the first face edge information feature and the second face edge information feature.

[0011] Furthermore, the semantic segmentation model processes the image to be identified, including: obtaining a background image, inputting the background image and the image to be identified into the trained semantic segmentation model to obtain a face image; wherein the training of the semantic segmentation model includes: obtaining training samples, and enhancing the images in the training samples; wherein the training samples are the background image and the image to be identified acquired by the image acquisition device; performing multi-scale feature extraction on the enhanced image to be identified and the background image, respectively, to obtain background feature maps of different scales and image feature maps of different scales; fusing all background feature maps of different scales to obtain a fused background feature map; fusing all image feature maps of different scales to obtain a fused feature map; setting a pixel difference threshold, comparing the pixel difference between the fused background feature map and the fused feature map, setting the pixel values ​​less than the pixel difference threshold to zero, to obtain a face image; calculating the model loss function based on the face image, continuously adjusting the model parameters, and completing the model training when the loss function converges.

[0012] Furthermore, setting the pixel difference threshold includes: obtaining image sample data, and inputting the image sample data into the semantic segmentation model to obtain the fused background feature map and fused feature map corresponding to all sample images; using the optimized genetic algorithm to iteratively process the fused background feature map and the fused feature map to obtain the optimal pixel difference, and using the difference as the pixel difference threshold.

[0013] Furthermore, the optimized genetic algorithm is used to iteratively process the fused background feature map and the fused feature map, including:

[0014] Step 1: Initialize the population number, where the population number is all the fused background feature maps and fused feature maps; set relevant parameters, including population size, selection probability, crossover probability, mutation probability, and number of iterations;

[0015] Step 2: Set the similarity range value, and calculate the similarity between the fused background feature map and the fused feature map respectively; compare all calculated similarity values ​​with the set range value, fuse all fused background feature maps within the similarity range value to obtain a new fused background feature map, and fuse all feature maps within the similarity range value to obtain a new fused feature map; use the new fused background feature map and the new fused feature map as new populations;

[0016] Step 3: Perform selection, crossover, and mutation operations on the new population to obtain a new generation of population; the number of iterations increases by 1;

[0017] Step 4: Perform genetic screening on the new generation population to obtain individuals with optimal genes;

[0018] Step 5: Calculate the individual fitness function value of the optimal gene, which is the pixel difference between the new fused background feature map and the new fused feature map;

[0019] Step 6: Use the attenuation strategy to update the similarity range value;

[0020] Step 7: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the individual fitness function value of the optimal gene in the new generation population. This value is the pixel difference threshold. Otherwise, return to step 2.

[0021] Furthermore, removing facial information from the two face images includes: performing face detection on the two face images and marking the largest face frame; fixing the center point of the largest face frame, scaling down the largest face frame so that the face frame only contains information about facial features, and using the face frame as a corrected face frame; and setting all pixel values ​​in the corrected face frame to zero to obtain a face edge image.

[0022] Furthermore, the face edge information feature extraction model adopts a convolutional neural network model, including a convolutional layer, a pooling layer, a batch normalization layer, a fully connected layer, and a dropout layer; the convolutional layer contains multiple convolution kernels, each of which detects different features in the image; the pooling layer is used to reduce the size of the feature map; the batch normalization layer is used to accelerate the training process and improve the robustness of the model; the fully connected layer is used to map the output of the convolutional layer to a specific category or target; the dropout layer is used to prevent the model from overfitting.

[0023] Preferably, authenticating the user based on the first facial edge information feature and the second facial edge information feature includes: inputting the facial image into the facial recognition model for facial feature matching, and when the recognition result is clear and unique, directly using the recognition result as the identity authentication result; if the recognition result is ambiguous or fails, extracting all confused persons to be inspected whose facial features are similar to the current person to be inspected, and comparing the second facial edge information feature of the current person to be inspected with the first facial edge information feature of all confused persons to be inspected one by one; calculating the facial edge information feature comparison score of the current person to be inspected and each confused person to be inspected, and screening out the confused person corresponding to the highest facial edge information feature comparison score to determine the identity information of the current person to be inspected.

[0024] Furthermore, the comparison score formula between the first face edge information feature and the second face edge information feature adopts the cosine similarity formula, which is expressed as follows:

[0025] similarity_score=(A·B) / (||A||*||B||)

[0026] Where A·B is the inner product of vector A and vector B, and ||A|| and ||B|| are the moduli of vector A and vector B, respectively.

[0027] Beneficial effects of the present invention:

[0028] The auxiliary identity authentication method proposed in the present invention cleverly utilizes facial edge information such as face shape, hairstyle, clothing, accessories, etc. that has short-term specificity but has not been fully utilized, and can provide additional and effective judgment basis for the identity authentication system; the auxiliary identity authentication method proposed in the present invention improves the identity authentication system's ability to distinguish the identities of people with similar faces to be inspected, to a large extent solves the difficult problems in existing face recognition technology methods, and further improves the recognition accuracy of face recognition algorithms; the auxiliary identity authentication method proposed in the present invention has low deployment difficulty and high practical value, which helps to improve the overall security and accuracy of the identity authentication system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is the overall flow chart of the present invention;

[0030] Figure 2 This is a schematic diagram of the airport security inspection channel area of ​​the present invention. DETAILED DESCRIPTION

[0031] 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.

[0032] An identity verification method based on short-interval face image edge information feature comparison, such as Figure 1 As shown, the method includes:

[0033] S1: Set the time interval; use the image acquisition device to capture the face of the person to be authenticated according to the time interval to obtain two images to be identified;

[0034] S2: Use the trained semantic segmentation model to perform background segmentation on the two images to be recognized, and obtain two face images;

[0035] S3: removing facial information of the two face images respectively to obtain a first face edge image and a second face edge image;

[0036] S4: Using the pre-trained face edge information feature extraction model to extract edge features from the first face edge image and the second face edge image, to obtain a first face edge information feature and a second face edge information feature;

[0037] S5: Authenticate the user based on the first face edge information feature and the second face edge information feature.

[0038] In this embodiment, a process of processing an image to be recognized by a semantic segmentation model is disclosed, including: obtaining a background image, inputting the background image and the image to be recognized into a trained semantic segmentation model to obtain a face image; wherein training the semantic segmentation model includes: obtaining training samples, and enhancing the images in the training samples; wherein the training samples are the background image and the image to be recognized acquired by an image acquisition device; performing multi-scale feature extraction on the enhanced image to be recognized and the background image respectively to obtain background feature maps of different scales and image feature maps of different scales; fusing all background feature maps of different scales to obtain a fused background feature map; fusing all image feature maps of different scales to obtain a fused feature map; setting a pixel difference threshold, comparing the pixel difference between the fused background feature map and the fused feature map, setting pixel values ​​less than the pixel difference threshold to zero, to obtain a face image; calculating a model loss function based on the face image, continuously adjusting model parameters, and completing model training when the loss function converges.

[0039] Setting the pixel difference threshold includes: obtaining image sample data and inputting the image sample data into the semantic segmentation model to obtain the fused background feature map and fused feature map corresponding to all sample images; using the optimized genetic algorithm to iteratively process the fused background feature map and the fused feature map to obtain the optimal pixel difference, and using the difference as the pixel difference threshold.

[0040] The optimized genetic algorithm is used to iteratively process the fused background feature map and the fused feature map, including:

[0041] Step 1: Initialize the population number, where the population number is all the fused background feature maps and fused feature maps; set relevant parameters, including population size, selection probability, crossover probability, mutation probability, and number of iterations;

[0042] Step 2: Set the similarity range value, and calculate the similarity between the fused background feature map and the fused feature map respectively; compare all calculated similarity values ​​with the set range value, fuse all fused background feature maps within the similarity range value to obtain a new fused background feature map, and fuse all feature maps within the similarity range value to obtain a new fused feature map; use the new fused background feature map and the new fused feature map as new populations;

[0043] Step 3: Perform selection, crossover, and mutation operations on the new population to obtain a new generation of population; the number of iterations increases by 1;

[0044] Step 4: Perform genetic screening on the new generation population to obtain individuals with optimal genes;

[0045] Step 5: Calculate the individual fitness function value of the optimal gene, which is the pixel difference between the new fused background feature map and the new fused feature map;

[0046] Step 6: Use the attenuation strategy to update the similarity range value;

[0047] Step 7: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the individual fitness function value of the optimal gene in the new generation population. This value is the pixel difference threshold. Otherwise, return to step 2.

[0048] Decay strategies include:

[0049]

[0050] Among them, ω t represents the decay weight, T max Indicates the maximum similarity range value under the current number of iterations, T min It represents the minimum similarity range value under the current number of iterations, t represents the current number of iterations, and ΔT represents the cycle period.

[0051] Removing facial information from two face images includes: performing face detection on the two face images and marking a maximum face frame; fixing the center point of the maximum face frame, scaling down the maximum face frame so that the face frame only contains information about facial features, and using the face frame as a corrected face frame; and setting all pixel values ​​in the corrected face frame to zero to obtain a face edge image.

[0052] Authentication of a user based on the first and second facial edge information features includes: inputting the first and second facial edge information features into a facial recognition model, respectively; when the facial recognition model outputs a result, the result is used as the final recognition result, and authentication is performed; when the facial recognition result is not a result, executing a recognition strategy based on the comparison of facial edge information features to complete user authentication. The recognition strategy based on the comparison of facial edge information features includes: calculating a comparison score between the first and second facial edge information features; weightedly fusing the facial edge information feature comparison score of the current person to be inspected and all other persons to be inspected with the facial feature comparison score, to obtain a facial comprehensive feature comparison score between the current person to be inspected and all other persons to be inspected; and obtaining an identification result based on the comparison score.

[0053] The formula for comparing the first face edge information feature and the second face edge information feature uses the cosine similarity formula, which is expressed as:

[0054] similarity_score=(A·B) / (||A||*||B||)

[0055] Where A·B is the inner product of vector A and vector B, and ||A|| and ||B|| are the moduli of vector A and vector B, respectively.

[0056] In this embodiment, taking the airport as an example, the airport security inspection channel area is as follows: Figure 2 As shown. An identity verification method based on short-interval facial image edge information feature comparison uses an algorithm to collect two sets of on-site facial images of the same group of people to be inspected within a short time interval, extract facial edge information from the images for feature matching, and use the obtained results as an auxiliary reference for other identity verification systems such as facial recognition. In particular, when faced with situations where facial similarity is high and relying solely on facial recognition makes it difficult to distinguish the identities of the people to be inspected, the integrated identity verification system can make more comprehensive and accurate judgments. Specifically, it involves collecting two sets of short-interval facial images, removing image background and facial information, extracting facial edge information features, comparing facial edge information features, and assisting facial recognition identity verification.

[0057] Short-interval facial image acquisition: First, the first facial image of all persons to be inspected is collected, which is recorded as the "first on-site photo". There can be multiple imaging devices (cameras) used for image acquisition (such as corresponding to multiple verification entrances), but the position and orientation of each imaging device are relatively fixed. The faces in the picture are detected in real time, and through appropriate guidance, the frontal facial image of the person to be inspected located in the center area of ​​the predetermined picture is collected; then after a short time interval, the second facial image of the person to be inspected is collected, which is recorded as the "second on-site photo". There can be multiple imaging devices (cameras) used for image acquisition (such as corresponding to multiple security inspection channels), but the position and orientation of each imaging device are relatively fixed. The faces in the picture are detected in real time, and through appropriate guidance, the frontal facial image of the person to be inspected located in the center area of ​​the predetermined picture is collected.

[0058] The process of removing the background of a face image includes:

[0059] S2.1. The imaging device used to capture the "first scene photo" in step S1.1, whenever the device is not currently detecting a person (when the device is just being used or has not detected a large face for a period of time), captures a scene photo as the background photo of the current environment, recorded as the "first background photo";

[0060] S2.2. The imaging device used to capture the "second scene photo" in step S1.2, whenever the device is not currently detecting a person (when the device is just being used or has not detected a large face for a period of time), captures a scene photo as the background photo of the current environment, recorded as the "second background photo";

[0061] S2.3, based on the "first scene photo" and "first background photo" collected in steps S1.1 and S2.1, and the "second scene photo" and "second background photo" collected in steps S1.2 and S2.2, or other available datasets, train a corresponding semantic segmentation model to separate the foreground and background of the person image;

[0062] S2.4. Generally, the trained semantic segmentation model is used (or more simply, only the image processing method is used to directly compare the pixel differences between the two sets of on-site photos and the corresponding background photos, and the pixel values ​​with smaller difference changes are set to zero), and the foreground of the people in the "first on-site photo" and the "second on-site photo" are respectively extracted (all foreground pixel values ​​belonging to the person to be inspected in the image are retained, and the remaining background pixel values ​​are all set to zero), and recorded as the "first on-site photo with background removed" and the "second on-site photo with background removed", respectively.

[0063] Facial information removal includes:

[0064] S3.1. Perform face detection on the first and second background-removed photos, using either a separate training model or other general face detection model based on the on-site data. Mark the largest face frame in the image (corresponding to the face of the person currently being inspected).

[0065] S3.2. Based on the largest face frame detected in the first background-removed photo and the second background-removed photo in step S3.1, the center of the largest face frame is kept unchanged and the largest face frame is appropriately reduced in proportion to include only the main facial information such as eyebrows, eyes, nose, and mouth, thereby obtaining the corresponding corrected face frame;

[0066] S3.3. According to the corrected face frames obtained in the "first background-removed scene photo" and the "second background-removed scene photo" in step S3.2, the corresponding facial information is removed (all pixel values ​​within the corrected face frame are set to zero, and the pixels outside the corrected face frame remain unchanged), and they are recorded as the "first face edge image" and the "second face edge image" respectively.

[0067] Face edge information feature extraction includes:

[0068] S4.1. Before deploying the application, collect a batch of on-site photos of different people to be inspected, pre-process the images, and use identity labels as data samples to train the facial edge information feature extraction model;

[0069] S4.2. Using the trained facial edge information feature extraction model, extract facial edge information features from the first facial edge image and the second facial edge image, denoting these features as the first facial edge information feature and the second facial edge information feature, respectively.

[0070] S4.3. When auxiliary comparison is required, the cosine similarity or Euclidean distance between the "first facial edge information feature" and the "second facial edge information feature" is calculated as a comparison score to provide additional reference.

[0071] Assisted facial recognition identity verification includes: in most cases where identification is still possible, relying solely on the facial feature comparison results of the facial recognition system to make a direct judgment on the identity information of the person to be inspected. In cases where the facial information of several persons to be inspected is relatively similar, causing the facial recognition algorithm to produce ambiguous comparison results (i.e., the facial features of the current person to be inspected at the second collection are confused with the facial features of more than one person to be inspected at the first collection, and the calculated comparison scores are all above the judgment threshold and relatively close), and it is impossible to reach a definitive judgment, all confused persons to be inspected whose facial features are similar to those of the current person to be inspected are extracted, and the "second facial edge information features" of the current person to be inspected are compared one by one with the "first facial edge information features" of all confused persons to be inspected. The facial edge information feature comparison scores of the current person to be inspected and each confused person to be inspected are calculated, and the identity information of the current person to be inspected is judged based on the comparison results (i.e., the identity of the current person to be inspected is judged to be the person with the highest facial edge information feature comparison score among all confused persons to be inspected). If confusion occurs after comparing the facial recognition features and the facial edge information features in sequence, and it is still difficult to reach an accurate judgment, the system will issue a prompt and the on-site staff will conduct manual verification.

[0072] Another implementation method can make a comprehensive judgment by considering the comparison results of facial recognition facial features and facial edge information features at the time of each identity verification. Specifically, after the current person to be inspected completes the second image acquisition, the "second facial edge information features" of the current person to be inspected are extracted in sequence, and the comparison scores of the "second facial edge information features" of the current person to be inspected and the "first facial edge information features" of all persons to be inspected are calculated respectively; the facial edge information feature comparison scores of the current person to be inspected and all persons to be inspected are weighted and fused with the facial feature comparison scores according to a certain proportional coefficient to obtain the facial comprehensive feature comparison scores of the current person to be inspected and all persons to be inspected; if confusion occurs when comparing the facial comprehensive features and it is difficult to make an accurate judgment, the system will issue a prompt and the on-site staff will conduct manual verification.

[0073] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation plans of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An identity authentication method based on short-interval facial image edge information feature comparison, characterized in that: include: S1: Over a period of time, an image acquisition device is used to capture the face of the person to be authenticated, and two images to be identified are obtained; S2: Use the trained semantic segmentation model to perform background segmentation on the two images to be identified, and obtain two face images; S3: removing facial information of the two face images respectively to obtain a first face edge image and a second face edge image; S4: Using the pre-trained face edge information feature extraction model to extract edge features from the first face edge image and the second face edge image, to obtain a first face edge information feature and a second face edge information feature; S5: Authenticate the user based on the first facial edge information feature and the second facial edge information feature; specifically including: inputting the facial image into the facial recognition model for facial feature matching, and when the recognition result is clear and unique, directly using the recognition result as the identity authentication result; if the recognition result is unclear or fails, extracting all confused persons to be inspected whose facial features are similar to the current person to be inspected, and comparing the second facial edge information feature of the current person to be inspected with the first facial edge information feature of all confused persons to be inspected one by one; calculating the facial edge information feature comparison score of the current person to be inspected and each confused person to be inspected, and screening out the confused person corresponding to the highest facial edge information feature comparison score to determine the identity information of the current person to be inspected.

2. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 1, characterized in that: The semantic segmentation model processes the image to be identified, including: obtaining a background image, inputting the background image and the image to be identified into the trained semantic segmentation model, and obtaining a face image; wherein the semantic segmentation model is trained, including: obtaining training samples, and performing image enhancement processing on the training samples; wherein the training samples are the background image and the image to be identified acquired by the image acquisition device; performing multi-scale feature extraction on the enhanced image to be identified and the background image, respectively, to obtain background feature maps of different scales and image feature maps of different scales; fusing all background feature maps of different scales to obtain a fused background feature map; fusing all image feature maps of different scales to obtain a fused feature map; setting a pixel difference threshold, performing pixel difference comparison between the fused background feature map and the fused feature map, setting pixel values ​​less than the pixel difference threshold to zero, and obtaining a face image; calculating the model loss function based on the face image, continuously adjusting the model parameters, and completing the model training when the loss function converges.

3. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 2, characterized in that: Setting the pixel difference threshold includes: obtaining image sample data and inputting the image sample data into the semantic segmentation model to obtain the fused background feature map and fused feature map corresponding to all sample images; using the optimized genetic algorithm to iteratively process the fused background feature map and the fused feature map to obtain the optimal pixel difference value, and using the optimal pixel difference value as the pixel difference threshold.

4. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 3, characterized in that: The optimized genetic algorithm is used to iteratively process the fused background feature map and the fused feature map, including: Step 1: Initialize the population number, where the population number is all the fused background feature maps and fused feature maps; set relevant parameters, including population size, selection probability, crossover probability, mutation probability, and number of iterations; Step 2: Set the similarity range value, and calculate the similarity between the fused background feature map and the fused feature map respectively; compare all calculated similarity values ​​with the set range value, fuse all fused background feature maps within the similarity range value to obtain a new fused background feature map, and fuse all fused feature maps within the similarity range value to obtain a new fused feature map; use the new fused background feature map and the new fused feature map as new populations; Step 3: Perform selection, crossover, and mutation operations on the new population to obtain a new generation of population; the number of iterations increases by 1; Step 4: Perform genetic screening on the new generation population to obtain individuals with optimal genes; Step 5: Calculate the individual fitness function value of the optimal gene, which is the pixel difference between the new fused background feature map and the new fused feature map; Step 6: Use the attenuation strategy to update the similarity range value; Step 7: Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the individual fitness function value of the optimal gene in the new generation population. This value is the pixel difference threshold. Otherwise, return to step 2.

5. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 4, characterized in that: Decay strategies include: Among them, ω t represents the decay weight, T max Indicates the maximum similarity range value under the current number of iterations, T min It represents the minimum similarity range value under the current number of iterations, t represents the current number of iterations, and ΔT represents the cycle period.

6. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 1, characterized in that: Removing facial information from two face images includes: performing face detection on the two face images and marking a maximum face frame; fixing the center point of the maximum face frame, scaling down the maximum face frame so that the face frame only contains information about facial features, and using the face frame as a corrected face frame; and setting all pixel values ​​in the corrected face frame to zero to obtain a face edge image.

7. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 1, characterized in that: The face edge information feature extraction model uses a convolutional neural network model, including convolutional layers, pooling layers, batch normalization layers, fully connected layers, and dropout layers. The convolutional layer contains multiple convolution kernels, each of which detects different features in the image. The pooling layer is used to reduce the size of the feature map. The batch normalization layer is used to accelerate the training process and improve the robustness of the model. The fully connected layer is used to map the output of the convolutional layer to a specific category or target. The dropout layer is used to prevent the model from overfitting.

8. The identity authentication method based on short-interval facial image edge information feature comparison according to claim 1, characterized in that: The formula for comparing the first face edge information feature and the second face edge information feature uses the cosine similarity formula, which is expressed as: similarity_score=(A·B) / (||A||*||B||) Where A·B is the inner product of vector A and vector B, where vector A is a vector composed of the edge information features of the first face, vector B is a vector composed of the edge information features of the second face, and ||A|| and ||B|| are the moduli of vector A and vector B, respectively.

Images