Face recognition method and system based on cross-age features

Abstract

The invention discloses a face recognition method and system based on cross-age features, belongs to the technical field of face recognition, and is used for solving the problems that traditional face recognition is slow in old people who need to see a doctor through face recognition, slow recognition and recognition errors are easily caused for people with large age spans, and the face recognition efficiency is high. And the medical seeing efficiency of the user is influenced. The method comprises the following steps: performing marking processing on cortical layer texture features and variable epidermal layer features on an acquired initial facial image, and determining a facial cortex feature image; carrying out dynamic stripping processing on variable epidermal layer features in the facial cortex feature image, and carrying out physical optical constraint on dermal layer texture features; carrying out extraction processing on facial feature points under related topological structures and shapes on the optimized facial image; performing age attribute editing diffusion processing on the optimized face image; and performing feature dimension reduction processing on the face key features, and determining social security identity information of the medical seeking personnel.

Description

Technical Field

[0001] This application relates to the field of face recognition, and particularly to a face recognition method and system based on cross-age features. Background Art

[0002] With the increasingly rapid development of medical automation, seeing a doctor and buying medicine have become more convenient. By using technologies such as face recognition, the entire process can be completed, such as registering, picking up medicine, and settlement. Even without an ID card or social security card, "seeing a doctor by face" can still be achieved.

[0003] Among the group of people seeking medical treatment, a large proportion are the elderly. For complex medical treatment processes, more procedures may be required. At the same time, the elderly are more likely to forget to carry information such as social security cards or ID cards. In some emergency situations, such as sudden illnesses, they often rush to see a doctor and neglect to carry valid documents. Therefore, "seeing a doctor by face" is an effective way to solve this pain point, which can greatly reduce the medical treatment process and the use of documents. It only requires the elderly to verify their identity information by face.

[0004] However, the identity information previously entered by the elderly is often relatively old. The social security or identity information is often the face information from a long time ago, which has a relatively large difference from the face features at the current medical treatment time point. When performing face recognition on the elderly, it is easy to cause face recognition errors or failure to recognize face information. Traditional face recognition algorithms are slow and may require multiple attempts to correctly recognize face information, which easily leads to slow medical treatment or delays in registration time. That is, traditional face recognition relies relatively heavily on static texture features and has certain limitations. The facial feature changes brought about by age changes in cross-age recognition will affect the accurate face recognition efficiency of current users to a certain extent, and then affect the medical treatment efficiency of users. Summary of the Invention

[0005] The embodiments of this application provide a face recognition method and system based on cross-age features to solve the following technical problems: For the elderly population who need to see a doctor by face, traditional face recognition is relatively slow, and it is easy to cause slow recognition and recognition errors for people with a large age span, affecting the medical treatment efficiency of users.

[0006] The embodiments of this application adopt the following technical solutions:

[0007] On the one hand, an embodiment of the present application provides a face recognition method based on cross-age features, including: performing marking processing on the initial facial image of the medical treatment personnel collected for dermal layer texture features and variable epidermal layer features to determine a facial cortex feature image; performing dynamic peeling processing on the variable epidermal layer features in the facial cortex feature image and performing physical optical constraints on the dermal layer texture features to obtain an optimized facial image; extracting facial feature points related to topological structure and shape from the optimized facial image through a multi-scale temporal deformation model to obtain facial deformation features; performing diffusion processing on age attribute editing of the optimized facial image to determine facial growth constraint features; through a preset feature enhancement extraction network, performing attention fusion processing on the facial deformation features and the facial growth constraint features and outputting facial key features; performing feature dimensionality reduction and restoration processing on the facial key features related to the real facial image to obtain cross-age facial features; and performing identity information query processing on the cross-age facial features to determine the social security identity information of the medical treatment personnel.

[0008] Through the marking processing of dermal layer texture features and variable epidermal layer features in the embodiment of the present application, facial features related to age can be extracted more accurately, which is crucial for cross-age face recognition. Utilizing dynamic peeling of variable epidermal layer features and physical optical constraints on dermal layer texture features helps to optimize the facial image, reduce the interference of non-age-related features, and improve the recognition accuracy. And using a multi-scale temporal deformation model to extract facial feature points can capture the changes in facial shapes at different ages, thereby improving the robustness of recognition. It is also possible to determine facial growth constraint features by editing age attributes, which helps to identify individuals of different age groups, especially in the case of large age changes. The feature enhancement extraction network can enhance the quality of features and the performance of the recognition system by performing attention fusion processing on facial deformation features and growth constraint features. Performing dimensionality reduction processing on facial key features can reduce the computational complexity while retaining key information and improving the recognition efficiency. At the same time, by extracting cross-age facial features, effective recognition can be carried out among individuals of different age groups, expanding the application scope of face recognition. Finally, through identity information query processing, the social security identity information of the medical treatment personnel can be determined quickly and accurately, improving the efficiency and security of medical services.

[0009] In a feasible implementation, the initial facial image of the medical personnel is marked with respect to the dermal layer texture features and the variable epidermal layer features to determine the facial cortex feature image, which specifically includes: irradiating the facial area of the medical personnel with light of different polarization angles in sequence through the multi-band polarization light source pre-installed in the intelligent medical assistance machine, and synchronously collecting multi-angle polarization images of the facial area through a preset high-resolution polarization camera; based on the infrared band in the multi-band polarization light source, performing penetration imaging processing on the anisotropic textures in the multi-angle polarization images to determine the dermal layer texture features; wherein, the dermal layer texture features include: facial pore distribution features and facial blood vessel distribution features; compensating for environmental light interference on the multi-angle polarization images through a non-polarized reference light source, and performing feature marking processing on the multi-angle polarization images under high-scattering characteristics to determine the variable epidermal layer features; the variable epidermal layer features include: wrinkle distribution features, spot distribution features, and color spot distribution features; determining the initial facial image based on the anisotropic textures and the multi-angle polarization images processed by the high-scattering characteristics; performing dimensional stratification processing on the dermal layer texture features and the variable epidermal layer features in the initial facial image according to the RGB color channels to generate the facial cortex feature image.

[0010] In the embodiment of the present application, through the analysis of the physical characteristics of polarized light and the decoupling of neural network features, the extraction of dermal layer biometric features with age robustness is realized, solving the problems that traditional facial recognition is interfered by age-related features (such as wrinkles and color spots) in the epidermal layer, resulting in feature drift, existing optical imaging methods being unable to separate the multi-layer scattering characteristics of skin tissue, and conventional adversarial networks lacking physical model constraints, resulting in uncontrollable feature decoupling processes. Moreover, by using the RGB color channels, dimensional stratification processing of the dermal layer texture features and the variable epidermal layer features is realized, which is beneficial to generating a facial cortex feature image with multi-dimensional texture features.

[0011] In a feasible implementation, before dynamically peeling off the variable epidermal layer features in the facial cortex feature image and performing physical optical constraints on the dermal layer texture features to obtain an optimized facial image, the method further includes: inputting the historical facial surface feature image set into the U-Net architecture of the generator; wherein, the historical facial surface image is the input end, and both the dermal layer texture features and the variable epidermal layer features are the output ends; based on the dimensional stratification characteristics between the dermal layer texture features and the variable epidermal layer features, constructing a cortex discriminator under a double-branch structure; wherein, the cortex discriminator is used to discriminate the physiological texture authenticity of the dermal layer texture features and the dynamic noise distribution of the variable epidermal layer features; performing generation processing of an adversarial network on the U-Net architecture and the cortex discriminator to obtain an adversarial decomposition network; according to Ladv = E[logD(G(x))] + E[log(1 - D(x))], to obtain the adversarial loss L of the adversarial decomposition network adv ; where D is the cortical discriminator; G(x) is the dermal texture feature and the variable epidermal feature output by the generator; E is represented by a mathematical symbol; by using a preset polarized light transmission function as a regular term, the dimensional hierarchical constraint loss calculation related to the RGB color channel is performed on the dermal texture feature and the variable epidermal feature to obtain the physical constraint loss; control the cycle consistency between the dermal texture feature and the variable epidermal feature, and determine the cycle consistency loss; through the adversarial loss, the physical constraint loss, and the cycle consistency loss, perform adversarial training on the adversarial decomposition network to obtain an optimized adversarial decomposition network.

[0012] In the embodiment of the present application, through the collaborative design of the U-Net architecture and the dual-branch discriminator, the physical interpretable separation of the dynamic noise of the facial epidermal layer (such as light reflection, makeup interference) and the dermal biometric features (such as capillary distribution, collagen fiber orientation) is realized. After using the polarized light transmission function as a regular term, the physiological authenticity discrimination of the dermal texture feature is more accurate. By introducing an epidermal layer dynamic noise distribution discrimination module in the adversarial training, through the joint optimization of the adversarial loss and the physical constraint loss, the influence of environmental light changes (such as highlights / shadows) and short-term epidermal state fluctuations (such as sweat, oiliness) on the core biometric features can be effectively suppressed. Through the cycle consistency constraint, the closed-loop reconstruction of the epidermal-dermal dual-channel features is realized, and the mathematical mapping relationship between the light field transmission equation and the biological tissue characteristics is established, which is beneficial to the adversarial training of the resistance decomposition network and increases the adversarial performance of the network.

[0013] In a feasible implementation manner, the variable epidermal feature in the facial cortical feature image is dynamically peeled off, and the physical optical constraint is performed on the dermal texture feature to obtain an optimized facial image, which specifically includes: performing a reflectivity calculation related to the polarization angle and polarization degree on the current facial surface feature image to obtain the polarization reflection relationship between the high-scattering characteristic and the anisotropic texture; through the optimized adversarial decomposition network, and based on the polarization reflection relationship, perform real-time feedback on the signal-to-noise ratio of the variable epidermal feature in the facial surface feature image, and dynamically adjust the attention weight of the cortical discriminator to complete the dynamic peeling of the variable epidermal feature; based on non-negative matrix factorization, perform cross-interference suppression processing on the dynamic peeling result to obtain the optimized peeled dermal texture feature; perform physical optical constraint on the Fresnel reflection boundary of the dermal texture feature, and perform adaptive histogram equalization on the image feature corresponding to the dermal texture feature to obtain the optimized facial image with immutable texture features.

[0014] In the embodiments of the present application, by dynamically peeling off the characteristics of the variable epidermal layer, unstable characteristics caused by factors such as light and expression changes can be removed, thereby improving the stability and accuracy of face recognition. Calculating the reflectivity of the polarization angle and polarization degree, and analyzing the polarization reflection relationship between the high-scattering characteristics and the anisotropic texture helps to understand the optical characteristics of the facial surface layer more deeply and provides a basis for subsequent processing. It can real-time feedback the signal-to-noise ratio of the variable epidermal layer characteristics and dynamically adjust the attention weights of the cortical discriminator, enabling real-time adaptation to image changes and improving the real-time performance and adaptability of the recognition system. An optimized adversarial decomposition network can also be used to more effectively separate and extract facial features, enhancing the performance of the recognition system.

[0015] In a feasible implementation manner, through a multi-scale temporal deformation model, the optimized facial image is processed to extract facial feature points related to the topological structure and shape, obtaining facial deformation features, specifically including: extracting feature points from the optimized facial image according to the key facial feature points in the face recognition model, obtaining the original facial feature points; wherein, the key facial feature points at least include: mouth feature points, eye feature points, nose feature points, facial contour feature points, zygomatic contour feature points, and mandibular angle feature points; through the short-term dynamic scale in the multi-scale temporal deformation model, screening processing is performed on the muscle texture feature points related to neuromuscular control in the original facial feature points under the influence of age change factors, obtaining muscle texture feature points irrelevant to the age change factors; according to the medium-term dynamic scale in the multi-scale temporal deformation model, and through a non-rigid algorithm, calculating the distances between the feature points related to skeletal development features in the original facial feature points, obtaining a feature point distance term; and based on the recombination transformation of the feature point distance term, determining a key point term; performing a rigid calculation on the original facial feature points related to the topological map of the skeletal development feature points through an arc-node incidence matrix, obtaining a feature point rigidity term; performing a double-layer loop iterative calculation based on an energy function on the feature point distance term, the feature point rigidity term, and the key point term to determine the facial bone deformation trajectory; performing facial evolution modeling on the muscle texture feature points and the facial bone deformation trajectory through a topology-preserving manifold learning algorithm, obtaining a facial evolution model; and extracting the facial deformation features in the facial evolution model.

[0016] The embodiments of the present application effectively decouple age-related deformations and individual inherent characteristics by integrating the dual mechanisms of short-term dynamic scale screening and medium-term skeletal development modeling. The dynamic screening technology of neuromuscular control muscle texture features (such as LBP texture analysis) can eliminate the interference of age factors on skin texture and improve the robustness of the face recognition system in cross-age scenarios. Through the rigid constraint calculation of the arc-node incidence matrix and the double-layer iterative optimization of the energy function, while capturing the non-linear deformation of the bones, the stability of the facial topological structure is maintained.

[0017] In a feasible implementation manner, diffusion processing for age attribute editing of the optimized facial image is performed to determine facial growth constraint features, which specifically includes: identifying the existing target attribute regions in the optimized facial image through a GAN model and a diffusion model, and through a complementary attention branch, the supplementary attribute regions in the optimized facial image where facial features are missing and need to add attributes; wherein, the target attribute regions are attribute regions with insignificant changes in facial features with age; the supplementary attribute regions are attribute regions with significant changes in facial features with age; through the attention mask of the GAN model, feature generation processing for cross-age regions is performed on the supplementary attribute regions to obtain a first masked image, and according to a color mask, age change effect processing for cross-age regions is performed on the target attribute regions to obtain a second masked image; the first masked image and the second masked image are combined by masking to generate a facial growth image after age editing operations; constraint processing under an attention mechanism is performed on the key facial growth features in the facial growth image, and facial growth constraint features are extracted.

[0018] By combining the use of a GAN model and a diffusion model in the embodiments of the present application, age editing processing can be completed more efficiently, generating facial growth images with higher quality and more naturalness, and better using the attention mask to capture age change attributes. At the same time, based on the combined use of attribute regions with insignificant changes in facial features and attribute regions with significant changes in facial features, the facial features of a person in historical ages can be predicted more accurately, and combined with the constraint processing under the attention mechanism, the key facial growth features can be accurately constrained, thereby reducing computational power and improving the processing speed of the computer.

[0019] In a feasible implementation manner, through a preset feature enhancement extraction network, the facial deformation feature and the facial growth constraint feature are subjected to attention fusion processing, and facial key features are output, specifically including: through a spatial attention network, the facial deformation feature and the facial growth constraint feature are respectively subjected to spatial feature extraction processing to obtain a first spatial feature and a second spatial feature; through a channel attention network, the first spatial feature and the second spatial feature are respectively subjected to channel feature extraction processing to obtain a first channel feature and a second channel feature; according to the importance degree of the channels after expert scoring, weight values are assigned to the first channel feature and the second channel feature, and cross-age channel features with cross-age facial feature maps are marked; based on a fusion loss function and the cross-age channel features, the facial deformation feature and the facial growth constraint feature are subjected to feature merging processing under multi-head attention, and the face distributed features after learning and training through a fully connected layer are predicted and mapped into an output space to generate a cross-age facial image corresponding to the medical personnel; wherein, the cross-age facial image is a predicted facial image of the medical personnel in the historical age group; according to the key facial feature points in the face recognition model, feature points are extracted from the cross-age facial image to obtain the facial key features.

[0020] In the embodiment of the present application, by performing attention fusion processing on the facial deformation feature and the facial growth constraint feature, a fusion loss forced attention mechanism can be added on the basis of multi-head attention, and the facial deformation feature that does not change with age after adversarial decomposition and the facial growth constraint feature after age editing are subjected to comprehensive feature fusion processing, and then the facial key features in the cross-age stage are generated. Combining these facial key features, the corresponding personnel facial image can be quickly and accurately recognized in the subsequent facial information of the historical age group. At the same time, through spatial attention and channel attention, and the response of the feature enhancement extraction network to the key feature information, multiple different attention regions can be captured, realizing the spatio-temporal fusion of facial features, thereby outputting a more realistic cross-age facial image.

[0021] In a feasible implementation manner, the facial key features are subjected to feature dimensionality reduction and reduction processing with respect to a real facial image to obtain cross-age facial features, specifically including: according to Obtain the cross-age facial features at each age group Wherein, k is the feature type in the facial key features, N is the cross-age change attribute parameter corresponding to the facial deformation feature in the facial key features; C is the cross-age change attribute parameter corresponding to the facial growth constraint feature in the facial key features; both i and j are mathematical constants; Is the pooling window size of the real facial image; Dage is the age feature size parameter of the cross - age facial image; V age is the age range.

[0023] In a feasible implementation manner, identity information query processing is performed on the cross - age facial features to determine the social security identity information of the medical treatment personnel, which specifically includes: extracting the cross - age facial features corresponding to the medical treatment personnel; generating an identity code for the cross - age facial features to obtain the identity code information to be queried; inputting the identity code information to be queried into the social security personnel information database, and arranging the identity information within the matching degree threshold range according to the identity information matching degree; sequentially displaying the social security identity information to be determined according to the arrangement order of the identity information; and obtaining the social security identity information matching the medical treatment personnel based on the socially - security identity information after manual confirmation.

[0024] On the other hand, the embodiment of the present application also provides a face recognition system based on cross - age features. The system includes: a texture recognition module, configured to perform marking processing on the initial facial image of the medical treatment personnel collected for the dermal layer texture features and variable epidermis layer features, and determine a facial cortex feature image; dynamically stripping the variable epidermis layer features in the facial cortex feature image, and performing physical - optical constraint on the dermal layer texture features to obtain an optimized facial image; a facial topology recognition module, configured to perform extraction processing on the facial feature points related to the topological structure and shape of the optimized facial image to obtain facial deformation features; a facial growth prediction module, configured to perform diffusion processing on the age attribute editing of the optimized facial image to determine facial growth constraint features; a feature fusion module, configured to perform attention fusion processing on the facial deformation features and the facial growth constraint features, and output facial key features; an identity information query module, configured to perform feature dimensionality reduction and restoration processing on the facial key features for the real facial image to obtain cross - age facial features; and performing identity information query processing on the cross - age facial features to determine the social security identity information of the medical treatment personnel.

[0025] The present application provides a face recognition method and system based on cross - age features. Compared with the prior art, the embodiment of the present application has the following beneficial technical effects:

[0026] 1. Feature marking and processing: By performing marking processing on the dermal layer texture features and variable epidermis layer features, facial features related to age can be extracted more precisely, which is crucial for cross - age face recognition.

[0027] 2. Dynamic stripping and physical - optical constraint: Dynamically stripping the variable epidermis layer features and physically - optically constraining the dermal layer texture features helps to optimize the facial image, reduce the interference of non - age - related features, and improve the recognition accuracy.

[0028] 3. Multi-scale temporal deformation model: Using the multi-scale temporal deformation model to extract facial feature points can capture the changes in facial shapes at different ages, thereby improving the robustness of recognition.

[0029] 4. Diffusion processing for age attribute editing: By editing age attributes and determining facial growth constraint features, it helps to identify individuals of different age groups, especially in cases where age changes are significant.

[0030] 5. Feature enhancement extraction network: The feature enhancement extraction network can enhance the quality of features and improve the performance of the recognition system by fusing facial deformation features and growth constraint features through attention.

[0031] 6. Feature dimensionality reduction and restoration processing: Performing dimensionality reduction on key facial features can reduce computational complexity while retaining key information and improving recognition efficiency.

[0032] 7. Cross-age facial feature extraction: By extracting cross-age facial features, this method can effectively identify individuals of different age groups, expanding the application scope of face recognition.

[0033] 8. Identity information query processing: Finally, through identity information query processing, the social security identity information of medical personnel can be quickly and accurately determined, improving the efficiency and security of medical services. Description of the Drawings

[0034] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0035] Figure 1 It is a flowchart of a face recognition method based on cross-age features provided by an embodiment of the present application;

[0036] Figure 2 It is a schematic diagram of the structure of a recognition network integrating feature enhancement provided by an embodiment of the present application;

[0037] Figure 3 It is a schematic diagram of the structure of a face recognition device based on cross-age features provided by an embodiment of the present application. Detailed Embodiments

[0038] To enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0039] The embodiment of this application provides a face recognition method based on cross-age features. As Figure 1 shown, the face recognition method based on cross-age features specifically includes steps S101 - S106:

[0040] S101. Perform marking processing on the initial facial image of the medical personnel collected for dermal layer texture features and variable epidermal layer features to determine the facial cortex feature image.

[0041] It should be noted that in the application of the polarized light system, the facial skin computer-aided imaging system has both cross-polarized light and parallel polarized light sources. Parallel polarized light is mostly used to enhance the detailed features on the skin surface, including fine lines, pores, and spots, etc. Cross-polarized light is used to enhance the tissue features under the skin surface, including pigments, blood vessels, etc. The facial skin computer-aided imaging system often analyzes the white light and polarized light images through specific algorithms, and finally presents the calculation results in the form of pictures and data in multiple modes.

[0042] Specifically, first, it is necessary to sequentially irradiate the facial area of the medical personnel with light of different polarization angles through the multi-band polarized light source pre-installed in the intelligent medical assistance machine (such as an intelligent self-service window, etc.), and synchronously collect multi-angle polarized images of the facial area through a preset high-resolution polarized camera.

[0043] Furthermore, based on the infrared band in the multi-band polarized light source, perform penetration imaging processing on the anisotropic texture in the multi-angle polarized images to determine the dermal layer texture features. Among them, the dermal layer texture features include: facial pore distribution features and facial blood vessel distribution features.

[0044] Furthermore, it is also necessary to perform ambient light interference compensation on the multi-angle polarized images through a non-polarized reference light source, and perform feature marking processing on the multi-angle polarized images under high scattering characteristics to determine the variable epidermal layer features. The variable epidermal layer features include: wrinkle distribution features, spot distribution features, and freckle distribution features.

[0045] Furthermore, based on the multi-angle polarized images processed by the anisotropic texture and high scattering characteristics, determine the initial facial image.

[0046] Further, finally, according to the RGB color channels, dimensional stratification processing is performed on the dermal layer texture features and the variable epidermis layer features in the initial facial image to generate a facial cortex feature image. That is, this facial cortex image can be more consistently bound to the identity information of the medical personnel, and at the same time, it can deeply identify the detailed features of the human face, avoiding malicious use of other human faces by others.

[0047] In one embodiment, the multi-angle polarized light source array can adopt a 450nm / 630nm dual-wavelength LED ring light source, configured with 0° / 45° / 90° / 135° four-way linear polarizers. The high-resolution polarization camera can be equipped with a rotating polarizer module (continuously rotating 0-180°) to collect a sequence of 16bit RAW format polarization images. An optical preprocessing module can also be pre-loaded: the degree of polarization (DoP) map and the angle of polarization (AoP) map are obtained through Stokes vector calculation. Then, a layered reflection model of the skin tissue can be established. This model can use the infrared band in the multi-band polarized light source and the unpolarized reference light source, combined with the epidermal / dermal layer reflection coefficient matrix and the dermal layer scattering coefficient, to respectively identify the dermal layer texture features and the variable epidermis layer features. Then, using the RGB color channels, the identified cortex features located in different dimensional levels are marked in layers, and a marked facial cortex feature image is generated.

[0048] S102. Dynamically strip the variable epidermis layer features in the facial cortex feature image, and perform physical-optical constraints on the dermal layer texture features to obtain an optimized facial image.

[0049] Specifically, the historical facial surface feature image set needs to be input into the U-Net architecture of the generator first. Among them, the historical facial surface image is the input end, and both the dermal layer texture features and the variable epidermis layer features are the output ends.

[0050] In one embodiment, the historical facial surface feature image set (RGB images taken with polarized light, resolution 1024×1024) can be input first to complete the pre-training process of the pre-stage adversarial decomposition network. In the generator (G), the U-Net architecture is adopted. The input is the facial surface image, and the output is two-branch features: Dermal layer texture features: including physiological textures (such as capillary distribution, pore structure); Variable epidermis layer features: dynamic noise (such as sweat, oil, short-term wrinkles). For example: input a multi-angle facial image with polarization angles of 0°, 45°, 90°, and 135°. The generator outputs a dermal layer feature map (512×512 grayscale image) and an epidermis layer noise map (dynamically changing RGB noise distribution).

[0051] Furthermore, based on the dimensional stratification characteristics between the dermal layer texture features and the variable epidermal layer features, a cortical discriminator under a dual-branch structure is constructed. Among them, the cortical discriminator is used to discriminate the physiological texture authenticity of the dermal layer texture features and the dynamic noise distribution of the variable epidermal layer features.

[0052] In one embodiment, regarding the construction of the dual-branch cortical discriminator (D), the dermal layer discrimination branch can be configured first: to discriminate whether the generated dermal layer texture conforms to physiological authenticity (such as the continuity and fractal characteristics of capillaries). Then, the epidermal layer discrimination branch is configured: to discriminate whether the dynamic noise distribution conforms to the statistical law of natural epidermal noise (such as Gaussian-Poisson mixture distribution). Combining adversarial design: The generator and the discriminator are optimized through adversarial training (GAN), the generator tries to deceive the discriminator, and the discriminator improves its discrimination ability. Then, during subsequent cortical discrimination, the dermal layer discriminator extracts the texture direction consistency features through a convolutional network, and the epidermal layer discriminator verifies the noise spectrum distribution through frequency domain analysis.

[0053] Furthermore, the U-Net architecture and the cortical discriminator are further processed for generating an adversarial network to obtain an adversarial decomposition network. According to L adv =E[log D(G(x))]+E[log(1 - D(x))], the adversarial loss of the adversarial decomposition network is obtained. adv Among them, D is the cortical discriminator; G(x) is the dermal layer texture features and variable epidermal layer features output by the generator; E is represented by a mathematical symbol.

[0054] Furthermore, it is also necessary to use the preset polarization light transmission function as a regularization term to calculate the dimensional hierarchical constraint loss of the dermal layer texture features and the variable epidermal layer features with respect to the RGB color channels to obtain the physical constraint loss. Control the cycle consistency between the dermal layer texture features and the variable epidermal layer features, and determine the cycle consistency loss. Then, through the adversarial loss, the physical constraint loss, and the cycle consistency loss, the adversarial decomposition network is adversarially trained to obtain an optimized adversarial decomposition network.

[0055] As a feasible implementation method, in order to constrain the reflection characteristics of the dermal layer and the epidermal layer in the RGB channels based on the polarization light transmission function:

[0056] The physical constraint loss (Physics-based Loss) L phy can be obtained, where Φ c is the channel constraint corresponding to the polarization light transmission function. In order to ensure that the dermal layer and epidermal layer features can be recombined to restore the original input image: L cycle =||G -1(G(x)) - x || 1 can obtain the Cycle Consistency Loss.

[0057] Further, perform reflectivity calculations on the current facial surface feature image regarding the polarization angle and polarization degree to obtain the polarization reflection relationship between the high-scattering property and the anisotropic texture.

[0058] Further, through the optimized adversarial decomposition network and based on the polarization reflection relationship, perform real-time feedback on the signal-to-noise ratio of the variable epidermal layer features in the facial surface feature image, and dynamically adjust the attention weights of the cortical discriminator to complete the dynamic peeling of the variable epidermal layer features.

[0059] Further, based on non-negative matrix factorization, perform cross-interference suppression processing on the dynamic peeling result to obtain the optimized dermal layer texture features after peeling.

[0060] In one embodiment, calculate the polarization reflection parameters of the high-scattering property (dermal layer) and the anisotropic texture (epidermal layer) according to the polarization angle and polarization degree of the facial surface. For example: the polarization degree of the epidermal layer is lower than that of the dermal layer (due to oil scattering), and dynamically adjust the discriminator weights through the polarization reflection coefficient. Then perform non-negative matrix factorization (NMF) to suppress cross-interference, that is, perform non-negative decomposition on the dermal layer feature matrix after peeling to eliminate residual epidermal noise.

[0061] Further, finally perform physical optical constraints on the Fresnel reflection boundary of the dermal layer texture features, and perform adaptive histogram equalization on the image features corresponding to the dermal layer texture features to obtain an optimized facial image with immutable texture features. Among them, the immutable texture features are facial texture features that do not change with age.

[0062] In one embodiment, the skin refractive index in the Fresnel constraint can be used to constrain the physical reflection boundary of the dermal layer texture and suppress optical artifacts. Then perform adaptive histogram equalization, that is, enhance the local contrast of the dermal layer image to highlight the facial texture features that do not change significantly with age.

[0063] As a feasible implementation, cross-interference is suppressed through non-negative matrix factorization, which can ensure that the texture features of the dermis layer after peeling are more pure, reducing the influence of noise and interference. Moreover, physical optical constraints of the Fresnel reflection boundary on the texture features of the dermis layer can simulate the optical behavior in a real environment, improving the robustness of the recognition system under complex lighting conditions. At the same time, adaptive histogram equalization of the image features can improve the image contrast, making the image more uniform visually, which is beneficial to subsequent feature extraction and recognition. The finally obtained optimized facial image has immutable texture features, which means that even if the facial expression or lighting conditions change, these texture features still remain stable, helping to improve the accuracy of face recognition.

[0064] S103. Through a multi-scale temporal deformation model, perform extraction processing on the optimized facial image for facial feature points under the topological structure and shape to obtain facial deformation features.

[0065] Specifically, according to the key facial feature points in the face recognition model, perform extraction of feature points on the optimized facial image to obtain the original facial feature points. Among them, the key facial feature points at least include: mouth feature points, eye feature points, nose feature points, face contour feature points, zygomatic contour feature points, and mandibular angle feature points.

[0066] In one embodiment, first input the optimized facial image (the texture feature map of the dermis layer output from the previous adversarial decomposition network, with a resolution of 1024×1024). Then, based on the pre-trained model of HRNet (High-Resolution Network), locate 68 key feature points. Examples of key point classification: Mouth: 20 points (lip contour and corners of the mouth). Eyes: 12 points (upper and lower eyelids and corners of the eyes). Nose: 9 points (bridge of the nose, nostrils, and tip of the nose). Face contour: 17 points (zygomatic bone, mandibular angle, and chin). Finally, output the set of original facial feature points P = {p1, p2,..., p68}; P = {p1, p2,..., p68}. For example: For a facial image of a 50-year-old male, extract the coordinates of the mandibular angle points (p16, p17) and calculate the change in mandibular width.

[0067] Furthermore, through the short-term dynamic scale in the multi-scale temporal deformation model, perform screening processing on the texture feature points related to neuromuscular control in the original facial feature points under age change factors to obtain texture feature points that are irrelevant to age change factors.

[0068] In one embodiment, the screening of texture feature points independent of age change factors is completed using the Short-term Dynamic Scale. The short-term dynamic changes of feature point micro-movements (such as blinking and smiling) can be analyzed using a neuromuscular control model. If the displacement variance of a feature point in the time series is less than a threshold, it is determined as a stable texture point independent of age. For example, the wrinkle points around the eyes (greatly affected by age change) are excluded, while the bone points on the nose bridge (less affected by age change) are retained.

[0069] Further, according to the Mid-term Dynamic Scale in the multi-scale time series deformation model, and through a non-rigid algorithm, the distance calculation between the original facial feature points related to the bone development feature points is performed to obtain the feature point distance term. And based on the recombination transformation of the feature point distance term, the key point term is determined.

[0070] In one embodiment, it is also necessary to quantify the change in the distance between feature points caused by bone development. Use Thin Plate Spline (TPS) to perform non-rigid alignment on the feature points and calculate the change amount of the distance between bone points. Then use the feature point distance term: Determine the feature point p i And the feature point p j The distance term between the reference age t0 and the target age t. Then select point pairs sensitive to bone deformation (such as the mandibular angle-zygomatic bone point pair) to construct a set of key point terms.

[0071] Further, through the Arc-Node Matrix, the rigid calculation of the topological graph of the original facial feature points related to the bone development feature points is performed to obtain the feature point rigidity term. Then, the feature point distance term, the feature point rigidity term, and the key point term are subjected to a double-layer loop iterative calculation based on the energy function to determine the facial bone deformation trajectory.

[0072] In one embodiment, regarding the Arc-Node Matrix, the bone feature points need to be connected into a rigid structure (such as a triangular mesh of the zygomatic bone - nose bridge - mandible), and then the rigid term calculation is performed: through Procrustes analysis, the rotation and translation invariance errors of the topological graph are calculated. Finally, combined with the above feature point distance term, feature point rigidity term, and key point term, a double-layer loop iterative calculation based on the energy function is performed, that is, the double-layer loop optimization of the energy function (the outer layer iterates the age step size, and the inner layer iterates the deformation parameters), and the L-BFGS algorithm is used to minimize the energy function to output the bone deformation trajectory (such as the forward movement rate of the zygomatic bone and the blunt curve of the mandibular angle).

[0073] Furthermore, through a topology-preserving manifold learning algorithm, a facial evolution model is built for the texture feature points and the facial bone deformation trajectories, and the facial deformation features in the facial evolution model are extracted. That is, in topology-preserving manifold learning, algorithms such as t-SNE or UMAP can be used to map the texture feature points and bone trajectories to a low-dimensional manifold space while preserving the local neighborhood relationship. Combining the bone deformation subspace in the facial evolution model, which characterizes bone development (such as the opening angle of the mandibular angle with age), and the texture aging subspace, which characterizes skin elasticity decay (such as the depth of nasolabial folds), the data support processing of the facial evolution model is completed. Finally, the principal components (PCA) are extracted from the manifold space to obtain the cross-age deformation basis vectors, which are the facial deformation features in the facial evolution model.

[0074] As a feasible implementation, multi-scale modeling combining short-term muscle dynamics (<5 years) and mid-term bone development dynamics (5 - 15 years) can be adopted. The facial evolution model constructed through manifold learning can predict the facial deformation trajectories in the previous 10 - 20 years. The feature point distance term recombination transformation and key point term dynamic pruning techniques are used to reduce the computational complexity while maintaining the accuracy of the non-rigid algorithm. Through the topology-preserving manifold learning algorithm, the spatio-temporal alignment of bone development features (rigid terms) and muscle movement features (non-rigid terms) is achieved, which is beneficial to the generation of the facial evolution model.

[0075] S104. Perform diffusion processing on the optimized facial image for age attribute editing to determine the facial growth constraint features.

[0076] Specifically, through the GAN model and the diffusion model, the existing target attribute regions in the optimized facial image are recognized, and the complementary attention branch is used to optimize the supplementary attribute regions that are missing in the facial image and need to add attributes. Among them, the target attribute region is the attribute region where the facial features change insignificantly with age. The supplementary attribute region is the attribute region where the facial features change significantly with age.

[0077] In one embodiment, the optimized facial image (the immutable texture feature map output from the previous step) is first input into the GAN model and the diffusion model. Then, attribute region recognition is performed. Among them, the target attribute regions (where age changes are not significant): bone contour, eye shape, nose bridge structure (segmented by a pre-trained UNet, with the mask denoted as Mtarget). Supplementary attribute regions (where age changes are significant): skin texture, wrinkles, hairline (mask denoted as Msupp). Then, using the complementary attention branch, the StyleGAN3 generator can be employed, with Msupp as the attention guidance, to generate age features of the supplementary region (such as removing wrinkles, skin texture, etc.). Then, the implicit denoising network of Stable Diffusion is used, with the Mtarget mask as the constraint, to maintain the stability of the bone structure. For example: keep the shape of the target regions (cheekbones, mandible) of a 60-year-old female image, and then use the diffusion model to generate the dermal layer texture of a 30-year-old (Msupp region).

[0078] Furthermore, through the attention mask of the GAN model, feature generation processing for the supplementary attribute region across age regions is performed to obtain the first masked image, and according to the color mask, age change effect processing for the target attribute region across age regions is performed to obtain the second masked image.

[0079] In one embodiment, the first masked image (generated by GAN) is: the current facial image + the target age label (such as "-20 years old"), and then using mask multiplication, the rejuvenated features of the supplementary region are generated (such as removing crow's feet, nasolabial folds). The second masked image (generated by the diffusion model) is: the same image + age prompt words (such as "wrinkles, eye bags"). Color mask constraints are also required, that is, skin color correction is performed to conform to the skin pigment deposition law of young people.

[0080] Furthermore, the first masked image and the second masked image are combined by masking to generate the facial growth image after the age editing operation. That is, the first masked image and the second masked image can be fused in the gradient domain at the mask boundary to eliminate the seam, thereby realizing Poisson image fusion. Then, the facial growth image after age editing is output, increasing age-related changes while retaining the original identity features.

[0081] Furthermore, constraint processing under the attention mechanism is performed on the key facial growth features in the facial growth image, and the facial growth constraint features are extracted.

[0082] As a feasible implementation, a Vision Transformer (ViT) can be used to extract multi-scale feature maps of facial growth images, calculate the attention weight matrix, and then constrain the consistency between key growth regions (such as nasolabial fold depth, drooping eye corners, hairline, and mouth position) and the biomechanical model. The constrained features are compressed into low-dimensional vectors by an autoencoder and used as cross-age deformation descriptions, and facial growth constraint features are extracted.

[0083] That is, through the collaborative generation of GAN and diffusion models, the physical realism of Poisson fusion, and the attention mechanism of biomechanical constraints, high-precision and identity-preserving cross-age facial editing is achieved, solving the problems of excessive deformation or identity loss in traditional methods, and being able to accurately predict the facial growth constraint features of a person when they were young under cross-age deformation.

[0084] S105. Through a preset feature enhancement extraction network, perform attention fusion processing on the facial deformation features and the facial growth constraint features, and output the facial key features.

[0085] Specifically, through a spatial attention network, perform spatial feature extraction processing on the facial deformation features and the facial growth constraint features respectively, and obtain the first spatial feature and the second spatial feature respectively.

[0086] Furthermore, through a channel attention network, perform channel feature extraction processing on the first spatial feature and the second spatial feature respectively, and obtain the first channel feature and the second channel feature respectively.

[0087] Furthermore, according to the importance of channels after expert scoring, assign weight values to the first channel feature and the second channel feature, and mark the cross-age channel features with cross-age facial feature maps.

[0088] In one embodiment, Figure 2 is a schematic diagram of an identification network structure with fused feature enhancement provided by an embodiment of the present application. As Figure 2 shown, first use a pre-trained spatial attention network (such as SENet or CBAM) to process the facial deformation features and the facial growth constraint features. Then extract spatial features from the two feature sets respectively to obtain the first spatial feature (F1) and the second spatial feature (F2). Next, apply a channel attention network (such as CBAM or SENet) to F1 and F2 to extract channel features, and obtain the first channel feature (C1) and the second channel feature (C2) respectively. Then, according to expert scoring, the importance of the channel features needs to be determined: First, assign weight values to C1 and C2 to reflect their importance in cross-age facial feature recognition. Second, mark the channel features with cross-age facial features to form cross-age channel features (CCF).

[0089] Further, based on the fusion loss function and cross-age channel features, the facial deformation features and facial growth constraint features are subjected to feature merging processing under multi-head attention, and the face distributed features after learning and training through a fully connected layer are predicted and mapped into the output space to generate a cross-age facial image corresponding to the medical treatment personnel. Among them, the cross-age facial image is the predicted facial image of the medical treatment personnel in the historical age range.

[0090] Further, according to the key facial feature points in the face recognition model, the feature points of the cross-age facial image are extracted to obtain the facial key features.

[0091] In one embodiment, as Figure 2 shown, in combination with the foregoing embodiments, it is also necessary to use the multi-head attention mechanism, in combination with the fusion loss function, to merge the features of F1 and F2 with the CCF, and through the learning of the fully connected layer, map the merged features into the face distributed feature space. Moreover, it is necessary to use the trained face distributed features, predict and map them into the output space to generate a cross-age facial image corresponding to the historical age range of the medical treatment personnel. Using the key facial feature point detection algorithm in the face recognition model (such as a deep learning-based facial key point detector), finally, the facial key features are extracted from the generated cross-age facial image.

[0092] S106. Perform feature dimensionality reduction and reduction processing on the facial key features with respect to the real facial image to obtain cross-age facial features. And perform identity information query processing on the cross-age facial features to determine the social security identity information of the medical treatment personnel.

[0093] Specifically, use to obtain the cross-age facial features for each age range where k is the feature type in the facial key features, N is the cross-age change attribute parameter corresponding to the facial deformation features in the facial key features; C is the cross-age change attribute parameter corresponding to the facial growth constraint features in the facial key features; both i and j are mathematical constants; is the pooling window size of the real facial image; D age is the age feature size parameter of the cross-age facial image; V age is the age range.

[0094] Further, it is also necessary to extract the cross-age facial features corresponding to the medical treatment personnel. Generate the identity encoding of the cross-age facial features to obtain the identity encoding information to be queried. Input the identity encoding information to be queried into the social security personnel information database, and arrange the identity information within the matching degree threshold range according to the identity information matching degree.

[0095] Further, in combination with the arrangement order of the identity information, the social security identity information to be determined is displayed in sequence. Based on the social security identity information after manual confirmation, the social security identity information matching the medical treatment personnel is obtained.

[0096] In one embodiment, the aforementioned face recognition model is used to extract cross-age face features from the face image of the medical treatment personnel. Then, the extracted features are standardized to facilitate subsequent encoding and comparison. Then, an autoencoder or variational autoencoder (VAE) model in deep learning is used to encode the standardized cross-age face features to generate identity encoding information. At the same time, the identity encoding information should contain sufficient features to distinguish different individuals, but at the same time be compact enough to reduce storage requirements. It is also necessary to utilize a pre-created social security personnel information database, which contains the face features and corresponding social security identity information of each person, to ensure that the face features in the information database have been converted into the same encoding format as the query features.

[0097] In one embodiment, it is also necessary to input the generated identity encoding information to be queried into the social security personnel information database. Similarity metrics (such as cosine similarity, Euclidean distance, etc.) can be used to calculate the matching degree between the query encoding and the encoding in the database. The social security identity information is sorted according to the matching degree, and the information with the highest matching degree is preferentially displayed. And a matching degree threshold is set, and only the identity information with a matching degree higher than this threshold is displayed. Then, the system sequentially displays the to-be-determined social security identity information after sorting. Then, through the user interface, the manual operator or the medical treatment personnel confirms the displayed information. Finally, based on the result of the manual confirmation, the social security identity information matching the medical treatment personnel is determined.

[0098] As a feasible implementation method, through cross-age face feature extraction and encoding, the system can process the images of medical treatment personnel of different age groups, improving the accuracy of identity verification. The identity encoding and matching degree calculation process are automated, reducing manual intervention and improving the processing speed. And by setting the matching degree threshold, the accuracy of the matching result is ensured, and the risk of incorrect matching is reduced. The system can also display the matching result through a clear interface, facilitating the manual operator or the medical treatment personnel to confirm the identity information. At the same time, the optimized identity recognition process reduces the query time and manual workload, improving the overall work efficiency. The security of the identity information database is strengthened, and personal privacy and data security are ensured through a strict matching process.

[0099] In addition, the embodiment of the present application also provides a face recognition system based on cross-age features, as Figure 3 shown, the face recognition system 300 based on cross-age features includes:

[0100] The texture recognition module 310 is used to perform marking processing on the initial facial image of the medical personnel collected, regarding the texture features of the dermal layer and the characteristics of the variable epidermal layer, to determine the facial cortex feature image. Dynamically strip the characteristics of the variable epidermal layer in the facial cortex feature image, and perform physical optical constraints on the texture features of the dermal layer to obtain an optimized facial image.

[0101] The facial topology recognition module 320 is used to perform extraction processing on the facial feature points under the topological structure and shape of the optimized facial image to obtain facial deformation features.

[0102] The facial growth prediction module 330 is used to perform diffusion processing on the age attribute editing of the optimized facial image to determine the facial growth constraint features.

[0103] The feature fusion module 340 is used to perform attention fusion processing on the facial deformation features and the facial growth constraint features, and output the facial key features.

[0104] The identity information query module 350 is used to perform dimensionality reduction and restoration processing on the facial key features regarding the real facial image to obtain cross-age facial features. And perform identity information query processing on the cross-age facial features to determine the social security identity information of the medical personnel.

[0105] In the embodiment of the present application, through the marking processing of the texture features of the dermal layer and the characteristics of the variable epidermal layer, facial features related to age can be extracted more accurately, which is crucial for cross-age face recognition. By dynamically stripping the characteristics of the variable epidermal layer and physically optically constraining the texture features of the dermal layer, it helps to optimize the facial image, reduce the interference of non-age-related features, and improve the recognition accuracy. And using a multi-scale temporal deformation model to extract facial feature points can capture the changes in facial shapes at different ages, thereby improving the robustness of recognition. It is also possible to determine the facial growth constraint features by editing the age attribute, which helps to identify individuals of different age groups, especially in the case of large age changes. The feature enhancement extraction network can enhance the quality of features and the performance of the recognition system by performing attention fusion processing on the facial deformation features and the growth constraint features. Performing dimensionality reduction processing on the facial key features can reduce the computational complexity, while retaining key information and improving the recognition efficiency. At the same time, by extracting cross-age facial features, effective recognition can be performed among individuals of different age groups, expanding the application scope of face recognition. Finally, through the identity information query processing, the social security identity information of the medical personnel can be determined quickly and accurately, improving the efficiency and security of medical services.

[0106] The various embodiments in this application are all described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the corresponding parts of the method embodiments for the relevant content.

[0107] The specific embodiments of the present application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order from that in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0108] The above description is only for the embodiments of the present application and is not intended to limit the present application. For those skilled in the art, various modifications and changes can be made to the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included within the scope of the claims of the present application.

Claims

1. A face recognition method based on cross-age features, characterized in that: The method comprises: The collected initial facial images of the medical personnel are marked with the dermis texture features and the variable epidermis features to determine the facial cortex feature image; Dynamically stripping the variable epidermal layer features in the facial cortical feature image and physically and optically constraining the dermal layer texture features to obtain an optimized facial image; Extracting facial feature points under relevant topological structure and shape from the optimized facial image through a multi-scale temporal deformation model to obtain facial deformation features; Performing diffusion processing of age attribute editing on the optimized facial image to determine facial growth constraint features; Through a preset feature enhancement extraction network, the facial deformation feature and the facial growth constraint feature are subjected to attention fusion processing, and key facial features are output; The facial key features are processed by feature dimensionality reduction and restoration related to the real facial image to obtain cross-age facial features; and identity information query processing is performed on the cross-age facial features to determine the social security identity information of the medical personnel.

2. The face recognition method based on cross-age features according to claim 1, characterized in that: The collected initial facial images of the medical personnel are marked with the dermis texture features and the variable epidermis features to determine the facial cortex feature image, including: The facial area of ​​the medical personnel is sequentially irradiated with light of different polarization angles by using the multi-band polarized light source pre-installed in the smart medical mutual aid machine, and the multi-angle polarized images of the facial area are synchronously collected by using a preset high-resolution polarization camera; Based on the infrared band in the multi-band polarized light source, the anisotropic texture in the multi-angle polarized image is subjected to penetration imaging processing to determine the texture features of the dermis layer; wherein the texture features of the dermis layer include: facial pore distribution features and facial blood vessel distribution features; By using a non-polarized reference light source, the multi-angle polarized image is compensated for ambient light interference, and the multi-angle polarized image is subjected to feature marking processing under high scattering characteristics to determine the variable epidermis layer characteristics; the variable epidermis layer characteristics include: wrinkle distribution characteristics, spot distribution characteristics and color spot distribution characteristics; Determining the initial facial image based on the anisotropic texture and the multi-angle polarization image processed with the high scattering characteristic; According to the RGB color channels, the dermis layer texture features and the variable epidermis layer features in the initial facial image are dimensionally layered to generate the facial cortex feature image.

3. The face recognition method based on cross-age features according to claim 1, characterized in that: Before dynamically stripping the variable epidermal layer features in the facial cortical feature image and performing physical optical constraints on the dermal layer texture features to obtain an optimized facial image, the method further includes: Input the historical facial surface feature image set into the U-Net architecture of the generator; the historical facial surface image is the input end, and the dermis texture feature and the variable epidermis feature are both output ends; Based on the dimensional hierarchical characteristics between the dermis texture features and the variable epidermis features, a cortex discriminator under a double-branch structure is constructed; wherein the cortex discriminator is used to discriminate the physiological texture authenticity of the dermis texture features and the dynamic noise distribution of the variable epidermis features; The U-Net architecture and the cortical discriminator are subjected to generation processing related to an adversarial network to obtain an adversarial decomposition network; According to L adv =E[log D(G(x))]+E[log(1-D(x))], and the adversarial loss L of the adversarial decomposition network is obtained. adv ; Wherein, D is the cortical discriminator; G(x) is the dermal texture feature and the variable epidermal feature output by the generator; E is a mathematical symbol representation; By using a preset polarized light transmission function as a regularization term, the dermis layer texture feature and the variable epidermis layer feature are subjected to a dimensional level constraint loss calculation related to the RGB color channel to obtain a physical constraint loss; Controlling the dermis layer texture feature and the variable epidermis layer feature to achieve cycle consistency, and determining the cycle consistency loss; The adversarial decomposition network is adversarially trained through the adversarial loss, the physical constraint loss and the cycle consistency loss to obtain an optimized adversarial decomposition network.

4. The face recognition method based on cross-age features according to claim 3, characterized in that: The variable epidermal layer features in the facial cortical feature image are dynamically stripped and the dermal layer texture features are physically and optically constrained to obtain an optimized facial image, specifically comprising: Performing reflectivity calculation on the current facial surface feature image in terms of polarization angle and polarization degree to obtain a polarization reflectivity relationship between high scattering characteristics and anisotropic texture; Through the optimized adversarial decomposition network and based on the polarization reflection relationship, the signal-to-noise ratio of the variable epidermal layer features in the facial surface feature image is fed back in real time, and the attention weight of the cortical discriminator is dynamically adjusted to complete the dynamic stripping of the variable epidermal layer features; Based on non-negative matrix decomposition, the dynamic peeling result is subjected to cross-interference suppression processing to obtain the texture features of the dermis layer after optimized peeling; The dermis layer texture features are subjected to physical optical constraints of Fresnel reflection boundaries, and the image features corresponding to the dermis layer texture features are subjected to adaptive histogram equalization to obtain the optimized facial image with immutable texture features.

5. The face recognition method based on cross-age features according to claim 1, characterized in that: Through the multi-scale temporal deformation model, the optimized facial image is subjected to extraction processing of facial feature points under relevant topological structure and shape to obtain facial deformation features, specifically including: Extracting feature points from the optimized facial image according to the key facial feature points in the face recognition model to obtain original facial feature points; wherein the key facial feature points include at least: mouth feature points, eye feature points, nose feature points, facial contour feature points, zygomatic contour feature points and mandibular angle feature points; Using the short-term dynamic scale in the multi-scale temporal deformation model, the texture feature points related to neuromuscular control in the original facial feature points are screened under the age change factor to obtain the texture feature points that are not related to the age change factor; According to the mid-term dynamic scale in the multi-scale temporal deformation model, and by using a non-rigid algorithm, the distance between the original facial feature points related to the skeletal development feature points is calculated to obtain a feature point distance term; and based on the reorganization transformation of the feature point distance term, a key point term is determined; Through the arc-node association matrix, the rigidity calculation of the original facial feature points related to the topological map of the bone development feature points is performed to obtain the feature point rigidity terms; The feature point distance term, the feature point rigidity term and the key point term are subjected to double-layer loop iterative calculation based on an energy function to determine a facial bone deformation trajectory; The facial evolution model is obtained by performing facial evolution modeling on the texture feature points and the facial bone deformation trajectory through a topology-preserving manifold learning algorithm; and facial deformation features in the facial evolution model are extracted.

6. The face recognition method based on cross-age features according to claim 1, characterized in that: Diffusion processing of age attribute editing is performed on the optimized facial image to determine facial growth constraint features, specifically including: By using the GAN model and the diffusion model, the target attribute region existing in the optimized facial image is identified and processed, and by using the complementary attention branch, the supplementary attribute region in the optimized facial image where the face is missing and needs to be added with attributes; wherein the target attribute region is an attribute region where the facial features do not change significantly with age; and the supplementary attribute region is an attribute region where the facial features change significantly with age; The supplementary attribute region is processed for feature generation across age regions by using the attention mask of the GAN model to obtain a first mask image, and the target attribute region is processed for age change effect across age regions according to the color mask to obtain a second mask image; Performing mask combination on the first mask image and the second mask image to generate a facial growth image after the age editing operation; The key facial growth features in the facial growth image are subjected to constraint processing under an attention mechanism, and facial growth constraint features are extracted.

7. The face recognition method based on cross-age features according to claim 1, characterized in that: Through a preset feature enhancement extraction network, the facial deformation features and the facial growth constraint features are subjected to attention fusion processing, and facial key features are output, specifically including: Through a spatial attention network, spatial feature extraction processing is performed on the facial deformation feature and the facial growth constraint feature to obtain a first spatial feature and a second spatial feature respectively; Through the channel attention network, performing channel feature extraction processing on the first spatial feature and the second spatial feature respectively, to obtain a first channel feature and a second channel feature respectively; According to the importance of the channels scored by the experts, weight values ​​are assigned to the first channel features and the second channel features, and cross-age channel features having cross-age facial feature maps are marked; Based on the fusion loss function and the cross-age channel features, the facial deformation features and the facial growth constraint features are subjected to feature merging processing under multi-head attention, and the distributed features of the face learned and trained by the fully connected layer are predicted and mapped to the output space to generate a cross-age facial image corresponding to the medical personnel; wherein the cross-age facial image is a predicted facial image under the historical age group of the medical personnel; According to the key facial feature points in the face recognition model, feature points are extracted from the cross-age facial image to obtain the key facial features.

8. The face recognition method based on cross-age features according to claim 1, characterized in that: The facial key features are processed by feature dimensionality reduction and restoration related to the real facial image to obtain cross-age facial features, specifically including: according to Get the cross-age facial features under each age group Wherein, k is the feature type in the facial key feature, N is the cross-age variation attribute parameter corresponding to the facial deformation feature in the facial key feature; C is the cross-age variation attribute parameter corresponding to the facial growth constraint feature in the facial key feature; i and j are both mathematical constants; is the pooling window size of the real face image; D age is the age feature size parameter of cross-age facial images; V age For age range.

9. The face recognition method based on cross-age features according to claim 1, characterized in that: Perform identity information query processing on the cross-age facial features to determine the social security identity information of the medical personnel, specifically including: Extracting cross-age facial features corresponding to the medical personnel; Performing identity coding on the cross-age facial features to obtain identity coding information to be queried; Input the identity code information to be queried into the social security personnel information database, and arrange the identity information within the matching degree threshold range according to the identity information matching degree; According to the arrangement order of the identity information, the social security identity information to be determined is displayed in sequence; Based on the manually confirmed social security identity information, social security identity information matching the medical person is obtained.

10. A face recognition system based on cross-age features, characterized in that: The system comprises: The texture recognition module is used to mark the dermis texture features and the variable epidermis features of the collected initial facial image of the medical personnel to determine the facial dermis feature image; dynamically peel off the variable epidermis features in the facial dermis feature image, and perform physical optical constraints on the dermis texture features to obtain an optimized facial image; A facial topology recognition module is used to extract facial feature points under the topological structure and shape of the optimized facial image to obtain facial deformation features; A facial growth prediction module, used for performing diffusion processing of age attribute editing on the optimized facial image to determine facial growth constraint features; A feature fusion module, used for performing attention fusion processing on the facial deformation feature and the facial growth constraint feature, and outputting facial key features; The identity information query module is used to perform feature dimensionality reduction and restoration processing on the facial key features related to the real facial image to obtain cross-age facial features; and perform identity information query processing on the cross-age facial features to determine the social security identity information of the medical personnel.

Images