A non-sensing identity verification and closed-loop management method and system for high-security scenarios

Abstract

The application relates to the technical field of palm vein security identification, and discloses a high-security-scene-oriented non-sensing identity verification and closed-loop management method and system, wherein the high-security-scene-oriented non-sensing identity verification and closed-loop management method comprises the following steps: collecting a user palm vein image through a near-infrared imaging device, extracting a blood vessel skeleton of each part by using a Frangi blood vessel enhancement filter and a morphological thinning algorithm, and constructing an individual vein topological atlas including bifurcation point coordinates, blood vessel curvatures and relative topological connection relationships based on the blood vessel skeleton. The user does not need to actively cooperate, does not need to pause and does not need a specific gesture, and the identity verification is automatically completed in a natural operation process, the verification process is embedded in a business flow, operation is verification, work efficiency is significantly improved, and identity inference is still performed through a forearm or upper arm vein, and the palm is identified by means of a graph neural network cross-region mapping model.

Description

Technical Field

[0001] This invention relates to the field of palm vein security recognition technology, specifically to a method and system for contactless identity verification and closed-loop management in high-security scenarios. Background Technology

[0002] A palm vein recognition system is an access control system that uses palm vein recognition technology. First, a vein recognition device obtains an individual's palm vein distribution map. Feature values ​​are extracted from this map using a specialized comparison algorithm. Infrared CCD cameras capture images of veins in the fingers, palms, and backs of the hands, storing these digital images and feature values ​​in a computer system. During vein comparison, a vein image is captured in real-time, feature values ​​are extracted, and advanced filtering, image binarization, and thinning techniques are used to extract features from the digital image. These features are then compared with the vein feature values ​​stored in the host computer. A complex matching algorithm is employed to match the vein features, thereby identifying and confirming the individual's identity.

[0003] However, in existing critical business processes, identity verification often exists as an independent step, which has the following drawbacks: identity verification is disconnected from business operations, making it impossible to guarantee that the operator is also the authenticator; the verification process requires active cooperation, affecting efficiency and making it easy to bypass; there is a lack of continuous identity monitoring throughout the entire operation, posing a risk of mid-process replacement; and log records only contain timestamps, lacking biometric-level audit evidence. Furthermore, the use of a one-time verification, long-term validity model makes it difficult to cope with high-level internal threats and process tampering attacks. Summary of the Invention

[0004] This invention provides a method and system for seamless identity verification and closed-loop management in high-security scenarios, which has the beneficial effects of supporting reuse in multiple scenarios and seamless verification.

[0005] This invention provides the following technical solution: a method and system for contactless identity verification and closed-loop management in high-security scenarios, comprising the following steps:

[0006] Images of veins in the user's palm are acquired using a near-infrared imaging device, and the vascular skeleton of each part is extracted using the Frangi vascular enhancement filter and morphological refinement algorithm.

[0007] Based on the aforementioned vascular skeleton, an individual vein topology map is constructed, including bifurcation point coordinates, vascular curvature, and relative topological connectivity relationships.

[0008] By jointly modeling the individual vein topology map using a graph neural network and learning a cross-regional mapping function, the vein features of any location can be used to infer the corresponding palm vein distribution features.

[0009] A lightweight hand key point detection model is run in real time to dynamically track the spatial position, orientation, and unfolding state of the user's hand.

[0010] When a palm is detected, at least one vein path is randomly selected from the pre-generated encrypted vein index pool in the local secure storage area. The region of interest is dynamically calculated based on the key points of the hand, and rotation and tilt of the region of interest are corrected by affine transformation.

[0011] Multiple frames of near-infrared images are continuously acquired within a predetermined time window. Feature extraction is performed on at least one selected vein segment in each frame, and the final verification features are generated based on a weighted fusion of image sharpness and blood flow signal intensity.

[0012] The final verification feature is compared with the corresponding encrypted template during registration using homomorphic encryption technology in the ciphertext domain. If the matching score exceeds the dynamic threshold, the identity verification is deemed successful.

[0013] As an optional solution to the non-intrusive identity verification and closed-loop management method and system for high-security scenarios described in this invention, the near-infrared imaging device is also used to acquire vein images of the user's forearm and upper arm.

[0014] If the palm is covered but the forearm or upper arm is partially exposed, the GNN cross-region mapping model is invoked to infer the corresponding palm vein feature distribution from the forearm or upper arm vein skeleton, and compare it with the encrypted palm template in the TEE to output a confidence score.

[0015] Based on the currently visible body parts, operational risk level, and confidence score, the adaptive strategy dynamically selects the verification mode and decides whether to allow the operation.

[0016] As an optional solution to the contactless identity verification and closed-loop management method and system for high-security scenarios described in this invention, the method for constructing an individual vein topology map includes:

[0017] Non-uniform illumination correction and background suppression are performed on the acquired raw vein images;

[0018] The linear structure was highlighted by the Frangi multi-scale vessel enhancement filter, and then Otsu threshold segmentation and morphological refinement were used to obtain a single-pixel wide vessel skeleton.

[0019] Key topological elements are extracted from the single-pixel wide blood vessel skeleton. These key topological elements include the origin of the main vein, bifurcation nodes, terminal endpoints, vessel segment length, and local curvature.

[0020] Encode the key topological elements into graph structure data;

[0021] A cross-part mapping model is trained using a graph convolutional network, with the input being a graph structure of the forearm and the output being a latent representation of a graph structure of the palm.

[0022] As an optional solution to the contactless identity verification and closed-loop management method and system for high-security scenarios described in this invention, it further includes random sampling encrypted matching, comprising the following steps:

[0023] During the registration phase, the complete palm vein image is segmented into several independent vein paths, and a local feature vector is generated for each path, and a unique encrypted index ID is assigned.

[0024] After the encrypted index ID is bound to the feature vector, it is encrypted using SM4 and stored in the local index pool;

[0025] During each verification, a one-time random number Nonce is generated, and at least one index is pseudo-randomly selected from the index pool based on the Nonce;

[0026] Among them, the Paillier homomorphic encryption scheme is used to calculate the cosine similarity in the ciphertext domain;

[0027] If the average matching score after multi-frame fusion is greater than or equal to the dynamic threshold, then the verification is successful.

[0028] As an optional solution to the contactless identity verification and closed-loop management method and system for high-security scenarios described in this invention, the adaptive strategy scheduling includes:

[0029] Random sampling of palm veins is applicable to all procedures when the palm is visible and the ROI quality score is ≥0.8.

[0030] For single-site inference, if only the forearm or upper arm is visible and the inference confidence level is ≥0.7, only low-risk operations are allowed;

[0031] Multi-site fusion inference shows that the forearm and upper arm are visible simultaneously, with a fusion confidence of ≥0.85, allowing for medium- to high-risk operations;

[0032] Emergency authorization: If full body covering or verification fails, two authorized personnel must complete the dual-person approval through a backup channel using a combination of voiceprint and temporary code.

[0033] The probability of the same vein path being selected repeatedly within a preset time decreases exponentially.

[0034] As an optional solution to the non-intrusive identity verification and closed-loop management method and system for high-security scenarios described in this invention, it further includes the recording of structured logs, which include task session ID, operation type code, risk level, verification mode identifier, encrypted index ID list of selected veins, inference confidence, live blood flow signal strength, device fingerprint, network status and timestamp.

[0035] All fields are concatenated and hashed using the national cryptographic standard SM3 to generate a digest, which is then signed using the device's private key via the SM2 algorithm.

[0036] The signed log is written to a local read-only log library and synchronized to a centralized auditing platform after connecting to the network.

[0037] As an optional solution to the contactless identity verification and closed-loop management method and system for high-security scenarios described in this invention, it further includes abnormal behavior linkage, including:

[0038] Real-time collection of user operation behavior sequences, including interface click coordinates, operation interval time, menu jump path and input speed;

[0039] A Gaussian mixture model is constructed based on historical normal behavior baselines to calculate the current behavior abnormality score;

[0040] If the abnormal score exceeds the threshold, at least one of the following responses will still be triggered even if authentication is successful;

[0041] The first response is to forcibly increase the K value of this verification;

[0042] The second response requires a second confirmation.

[0043] The third response is to suspend operations and notify the security administrator.

[0044] A system applying a seamless identity verification and closed-loop management method for high-security scenarios includes:

[0045] An image acquisition module is used to acquire vein images of a user's palm, forearm, or upper arm using a near-infrared imaging device.

[0046] The processing module includes a Frangi vessel enhancement filter, a morphological thinning algorithm, and a graph neural network model. The processing module is used to extract the vascular skeleton, construct an individual vein topology map, and perform joint modeling on the vein topology map.

[0047] The dynamic tracking module includes a camera array and a lightweight hand key point detection model, which is used to track the spatial position, orientation and unfolding state of the user's hand in real time.

[0048] An encrypted storage module is used to securely store a pre-generated encrypted vein index pool and dynamically calculate the region of interest based on key points of the hand during authentication.

[0049] The feature extraction and fusion module is used to continuously acquire multiple frames of near-infrared images within a predetermined time window, and extract and fuse features of selected vein segments.

[0050] A homomorphic encryption comparison module, which uses homomorphic encryption technology to compare the final verification features with the corresponding encrypted template at the time of registration within the ciphertext domain;

[0051] An adaptive strategy scheduling module dynamically selects a verification mode and decides whether to allow the operation based on the currently visible body part, the operation risk level, and the confidence score.

[0052] The structured log module is used to record task session ID, operation type code and risk level information, and to generate a digest using the national cryptographic SM3 hash and then sign and write it into the local read-only log library.

[0053] The abnormal behavior linkage module is used to collect user operation behavior sequences in real time, calculate the current behavior abnormal score based on the historical normal behavior baseline, and trigger corresponding response measures.

[0054] The present invention has the following beneficial effects:

[0055] 1. This seamless identity verification and closed-loop management method and system for high-security scenarios allows users to automatically complete identity verification during natural operation without active cooperation, pauses, or specific gestures. The verification process is embedded in the business flow, making operation and verification instantaneous, significantly improving work efficiency and completely eliminating the interference of traditional biometrics on work rhythm. Even if users wear double gloves, long-sleeved protective clothing, or even fully enclosed PPE equipment, identity can still be inferred through forearm or upper arm veins. With the help of graph neural network cross-regional mapping model, it realizes palm recognition by arm, solving the fundamental pain point that traditional palm vein or fingerprint recognition is completely ineffective in occluded scenarios. It is suitable for various scenarios to recognize triple liveness protection.

[0056] 2. This seamless identity verification and closed-loop management method and system for high-security scenarios monitors business events in real time and performs backtracking verification from task initiation to critical operations, ensuring that every operation is executed by a legitimate entity. It also supports dual-person collaborative authorization and linkage blocking of abnormal behavior, effectively preventing high-risk behaviors such as internal personnel impersonation, account sharing, and mid-process replacement. The verification mode is dynamically switched according to the visible location, operational risk, and abnormal behavior. Furthermore, a time decay random sampling mechanism is introduced to avoid targeted attacks on repeated paths. High-risk operations automatically raise the security threshold, while low-risk scenarios maintain a smooth experience, achieving elastic security.

[0057] 3. This seamless identity verification and closed-loop management method and system for high-security scenarios intelligently determines whether to allow the current operation based on the verification results of the selected mode. This truly achieves seamless verification, allowing users to automatically complete identity verification during natural operations without pauses or specific gestures. This significantly improves work efficiency and eliminates the interference of traditional biometrics on business processes. It possesses extremely strong liveness detection capabilities, combining blood flow signal detection, dynamic random sampling, and multi-frame temporal analysis. Attackers find it difficult to simultaneously forge vein structures, blood flow dynamics, and cross-site topological relationships, making the deception success rate close to zero. It not only completes identity recognition but also dynamically adjusts security strategies based on operational risks and intervenes in abnormal situations, effectively preventing internal risks such as account misuse and mid-operation replacement. Attached Figure Description

[0058] Figure 1 This is a flowchart of the vein topology map construction process of the present invention.

[0059] Figure 2 This is a flowchart of the abnormal behavior linkage response of the present invention.

[0060] Figure 3 This is a schematic diagram of the blood vessels on the palmar side of the hand according to the present invention.

[0061] Figure 4 This is a texture map of the palm surface of the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1

[0064] Please see Figures 1-4 One method for seamless identity verification and closed-loop management in high-security scenarios includes the following steps:

[0065] S1. Acquire images of veins in the user's palm using a near-infrared imaging device, and extract the vascular skeleton of each part using the Frangi vascular enhancement filter and morphological refinement algorithm.

[0066] Specifically, a near-infrared CMOS camera with a center wavelength of 850nm is used. This wavelength can penetrate the epidermis (1-3mm) and is strongly absorbed by hemoglobin, while the surrounding tissue reflects light strongly, thus forming a high-contrast vein image. Deoxyhemoglobin in the blood has a strong absorption of 850nm light, making the vein area appear dark and the background tissue appear bright, naturally creating a negative effect.

[0067] Image preprocessing employs non-uniform illumination correction, utilizing the Retinex algorithm or homomorphic filtering to eliminate ambient light interference. Background suppression uses Gaussian difference or adaptive thresholding to remove noise outside the hand contour.

[0068] The Frangi vessel enhancement filter formula is as follows:

[0069] ;

[0070] in, For bulkness, For structural strength, and These are Hessian eigenvalues. and These are adjustable parameters that control response sensitivity.

[0071] The vascular skeletonization uses the Zhang-Suen or Guo-Hall thinning algorithm to transform the enhanced vascular region into a single-pixel-wide, connected and unbroken skeleton map, preserving the topological structure.

[0072] S2. Based on the vascular skeleton, construct an individual vein topology map including bifurcation point coordinates, vascular curvature, and relative topological connection relationships.

[0073] Specifically, key element extraction:

[0074] Endpoint, a pixel with a degree of 1 (the end of a blood vessel).

[0075] Fork point, a pixel with a degree ≥ 3 (the main branch branches off from the main branch);

[0076] A blood vessel segment is a curved path connecting two key points.

[0077] Geometric feature calculation:

[0078] Curvature is determined by fitting a spline curve along the vessel segment and calculating the local curvature. ;

[0079] in, As the unit tangent vector, For arc length parameters, The rate of change of the unit tangent vector with respect to the arc length;

[0080] The length and direction are Euclidean distance and principal axis direction angle;

[0081] The relative position is with the palm of the hand as the origin, and the polar coordinates of each node are recorded.

[0082] Graph structure encoding models the venous system as an undirected graph G=(V,E):

[0083] Node V is a branch point or endpoint, with attributes = (x, y, type, mean curvature);

[0084] Edge E is a blood vessel segment with attributes = (length, mean curvature, direction vector);

[0085] This outputs a structured venous topology map, possessing individual uniqueness and physiological stability.

[0086] S3. Jointly model the individual vein topology map using a graph neural network, learn the cross-regional mapping function, so that the vein features of any location can be used to infer the corresponding palm vein distribution features.

[0087] Specifically, a graph convolutional network or graph attention network is used, with the input being the forearm / upper arm vein map Garm and the output being the latent embedding of the palm vein map Gpalm.

[0088] Training is as follows:

[0089] Data: three sets of paired images of the same user's palm, forearm, and upper arm;

[0090] Loss functions include reconstruction loss, such as MSE on node features, and topology consistency loss.

[0091] The goal is to learn mappings .

[0092] In occluded scenarios, with only a forearm image as input, the model outputs an inferred palm feature vector for subsequent comparison.

[0093] S4. Runs a lightweight hand key point detection model in real time to dynamically track the spatial position, orientation, and unfolding state of the user's hand.

[0094] S5. When a palm is detected, at least one vein path is randomly selected from the pre-generated encrypted vein index pool in the local secure storage area. The region of interest is dynamically calculated based on the key points of the hand, and rotation and tilt are corrected by affine transformation.

[0095] Specifically, the ROI is dynamically calculated by taking the palm center (joint 0) as the center and determining the direction of the main axis based on the line connecting the finger roots (nodes 5–17) to generate an elliptical ROI.

[0096] Calculate the rotation angle and scaling factor between the current palm axis and the standard template, construct an affine matrix, map the ROI to the standard coordinate system, and eliminate posture differences.

[0097] A random sampling mechanism is used to segment the palm veins into N independent paths (N≥20) during registration.

[0098] Each path generates a 128-dimensional feature vector and a unique ID;

[0099] The ID and feature are encrypted with SM4 and stored in the index pool within the TEE; during verification, a Nonce is generated, and K records are selected pseudo-randomly (K=3-5).

[0100] S6. Within a predetermined time window, continuously acquire multiple frames of near-infrared images, extract features from at least one selected vein segment in each frame, and generate final verification features based on weighted fusion of image clarity and blood flow signal intensity.

[0101] Multi-frame fusion strategies include:

[0102] Time window 200–500ms (covering one natural pause);

[0103] Sharpness rating, based on Laplacian variance;

[0104] Blood flow signal intensity, detected by inter-frame difference pulsation, must exceed a threshold to ensure liveness;

[0105] S7. The final verification feature is compared with the corresponding encryption template during registration using homomorphic encryption technology in the ciphertext domain. If the matching score exceeds the dynamic threshold, the identity verification is deemed successful.

[0106] Specifically, the close matching scheme uses Paillier homomorphic encryption;

[0107] By incorporating security protocols, cosine similarity can be calculated without decryption.

[0108] The base threshold is 0.80. During high-risk operations, the threshold is raised to 0.95. When abnormal behavior is triggered, the threshold is temporarily increased.

[0109] In summary, users do not need to actively cooperate, pause, or use specific gestures. Identity verification is automatically completed during natural operation. The verification process is embedded in the business flow, and operation is verification, which significantly improves work efficiency and completely eliminates the interference of traditional biometrics on the work rhythm. Even if users wear double gloves, long-sleeved protective clothing, or even fully enclosed PPE equipment, identity can still be inferred through the veins in the forearm or upper arm. With the help of the graph neural network cross-regional mapping model, it can realize palm recognition by arm, which solves the fundamental pain point that traditional palm vein or fingerprints are completely ineffective in occluded scenarios. It is suitable for various scenarios to recognize triple liveness protection.

[0110] One method is near-infrared blood flow signal detection and pulsation analysis;

[0111] Secondly, dynamic random sampling leads to unpredictable venous pathways;

[0112] Thirdly, it uses multi-frame temporal fusion, rejecting static photo or video playback;

[0113] In summary, the success rate of deception is close to zero. Attackers cannot predict the sampling location and must simultaneously forge blood flow dynamics and vein topology from multiple locations. Furthermore, from task initiation to critical operations, business events are monitored in real time and backtracked for verification, ensuring that every step is executed by a legitimate entity. It also supports dual-person collaborative authorization and abnormal behavior linkage blocking, effectively preventing high-risk behaviors such as internal personnel impersonation, account sharing, and mid-process replacement. Verification modes (palm vein, single-site inference, multi-site fusion, and dual-person emergency response) are dynamically switched based on visible locations, operational risks, and abnormal behavior. A time-decay random sampling mechanism is introduced to avoid targeted attacks on repetitive paths. High-risk operations automatically raise the security threshold, while low-risk scenarios maintain a smooth experience, achieving resilient security.

[0114] Example 2

[0115] This embodiment is an improvement upon embodiment 1. For details, please refer to [link / reference]. Figures 1-4 The near-infrared imaging device is also used to acquire vein images of the user's forearm and upper arm;

[0116] If the palm is covered but the forearm or upper arm is partially exposed, the GNN cross-region mapping model is invoked to infer the corresponding palm vein feature distribution from the forearm or upper arm vein skeleton, and compare it with the encrypted palm template in the TEE to output a confidence score.

[0117] Based on the currently visible body parts, operational risk level, and confidence score, the adaptive strategy dynamically selects the verification mode and decides whether to allow the operation.

[0118] A lightweight hand keypoint detection model, such as the improved MediaPipe Hands, is used to detect hand occlusion. For example, if the palm keypoint loss rate is too high or the image signal-to-noise ratio is too low, an occlusion response mechanism is immediately activated. At this point, the near-infrared image quality of the forearm and / or upper arm is analyzed to determine which areas are partially exposed and usable, and the vascular skeleton is extracted from them. This process uses the Frangi vascular enhancement filter combined with a morphological thinning algorithm to generate a single-pixel-wide, topology-preserving vein centerline map. Subsequently, a pre-trained graph neural network cross-region mapping model is invoked. This model, based on multi-site vein pairing data collected during the user registration phase, learns the physiological correlation of vein topology between the palm, forearm, and upper arm.

[0119] During the inference phase, the model takes the vein structure extracted from the forearm or upper arm as input, outputs an inferred palm vein feature vector, and simultaneously generates a confidence score between 0.0 and 1.0. This score reflects the reliability of the inference result and is determined by the model's attention entropy value and reconstruction error.

[0120] To ensure biometric privacy and security, the inferred palm feature vector is sent to a Trusted Execution Environment (TEE). Inside the TEE, the system decrypts the encrypted palm template stored during user registration and calculates the cosine similarity between the two as the final identity matching score. The entire comparison process is completed in a hardware-isolated secure environment, where the operating system and other applications cannot access the original biometric data, fundamentally eliminating the risk of data leakage.

[0121] Finally, based on three key factors—the currently visible body part (e.g., forearm only, forearm plus upper arm, full occlusion), the risk level of the operation to be performed, and the confidence score of the GNN inference output—the adaptive policy scheduling module dynamically selects the most suitable verification mode.

[0122] It also includes a random sampling mode for palm veins (when the palm is visible), a single-site vein inference mode (visible only in the forearm and with a confidence level ≥ 0.7, allowing only low-risk operations), a multi-site fusion inference mode (visible in both the forearm and upper arm and with a fusion confidence level ≥ 0.85, supporting medium- and high-risk operations), and a dual-person emergency authorization mode (when fully covered or verification fails, two authorized personnel are required to complete the approval through voiceprint and temporary code).

[0123] Based on the verification results of the selected mode, the system intelligently decides whether to allow the current operation. This truly achieves seamless verification, allowing users to automatically complete identity verification through natural operation without pauses or specific gestures, significantly improving work efficiency and eliminating the interference of traditional biometrics on business processes. It breaks through to support fully occluded scenarios, enabling the inference of hand identity from forearm or upper arm veins, solving the industry problem of traditional biometrics completely failing in extreme environments such as operating rooms, P4 laboratories, and hazardous chemical operations. It possesses extremely strong anti-spoofing capabilities for liveness detection. Combining blood flow signal detection, dynamic random sampling, and multi-frame temporal analysis, attackers find it difficult to simultaneously forge vein structures, blood flow dynamics, and cross-site topological relationships, resulting in a near-zero success rate for deception. It not only completes identity verification but also dynamically adjusts security strategies based on operational risks and intervenes in conjunction with blocking mechanisms in abnormal situations, effectively preventing internal risks such as account misuse and mid-operation replacement. Furthermore, the original vein images are not stored or uploaded; feature comparison is completed in the TEE (Tracking Equipment Environment), and audit logs are signed with national cryptographic algorithms and are irreversibly reproducible, fully complying with the strictest privacy compliance requirements both domestically and internationally, including the Personal Information Protection Law.

[0124] Example 3

[0125] This embodiment is an improvement upon embodiment 2. For details, please refer to [link / reference]. Figures 1-4 The method for constructing an individual vein topology map includes:

[0126] Non-uniform illumination correction and background suppression are performed on the acquired raw vein images;

[0127] The linear structure was highlighted by the Frangi multi-scale vessel enhancement filter, and then Otsu threshold segmentation and morphological refinement were used to obtain a single-pixel wide vessel skeleton.

[0128] Key topological elements are extracted from the single-pixel wide blood vessel skeleton. These key topological elements include the origin of the main vein, bifurcation nodes, terminal endpoints, vessel segment length, and local curvature.

[0129] Encode the key topological elements into graph structure data;

[0130] A cross-part mapping model is trained using a graph convolutional network, with the input being a graph structure of the forearm and the output being a latent representation of a graph structure of the palm.

[0131] First, the acquired raw vein images undergo non-uniform illumination correction and background suppression. Near-infrared imaging is susceptible to uneven ambient light distribution and reflections from the curved surface of the hand, often resulting in gradual changes in brightness or edge attenuation. Therefore, the system employs an illumination correction algorithm based on homomorphic filtering or Retinex theory, decomposing the image into illuminance and reflectance components to suppress low-frequency illuminance variations and enhance high-frequency vein structures.

[0132] Simultaneously, background noise outside the hand contour is effectively removed by using Gaussian difference or adaptive morphological top-hat transformation, preserving clear vein regions. Secondly, the Frangi multi-scale vessel enhancement filter highlights linear structures in the image. Based on eigenvalue analysis of the Hessian matrix, the Frangi filter calculates local structural responses at multiple scales (typically σ ranging from 1.0 to 3.0 pixels), significantly enhancing slender, continuous vessel signals while suppressing speckle, edge, and texture interference. In the enhanced image, veins appear as high-contrast bright lines, facilitating subsequent segmentation.

[0133] Next, Otsu adaptive thresholding segmentation is used to binarize the enhanced image. The Otsu method automatically determines the optimal segmentation threshold by maximizing the inter-class variance, without the need for manual parameter setting, and is suitable for vein images under different lighting conditions and individual differences.

[0134] Subsequently, morphological thinning operations (such as the Zhang-Suen or Guo-Hall algorithms) are performed on the binary image to gradually peel away boundary pixels until a single-pixel-wide, 8-connected, and topology-preserving vascular skeleton is obtained. This skeleton accurately reflects the central direction and branching structure of the vein and is the basis for constructing the topological map.

[0135] Based on this, key topological elements are extracted from the single-pixel wide blood vessel skeleton.

[0136] Elements include:

[0137] The main vein originates, usually near the proximal end of the wrist, serving as the root node of the venous network;

[0138] Bifurcation node: The intersection of two or more branches of a blood vessel, with a degree ≥ 3;

[0139] The distal end point, the location where the blood vessel terminates, is measured in degrees of 1.

[0140] The length of the blood vessel segment, the Euclidean distance between adjacent key points, or the actual path length along the skeleton;

[0141] Local curvature is calculated by fitting a spline curve to the vascular segment and determining its curvature value at each point.

[0142] Through multi-level image preprocessing and skeletonization, uneven illumination, noise interference, and individual differences are effectively overcome, ensuring the accuracy and stability of vein topology extraction. Encoding vein information into a graph structure not only preserves geometric features but also explicitly models physiological topological relationships, offering greater interpretability and resistance to deformation compared to traditional template matching or deep feature vectors. Furthermore, a cross-regional mapping model based on graph neural network learning enables the inversion of hand identity from forearm veins, breaking through the dependence of traditional palm vein recognition on complete palm exposure and greatly expanding application scenarios. Moreover, since the original vein image is destroyed immediately after feature extraction, retaining only encrypted graph structure parameters, the risk of biometric information leakage is fundamentally reduced. The constructed individual vein topology map supports subsequent random path sampling, multi-frame fusion comparison, and confidence assessment, enabling highly secure, seamless, and closed-loop identity management.

[0143] Example 4

[0144] This embodiment is an improvement upon embodiment 3. For details, please refer to [link / reference]. Figures 1-4 It also includes random sampling encrypted matching, including the following steps:

[0145] During the registration phase, the complete palm vein image is segmented into several independent vein paths, and a local feature vector is generated for each path, and a unique encrypted index ID is assigned.

[0146] After the encrypted index ID is bound to the feature vector, it is encrypted using SM4 and stored in the local index pool;

[0147] During each verification, a one-time random number Nonce is generated, and at least one index is pseudo-randomly selected from the index pool based on the Nonce;

[0148] Among them, the Paillier homomorphic encryption scheme is used to calculate the cosine similarity in the ciphertext domain;

[0149] If the average matching score after multi-frame fusion is greater than or equal to the dynamic threshold, then the verification is successful.

[0150] Specifically, random sampling encryption matching methods include:

[0151] During the registration phase, the acquired complete palm vein image is processed and segmented into several independent vein paths. Each path generates a local feature vector using the aforementioned method for constructing an individual vein topology map; these feature vectors represent the uniqueness of the vein path.

[0152] Assign a unique encrypted index ID to the feature vector corresponding to each vein path. This index ID is used to identify and retrieve a specific vein path and its feature vector.

[0153] SM4 encrypted storage binds each encrypted index ID to its corresponding local feature vector and encrypts it using the national commercial cryptography standard SM4 algorithm. The encrypted data is securely stored in a local index pool, ensuring that even if the data is leaked, it cannot be easily decrypted to obtain the original information.

[0154] During the verification phase, a one-time random number (Nonce) is generated for each verification. The Nonce guarantees the uniqueness and unpredictability of each verification process.

[0155] Pseudo-random index selection uses a pseudo-random algorithm to select at least one index from the index pool based on the generated nonce. This selection method increases the difficulty for attackers to predict or replay attacks.

[0156] For similarity calculation under homomorphic encryption, the Paillier homomorphic encryption scheme is used to compare the similarity between newly collected samples and registered samples without exposing feature vectors. This scheme allows cosine similarity to be calculated directly in the ciphertext domain without decrypting the original data, thus protecting user privacy while achieving efficient authentication.

[0157] Multi-frame fusion and dynamic threshold judgment, in practical applications, usually combine data collected from multiple times to improve verification accuracy. The matching scores between different frames are fused and averaged. If this average score is greater than or equal to a preset dynamic threshold, the verification is considered successful.

[0158] By using SM4 encryption technology and Paillier homomorphic encryption scheme, the security of data storage is guaranteed, and feature comparison under privacy protection is achieved. The Nonce mechanism and pseudo-random index selection strategy effectively resist replay attacks, improve overall security, support multi-frame fusion processing, and improve the accuracy and robustness of identity verification. It is especially suitable for biometric application scenarios in complex environments.

[0159] Example 5

[0160] This embodiment is an improvement upon embodiment 4. For details, please refer to [link / reference]. Figures 1-4 It also includes abnormal behavior linkages, including:

[0161] Real-time collection of user operation behavior sequences, including interface click coordinates, operation interval time, menu jump path and input speed;

[0162] A Gaussian mixture model is constructed based on historical normal behavior baselines to calculate the current behavior abnormality score;

[0163] If the abnormal score exceeds the threshold, at least one of the following responses will still be triggered even if authentication is successful;

[0164] The first response is to forcibly increase the K value of this verification;

[0165] The second response requires a second confirmation.

[0166] The third response is to suspend operations and notify the security administrator.

[0167] Specifically, multi-dimensional behavioral sequence data is collected in real time during user operations, including click or touch coordinates on the graphical user interface, time intervals between adjacent operations, menu navigation paths, text input speed, and mouse movement trajectories. This behavioral data is indexed by timestamps to form a structured operation log stream.

[0168] Based on users' historical normal operation data, the system offline constructs a personalized behavior baseline model. This model uses a Gaussian Mixture Model (GMM). GMM is a probability density estimation method that can effectively fit complex, non-unimodal behavioral distributions. Specifically, each operation sequence is mapped to a high-dimensional feature vector, including factors such as average click offset, operation rhythm variance, and frequency of common paths. Then, the GMM is trained using the Expectation-Maximization (EM) algorithm to obtain the weights, mean vectors, and covariance matrices of several Gaussian components, which together constitute the user's normal behavior fingerprint.

[0169] During the real-time validation phase, the current operation sequence is also converted into a feature vector and input into the trained GMM. Its log-likelihood probability is calculated using the following formula:

[0170] ;

[0171] in, For the observed sample vector, The number of Gaussian components, The weight of the k-th Gaussian component. Let be the probability density function of the Kth Gaussian distribution. It is the natural logarithm, which facilitates numerical calculation and optimization;

[0172] in, This is a GMM parameter; the lower the value, the more the current behavior deviates from the normal pattern. It is converted into an anomaly score using the following formula:

[0173]

[0174] And normalize to the interval [0, 1], where, For the observed sample vector, The parameters of the already trained model. To observe samples given model parameters.

[0175] If the abnormal score exceeds a preset threshold, which can be dynamically adjusted according to the operational risk level, the current session is determined to have an abnormal behavior risk. It should be noted that even if previous vein authentication has been successfully completed, a security response mechanism will still be triggered to prevent advanced threats such as account theft, abnormal operations by internal personnel, or coerced execution.

[0176] The triggered response measures include at least one of the following:

[0177] The first response is to forcibly increase the K value of this verification. K is the number of vein paths selected in the random sampling encryption matching phase. The default K=3, and it can be increased to K=5 or higher, thereby improving the confidence and anti-spoofing ability of biometric comparison and increasing the difficulty for attackers to bypass the verification.

[0178] The second response requires secondary confirmation, popping up a secondary authentication interface and requiring the user to confirm through alternative methods (such as entering a dynamic password, completing a gesture password, or showing their palm again), thus forming two-factor authentication and blocking automated scripts or remote control.

[0179] The third response involves suspending the current operation and notifying the security administrator to immediately freeze the current transaction (e.g., halting the drug dispensing process). Simultaneously, an alert is pushed to the security audit platform, including the session ID, details of the abnormal behavior, operation context, and biometric verification logs, for manual investigation by the administrator. These response strategies can be combined and enabled by the security policy engine based on risk levels; for example, a low anomaly score might only increase the K-value, while a high score would directly suspend the operation.

[0180] In summary, this approach achieves continuous authentication at the behavioral level, overcoming the limitations of traditional one-time verification and end-to-end trust. It dynamically assesses user credibility throughout the entire operational lifecycle, effectively addressing the risk of lateral movement after credential leakage, and accurately identifying abnormal operations. The GMM model can capture subtle differences in individual operating habits and is highly sensitive to scenarios such as impersonation and coerced operations. Deeply integrated with biometrics, it merges vein recognition and behavioral pattern verification to construct a multi-factor, multi-layered security defense. Furthermore, it supports a gradient response from mild intervention to increasing the K-value to strong blocking, balancing security and business continuity. Meeting compliance audit requirements, all abnormal events are recorded in structured logs and synchronized to the central audit platform, providing a complete chain of evidence for post-event tracing and liability determination.

[0181] Example 6

[0182] This example also provides a system for a contactless identity verification and closed-loop management method for high-security scenarios, characterized by including:

[0183] An image acquisition module is used to acquire vein images of a user's palm, forearm, or upper arm using a near-infrared imaging device.

[0184] The processing module includes a Frangi vessel enhancement filter, a morphological thinning algorithm, and a graph neural network model. The processing module is used to extract the vascular skeleton, construct an individual vein topology map, and perform joint modeling on the vein topology map.

[0185] The dynamic tracking module includes a camera array and a lightweight hand key point detection model, which is used to track the spatial position, orientation and unfolding state of the user's hand in real time.

[0186] An encrypted storage module is used to securely store a pre-generated encrypted vein index pool and dynamically calculate the region of interest based on key points of the hand during authentication.

[0187] The feature extraction and fusion module is used to continuously acquire multiple frames of near-infrared images within a predetermined time window, and extract and fuse features of selected vein segments.

[0188] A homomorphic encryption comparison module, which uses homomorphic encryption technology to compare the final verification features with the corresponding encrypted template at the time of registration within the ciphertext domain;

[0189] An adaptive strategy scheduling module dynamically selects a verification mode and decides whether to allow the operation based on the currently visible body part, the operation risk level, and the confidence score.

[0190] The structured log module is used to record task session ID, operation type code and risk level information, and to generate a digest using the national cryptographic SM3 hash and then sign and write it into the local read-only log library.

[0191] The abnormal behavior linkage module is used to collect user operation behavior sequences in real time, calculate the current behavior abnormal score based on the historical normal behavior baseline, and trigger corresponding response measures.

[0192] This specification also provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the multiple steps described in the above embodiments. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0193] This specification also provides a computer program product, including a computer program that, when executed by a processor, implements the multiple steps described in the above embodiments.

[0194] Where there is no conflict, the technical features in this embodiment and implementation scheme can be combined arbitrarily.

[0195] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes multiple computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating multiple available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital versatile discs (DVDs)), or semiconductor media (e.g., solid state disks (SSDs)).

[0196] When implemented through hardware or firmware, the aforementioned method flow is programmed into the hardware circuit to obtain the corresponding hardware circuit structure and achieve the corresponding function. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit, whose logic function is determined by the user programming the device. Designers can program a digital system onto a PLD themselves, eliminating the need for chip manufacturers to design and fabricate dedicated integrated circuit chips. Furthermore, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, similar to the software compiler used in program development. The original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There is not just one HDL, but many. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of the aforementioned hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0197] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0198] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for seamless identity verification and closed-loop management in high-security scenarios, characterized in that, Includes the following steps: Images of veins in the user's palm are acquired using a near-infrared imaging device, and the vascular skeleton of each part is extracted using the Frangi vascular enhancement filter and morphological refinement algorithm. Based on the aforementioned vascular skeleton, an individual vein topology map is constructed, including bifurcation point coordinates, vascular curvature, and relative topological connectivity relationships. By jointly modeling the individual vein topology map using a graph neural network and learning a cross-regional mapping function, the vein features of any location can be used to infer the corresponding palm vein distribution features. A lightweight hand key point detection model is run in real time to dynamically track the spatial position, orientation, and unfolding state of the user's hand. When a palm is detected, at least one vein path is randomly selected from the pre-generated encrypted vein index pool in the local secure storage area. The region of interest is dynamically calculated based on the key points of the hand, and rotation and tilt of the region of interest are corrected by affine transformation. Multiple frames of near-infrared images are continuously acquired within a predetermined time window. Feature extraction is performed on at least one selected vein segment in each frame, and the final verification features are generated based on a weighted fusion of image sharpness and blood flow signal intensity. The final verification feature is compared with the corresponding encrypted template during registration using homomorphic encryption technology in the ciphertext domain. If the matching score exceeds the dynamic threshold, the identity verification is deemed successful. The near-infrared imaging device is also used to acquire vein images of the user's forearm and upper arm; If the palm is obscured but the forearm or upper arm is partially exposed, the GNN cross-region mapping model is invoked to infer the corresponding palm vein feature distribution from the forearm or upper arm vein skeleton, and compare it with the encrypted palm template in the TEE to output a confidence score. Based on the currently visible body parts, operational risk level, and confidence score, the adaptive strategy dynamically selects the verification mode and decides whether to allow the operation.

2. The seamless identity verification and closed-loop management method for high-security scenarios according to claim 1, characterized in that, The method for constructing an individual vein topology map includes: Non-uniform illumination correction and background suppression are performed on the acquired raw vein images; The linear structure was highlighted by the Frangi multi-scale vessel enhancement filter, and then Otsu threshold segmentation and morphological refinement were used to obtain a single-pixel wide vessel skeleton. Key topological elements are extracted from the single-pixel wide blood vessel skeleton. These key topological elements include the origin of the main vein, bifurcation nodes, terminal endpoints, vessel segment length, and local curvature. Encode the key topological elements into graph structure data; A cross-part mapping model is trained using a graph convolutional network, with the input being a graph structure of the forearm and the output being a latent representation of a graph structure of the palm.

3. The seamless identity verification and closed-loop management method for high-security scenarios according to claim 2, characterized in that, It also includes random sampling encrypted matching, comprising the following steps: During the registration phase, the complete palm vein image is segmented into several independent vein paths, and a local feature vector is generated for each path, and a unique encrypted index ID is assigned. After the encrypted index ID is bound to the feature vector, it is encrypted using SM4 and stored in the local index pool; During each verification, a one-time random number Nonce is generated, and at least one index is pseudo-randomly selected from the index pool based on the Nonce; Among them, the Paillier homomorphic encryption scheme is used to calculate the cosine similarity in the ciphertext domain; If the average matching score after multi-frame fusion is greater than or equal to the dynamic threshold, then the verification is successful.

4. The seamless identity verification and closed-loop management method for high-security scenarios according to claim 3, characterized in that, The adaptive policy scheduling includes: Random sampling of palm veins is applicable to all procedures when the palm is visible and the ROI quality score is ≥0.

8. For single-site inference, if only the forearm or upper arm is visible and the inference confidence level is ≥0.7, only low-risk operations are allowed; Multi-site fusion inference shows that the forearm and upper arm are visible simultaneously, with a fusion confidence of ≥0.85, allowing for medium- to high-risk operations; Emergency authorization: If full body covering or verification fails, two authorized personnel must complete the dual-person approval through a backup channel using a combination of voiceprint and temporary code. The probability of the same vein path being selected repeatedly within a preset time decreases exponentially.

5. The seamless identity verification and closed-loop management method for high-security scenarios according to claim 4, characterized in that, It also includes recording structured logs, which include task session ID, operation type code, risk level, verification mode identifier, encrypted index ID list of selected veins, inferred confidence level, live blood flow signal strength, device fingerprint, network status and timestamp; All fields are concatenated and hashed using the national cryptographic standard SM3 to generate a digest, which is then signed using the device's private key via the SM2 algorithm. The signed log is written to a local read-only log library and synchronized to a centralized auditing platform after connecting to the network.

6. The seamless identity verification and closed-loop management method for high-security scenarios according to claim 1, characterized in that: This also includes abnormal behavior linkages, including: Real-time collection of user operation behavior sequences, including interface click coordinates, operation interval time, menu jump path and input speed; A Gaussian mixture model is constructed based on historical normal behavior baselines to calculate the current behavior abnormality score; If the abnormal score exceeds the threshold, at least one of the following responses will still be triggered even if authentication is successful; The first response is to forcibly increase the K value of this verification; The second response requires a second confirmation. The third response is to suspend operations and notify the security administrator.

7. A system that applies a seamless identity verification and closed-loop management method for high-security scenarios, characterized in that: include: An image acquisition module is used to acquire vein images of a user's palm, forearm, or upper arm using a near-infrared imaging device. The near-infrared imaging device is also used to acquire vein images of the user's forearm and upper arm; If the palm is obscured but the forearm or upper arm is partially exposed, the GNN cross-region mapping model is invoked to infer the corresponding palm vein feature distribution from the forearm or upper arm vein skeleton, and compare it with the encrypted palm template in the TEE to output a confidence score. Based on the currently visible body parts, operational risk level, and confidence score, the adaptive strategy scheduler dynamically selects the verification mode and decides whether to allow the operation. The processing module includes a Frangi vessel enhancement filter, a morphological thinning algorithm, and a graph neural network model. The processing module is used to extract the vascular skeleton, construct an individual vein topology map, and perform joint modeling on the vein topology map. The dynamic tracking module includes a camera array and a lightweight hand key point detection model, which is used to track the spatial position, orientation and unfolding state of the user's hand in real time. An encrypted storage module is used to securely store a pre-generated encrypted vein index pool and dynamically calculate the region of interest based on key points of the hand during authentication. The feature extraction and fusion module is used to continuously acquire multiple frames of near-infrared images within a predetermined time window, and extract and fuse features of selected vein segments. A homomorphic encryption comparison module, which uses homomorphic encryption technology to compare the final verification features with the corresponding encrypted template at the time of registration within the ciphertext domain; An adaptive strategy scheduling module dynamically selects a verification mode and decides whether to allow the operation based on the currently visible body part, the operation risk level, and the confidence score. The structured log module is used to record task session ID, operation type code and risk level information, and to generate a digest using the national cryptographic SM3 hash and then sign and write it into the local read-only log library. The abnormal behavior linkage module is used to collect user operation behavior sequences in real time, calculate the current behavior abnormal score based on the historical normal behavior baseline, and trigger corresponding response measures.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images