Palm print vein recognition method, electronic equipment and computer program product

Abstract

The invention provides a palm print vein recognition method, electronic equipment and a computer program product. The palmprint and vein recognition method comprises the steps that feature extraction is conducted on palmprint data and vein data of a user, and a first palmprint feature and a first vein feature are determined; fusing the first palmprint feature and the second palmprint feature to obtain a palmprint fusion feature; fusing the first vein feature and the second vein feature to obtain a vein fusion feature; comparing the first palmprint feature and the palmprint fusion feature with a palmprint feature and a palmprint fusion feature in a database, and determining a palmprint comparison score; comparing the first vein feature and the vein fusion feature with a vein feature and a vein fusion feature in a database, and determining a vein comparison score; and determining whether the user is a registered user in a database based on the palm print comparison score and the vein comparison score.

Description

Technical Field

[0001] This disclosure relates to a palm vein recognition method, electronic device, and computer program product. Background Technology

[0002] In the process of palm vein recognition, traditional methods have limited ability to represent palm veins and are difficult to cope with real-world interference such as complex lighting, pose changes, or image noise. While pure deep learning methods can automatically learn high-level semantic features and improve recognition accuracy, they are highly dependent on a large amount of labeled data. In actual deployment, the model often suffers from poor generalization ability and high risk of overfitting due to the scarcity of user registration samples or inconsistent equipment collection quality. Summary of the Invention

[0003] This disclosure provides a palm vein recognition method, an electronic device, and a computer program product.

[0004] According to one aspect of this disclosure, a palmprint vein recognition method is provided, comprising: extracting features from a user's palmprint data and vein data respectively to determine a first palmprint feature and a first vein feature, wherein the first palmprint feature includes image texture features and image pixel position features in the palmprint data; the first vein feature includes image texture features and image pixel position features in the vein data; fusing the first palmprint feature and the second palmprint feature to obtain a palmprint fusion feature, wherein the second palmprint feature is obtained by extracting features from the palmprint data through a deep learning network; fusing the first vein feature and the second vein feature to obtain a vein fusion feature, wherein the second vein feature is obtained by extracting features from the vein data through the deep learning network; comparing the first palmprint feature and the palmprint fusion feature with palmprint features and palmprint fusion features in a database to determine a palmprint comparison score; comparing the first vein feature and the vein fusion feature with vein features and vein fusion features in the database to determine a vein comparison score; and determining whether the user is a registered user in the database based on the palmprint comparison score and the vein comparison score.

[0005] Based on one aspect of palmprint and vein recognition methods, this paper processes user palmprint and vein data in parallel through dual channels, extracting and fusing first palmprint / first vein features and second palmprint / second vein features respectively. This constructs palmprint / vein fusion features that possess both structural stability and deep discriminative power. High-confidence palmprint and vein comparison scores are then generated through multi-granularity similarity comparison. Finally, a comprehensive decision is made based on the comparison scores to determine whether the user is a registered user in the database. This method effectively addresses interference from image blur, lighting variations, and device differences. Even when a single palmprint or vein modality fails, accurate recognition can still be achieved through another modality or fusion features, improving system security, robustness, and availability, and providing reliable identity authentication for high-security applications.

[0006] According to at least one embodiment of the palmprint and vein recognition method of this disclosure, feature extraction is performed on the user's palmprint data and vein data respectively to determine a first palmprint feature and a first vein feature, including: dividing the palmprint data or vein data to determine local image regions of the palmprint data or vein data; calculating the pixel gradient of image pixels in the local image regions of the palmprint data or vein data to determine the local pixel gradient of the local image regions of the palmprint data or vein data; statistically analyzing the local pixel gradient to determine the image texture statistical vector of the palmprint data or vein data; and constructing the first palmprint feature or the first vein feature based on the image texture statistical vector of the palmprint data or vein data.

[0007] According to at least one embodiment of the palmprint and vein recognition method of this disclosure, feature extraction is performed on the user's palmprint data and vein data to determine a first palmprint feature and a first vein feature, including: scanning image pixels in the palmprint data or vein data to determine the intersection points and endpoints of the ridges in the palmprint data or vein data; performing Gaussian response calculation on the palmprint data or vein data to determine the extreme points and target points of the palmprint data or vein data, wherein the target point is an image pixel whose Gaussian response value is greater than or equal to a target threshold; concatenating the coordinates, angles, and neighborhood information vectors of the intersection points, endpoints, extreme points, and target points to determine the ridge pixel distribution vector of the palmprint data or vein data, wherein the neighborhood information vector is used to describe the texture structure of the remaining image pixels within the target range centered on the intersection points, endpoints, extreme points, or target points; and constructing the first palmprint feature or the first vein feature based on the ridge pixel distribution vector of the palmprint data or vein data.

[0008] According to at least one embodiment of the palm vein recognition method of this disclosure, determining palm vein fusion features and vein fusion features includes: performing feature transformation on the first palm vein feature or the first vein feature to determine the transformation features of the first palm vein feature or the first vein feature; fusing the transformation features of the first palm vein feature with the second palm vein feature to determine palm vein fusion features; and fusing the transformation features of the first vein feature and the second vein feature to determine vein fusion features.

[0009] According to at least one embodiment of the palmprint vein recognition method of this disclosure, determining palmprint fusion features and vein fusion features further includes: performing feature separation on the palmprint fusion features or the vein fusion features to determine the separation features of the palmprint fusion features or the vein fusion features; and constructing a loss function of a deep learning network based on the palmprint fusion features or the vein fusion features, and respectively with the corresponding separation features and transformation features, wherein the loss function is used to drive the deep learning network to learn.

[0010] According to at least one embodiment of the palmprint vein recognition method of this disclosure, determining the palmprint comparison score and the vein comparison score includes: calculating the similarity between the first palmprint feature and palmprint features in the database to determine the palmprint similarity; calculating the similarity between the first vein feature and vein features in the database to determine the vein similarity; calculating the similarity between the palmprint fusion feature and palmprint fusion features in the database to determine the palmprint fusion similarity; calculating the similarity between the vein fusion feature and vein fusion features in the database to determine the vein fusion similarity; performing a weighted calculation on the palmprint similarity and the palmprint fusion similarity to determine the palmprint comparison score; and performing a weighted calculation on the vein similarity and the vein fusion similarity to determine the vein comparison score.

[0011] According to at least one embodiment of the palm vein recognition method of this disclosure, determining whether a user is a registered user in the database based on the palm vein comparison score and the vein comparison score includes: performing a weighted calculation on the palm vein comparison score and the vein comparison score to determine a comprehensive score; if the comprehensive score is greater than or equal to a threshold score, then the user is a registered user in the database; if the comprehensive score is less than the threshold score, then the user is not a registered user in the database.

[0012] According to at least one embodiment of the palm vein recognition method of this disclosure, the weighted calculation process includes: when the weighting coefficients are non-learnable parameters, performing weighted calculation through a piecewise function; when the weighting coefficients are learnable parameters, performing weighted calculation through a multilayer perceptron.

[0013] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a palm vein recognition method according to any embodiment of this disclosure.

[0014] According to another aspect of this disclosure, a readable storage medium is provided, wherein executable instructions are stored therein, which, when executed by a processor, are used to implement the palm vein recognition method of any embodiment of this disclosure.

[0015] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a palm vein recognition method according to any embodiment of this disclosure. Attached Figure Description

[0016] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0017] Figure 1 This is a schematic diagram of the overall process of a palm vein recognition method according to one embodiment of the present disclosure.

[0018] Figure 2 This is a schematic diagram of the process of constructing a first palm print feature or a first vein feature in a palm print vein recognition method according to one embodiment of the present disclosure.

[0019] Figure 3 This is a schematic diagram of the process for constructing a first palm print feature or a first vein feature in a palm print vein recognition method according to another embodiment of the present disclosure.

[0020] Figure 4 This is a schematic flowchart illustrating the process of determining palm print fusion features and vein fusion features in a palm print vein recognition method according to one embodiment of the present disclosure.

[0021] Figure 5 This is a schematic flowchart illustrating the construction of a loss function in a palm vein recognition method according to one embodiment of the present disclosure.

[0022] Figure 6 This is a schematic flowchart illustrating the process of determining palm print comparison scores and vein comparison scores in a palm print and vein recognition method according to one embodiment of the present disclosure.

[0023] Figure 7 This is a schematic diagram illustrating the process of determining whether a user's palm print and vein data exist in a database in a palm vein recognition method according to one embodiment of the present disclosure.

[0024] Figure 8 This is a schematic block diagram of a palm vein recognition method according to one embodiment of the present disclosure.

[0025] Figure 9 This is a schematic block diagram of adaptive feature-supervised learning in a palm vein recognition method according to one embodiment of the present disclosure.

[0026] Figure 10 This is a schematic block diagram of feature comparison and adaptive two-level cascaded decision-making in a palm vein recognition method according to one embodiment of the present disclosure.

[0027] Figure 11 This is a schematic structural block diagram of a palm vein recognition device according to one embodiment of the present disclosure.

[0028] Figure 12 This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0029] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0030] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] In practical applications of palmprint and vein recognition, such as financial ATMs, access control systems, or mobile payment terminals, users often face challenges such as uneven lighting, changes in hand posture, blurred images, or differences in acquisition devices. This leads to high false recognition rates for single-modality (palmprint only or vein only) or single-feature type (manually designed or deep learning only) recognition systems. At the same time, in scenarios with scarce data (such as a small number of new user registration samples), pure deep learning models are difficult to converge due to a lack of sufficient labeled data, resulting in poor generalization ability.

[0032] To address this issue, this disclosure proposes a palmprint and vein recognition method. By fusing manually designed features (i.e., first palmprint features and first vein features) with deep learning features (i.e., palmprint fusion features and vein fusion features), the robustness of the system under complex conditions such as uneven lighting, pose changes, image blurring, or device differences is effectively improved. To address the problem of insufficient generalization ability of pure deep learning models due to the scarcity of new user registration samples, manually designed texture and keypoint features are introduced as prior knowledge to guide the deep network to converge stably with limited data, enhancing its adaptability to unknown samples. Simultaneously, the independent extraction and fusion comparison mechanism of palmprint and vein dual-modal features achieves multi-source information complementarity, significantly reducing the risk of misidentification caused by single-modal failure. Ultimately, while ensuring high recognition accuracy, the usability and security of the system in real-world environments are improved.

[0033] The palm vein recognition method disclosed herein can be deployed on terminal devices (such as smartphones, smart locks, ATMs, or wearable devices) to achieve localized real-time identity authentication, meeting the needs of scenarios such as mobile payment, contactless access control, or privacy protection. It can also be deployed on the server side to provide high-concurrency, high-security centralized biometric comparison services for financial risk control, government platforms, or unified enterprise identity management, as well as barrier-free authentication for special groups such as the elderly or visually impaired individuals, achieving coverage from everyday convenience scenarios.

[0034] Figure 1 A schematic flowchart illustrating the overall process of a palm vein recognition method according to one embodiment of this disclosure is shown. Figure 1 The method M100 shown includes steps S110 to S140. This method can be executed by electronic devices such as mobile phones and tablets.

[0035] In step S110, feature extraction is performed on the user's palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature. The first palmprint feature includes image texture features and image pixel position features in the palmprint data; the first vein feature includes image texture features and image pixel position features in the vein data.

[0036] Feature extraction processes were performed independently on the user's palmprint and vein data. The extracted first palmprint and first vein features each contain two types of complementary information: statistical features reflecting the global image texture distribution and image pixel location features characterizing local structural semantics. By combining these two types of features, a primary feature vector with both global discriminative power and local detail representation capability was constructed, providing prior guidance for subsequent deep learning feature fusion and supervision.

[0037] Preferably, image pixel location features refer to obtaining the spatial location information of image pixels by detecting and locating the positions of image pixels with significant structural significance in palmprint data or vein data. Image pixel location features include the two-dimensional coordinates, principal orientation angle, and neighborhood information vector of the image pixel in the palmprint data or vein data, which together constitute a description of the local structural topological relationships.

[0038] Preferably, the image texture features are descriptive information extracted by quantifying the spatial distribution pattern, intensity of change and repetition pattern of pixel grayscale in palm print data or vein data, which is used to characterize visual attributes such as the direction of lines, thickness, density, contrast and structural periodicity of blood vessels.

[0039] Preferably, the palmprint data includes a palmprint image, and the vein data includes vein images of the same palmprint.

[0040] For example, in real-world deployment scenarios such as financial ATMs or corporate access control systems, image quality often degrades due to sweat, dirt, or aging acquisition equipment. In such cases, relying solely on a single texture feature is susceptible to interference, while relying solely on deep learning features is difficult to generalize due to insufficient samples. Step S110, by simultaneously extracting image texture features (strong noise resistance) and image pixel distribution features (stable spatial structure), enables the system to maintain a high recognition rate even under conditions of image blur, low contrast, or partial occlusion, effectively addressing non-ideal acquisition conditions in real-world environments.

[0041] In step S120, the first palmprint feature and the second palmprint feature are fused to obtain the palmprint fusion feature. The second palmprint feature is obtained by extracting features from the palmprint data through a deep learning network. The first vein feature and the second vein feature are fused to obtain the vein fusion feature. The second vein feature is obtained by extracting features from the vein data through a deep learning network.

[0042] In the palmprint modality, the first palmprint feature (i.e., manually designed feature) reflecting texture distribution and structural semantics is fused with the second palmprint feature automatically extracted by a deep neural network to generate a palmprint fusion feature. Similarly, in the vein modality, the first vein feature and the second vein feature are fused to generate a vein fusion feature. Fusion methods may include weighted concatenation, attention mechanisms, feature mapping alignment, or adaptive fusion networks to achieve complementary advantages and information gain between the two types of features.

[0043] For example, when palmprint or vein data is partially blurred, palmprint fusion features or vein fusion features may fail, but the image pixel positions and image textures in the first palmprint or first vein features can still provide effective clues. Conversely, when the first palmprint or first vein features are inaccurate due to pose changes, palmprint fusion features or vein fusion features can be compensated for by contextual relationships, thereby improving the usability and reliability of the system in real-world environments.

[0044] Preferably, the second palm print feature and the second vein feature can be obtained by the same deep learning network or by different deep learning networks.

[0045] In step S130, the first palm print feature and the palm print fusion feature are compared with the palm print features and palm print fusion features in the database to determine the palm print comparison score; the first vein feature and the vein fusion feature are compared with the vein features and vein fusion features in the database to determine the vein comparison score.

[0046] In the palmprint modality, the similarity between the first palmprint feature and the palmprint feature of the user in the database, as well as the similarity between the palmprint fusion feature and the palmprint fusion feature of the user in the database, are calculated to generate a palmprint comparison score. Similarly, in the vein modality, the first vein feature and the vein fusion feature are compared and fused with the vein feature and the vein fusion feature in the database in a dual-channel manner to generate a vein comparison score.

[0047] Optionally, the database contains palmprint data, vein data, palmprint features, vein features, palmprint fusion features, and vein fusion features of all registered users, typically stored in a structured format, such as user ID, feature vector, registration time, and device information. These features are extracted and standardized during the registration phase using manual design or deep learning methods for rapid retrieval and matching during subsequent real-time comparisons.

[0048] In step S140, based on the palm print comparison score and vein comparison score, it is determined whether the user is a registered user in the database.

[0049] The palmprint comparison score and the vein comparison score are fused together. Then, the palmprint data and vein data of all registered users in the database are sorted or filtered by threshold based on the comprehensive score to determine whether the user is a registered user in the database.

[0050] For example, in practical applications, identity authentication needs to be completed within milliseconds, and false recognition must be avoided. In this case, relying solely on a single modality (such as palm print) is prone to misjudgment due to environmental interference. Step S140, however, fuses the scores of two independent modalities—palm print and vein—to form dual verification. Even if one modality deviates due to hand dirt or device aging, the other modality can still provide effective support. When a user's palm print data is blurred due to sweat, the palm print comparison score may be low, but the vein image remains unaffected, and the vein comparison score remains stable. The fusion of the two modalities still correctly identifies the user, thereby improving the system's reliability and security in complex real-world environments.

[0051] Therefore, the palm vein recognition method disclosed in this paper improves the accuracy of palm vein identity authentication in real and complex scenarios through a phased, multimodal, and dual-feature collaborative recognition architecture. By extracting artificially designed features that combine global texture statistics and local key point structure, a stable and reliable prior expression is provided to the system, effectively dealing with acquisition interference such as image blurring and uneven lighting. Then, this artificial feature is fused with deep semantic features automatically extracted by deep learning, which not only retains the stability of traditional methods but also injects the powerful discriminative ability of data-driven models, achieving complementary advantages. A dual-channel parallel comparison mechanism is adopted to calculate the similarity between artificial features and fused features separately and then perform weighted fusion to form a more refined and reliable single-modal score. By fusing the two independent scores of palm print and vein across modalities, a dual verification mechanism is constructed to complete high-confidence identity screening within milliseconds. Even if one modality fails due to environmental factors, the other modality can still guarantee the recognition accuracy, thereby achieving anti-interference, low false recognition, and highly available practical performance.

[0052] Regarding step S110, feature extraction is performed on the palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature. In some embodiments of this disclosure, this may include, for example... Figure 2 Steps S1101 to S1104 are shown.

[0053] In step S1101, the palm print data or vein data is divided to determine the local image region of the palm print data or vein data.

[0054] The global image (i.e., palmprint data or vein data) is decomposed into multiple semantically meaningful local regions so that structured features such as texture and gradient can be extracted at the regional granularity, providing structured data input for achieving high-precision identity recognition.

[0055] Specifically, palmprint or vein data is divided into several local image patches of uniform size or dynamically adjusted according to texture density (e.g., 8×8 or 16×16 pixel regions). Each local image region serves as an independent analysis unit for subsequent calculation of local gradients or texture histograms. This localization process effectively captures subtle texture variations and structural distributions in palmprint or vein data, while reducing the impact of uneven illumination or pose shifts on the global image, thus enhancing the stability and discriminative power of feature extraction.

[0056] In step S1102, the pixel gradient of the image pixels in the local image region of the palm print data or vein data is calculated to determine the local pixel gradient of the local image region of the palm print data or vein data.

[0057] Directional and intensity information is extracted from local image regions of palmprint or vein data to quantify key biometric features such as texture direction, edge distribution, and structural abrupt changes. Relying solely on raw pixel grayscale values ​​is susceptible to uneven lighting or noise interference. For example, even if grayscale values ​​are distorted in a part of the palm due to dirt, the gradient direction of the ridges remains stable, thus maintaining the effectiveness of the features. Furthermore, in palmprint or vein data collected from different devices, gradient calculation can automatically adapt to brightness changes due to differences in lighting conditions, improving cross-device recognition consistency and thus enhancing recognition stability in complex real-world environments.

[0058] Specifically, within each segmented local image region, the gradient magnitude and orientation angle of each image pixel in the horizontal and vertical directions are calculated. This is further extended to multi-scale gradient analysis to enhance the responsiveness to texture structures of different frequencies. The obtained local pixel gradient information will serve as the basis for subsequent texture histogram statistics, used to describe the texture orientation distribution and energy intensity of the local image region.

[0059] Alternatively, the pixel gradient can be obtained by approximating it with the first derivative using Sobel, Prewitt, or the central difference operator.

[0060] In step S1103, the local pixel gradients are statistically analyzed to determine the image texture statistical vector of the palm print data or vein data.

[0061] The gradient distribution of local pixel gradients at the pixel level is transformed into statistically significant, highly comparable, and dimensionally controllable feature vectors, thereby achieving a quantitative description of image texture. By aggregating local gradient information, a stable expression of the overall texture direction, density, and orientation distribution of palm prints or vein data is formed.

[0062] Preferably, within each local image region, the calculated pixel gradient direction angle is normalized and statistically analyzed according to a preset angle range (e.g., 0°–180° divided into 9 bins), and weighted by gradient magnitude to generate a texture histogram for that local image region. Subsequently, the histograms of all local image regions are stitched together or concatenated to form an image texture statistical vector for the entire palmprint or vein data.

[0063] In step S1104, a first palm print feature or a first vein feature is constructed based on the image texture statistical vector of palm print data or vein data.

[0064] By analyzing and extracting unique texture information from palmprint or vein data, highly discriminative feature vectors are formed. These feature vectors not only accurately represent the unique biometrics of an individual but also maintain consistency across different environmental conditions and acquisition devices.

[0065] Therefore, the global image (palmprint data or vein data) is divided into multiple local image regions to effectively isolate noise interference and focus on structural details. Then, the pixel gradients within each local image region are calculated to accurately capture texture direction and edge intensity, improving robustness to illumination changes and local contamination. Subsequently, the gradient directions are weighted, statistically analyzed, and normalized to construct a standardized texture vector that characterizes the overall texture distribution. Finally, the first palmprint feature or the first vein feature is generated based on the standardized texture vector, forming a basic feature representation that combines stability, comparability, and strong discriminative power.

[0066] Regarding step S110, feature extraction is performed on the palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature. In some embodiments of this disclosure, this may include, for example... Figure 3 Steps S310 to S340 are shown.

[0067] In step S310, the image pixels in the palm print data or vein data are scanned to determine the intersection points and endpoints of the ridges in the palm print data or vein data.

[0068] A comprehensive scan of palmprint or vein data is performed to detect all potential feature points (i.e., image pixels), and to identify and locate intersections and endpoints in the palmprint or vein data.

[0069] Preferably, the palmprint or vein data is denoised, and directional filtering is used to enhance the ridges. Then, ridge segmentation and binarization are performed to obtain preprocessed palmprint or vein data. The image pixels in the preprocessed palmprint and vein data are then scanned.

[0070] In step S320, Gaussian response calculation is performed on the palmprint data or vein data to determine the extreme points and target points of the palmprint data or vein data. The target point is the image pixel whose Gaussian response value is greater than or equal to the target threshold.

[0071] Identify image pixels with high response values ​​from palmprint or vein data; these are the most significant and discriminative extreme points and target points in the image where texture or structural changes are most pronounced. Target points correspond to the geometric centers of biometric features such as ridge intersections, branches, endpoints, or blood vessel bifurcations, serving as key anchor points for constructing local descriptors and achieving identity matching.

[0072] Preferably, the target threshold is used to filter out target points with significant structural meaning from the Gaussian response map, representing the minimum judgment criterion for the system's assessment of the importance or salience of local structure in the image. The target threshold can be dynamically adjusted according to image quality, device type, or security level. For example, it can be set to a higher value when the image is clear to retain the most significant target points and improve matching accuracy; it can be appropriately lowered when the image is blurry or noisy to retain more potential target points and avoid missed detections.

[0073] In step S330, the coordinates, angles, and neighborhood information vectors of the intersection point, endpoint, extreme point, and target point are concatenated to determine the distribution vector of the ridge pixel points of the palm print data or vein data. The neighborhood information vector is used to describe the texture structure of the remaining image pixels within the target range centered on the intersection point, endpoint, extreme point, or target point.

[0074] For each type of keypoint (intersection, endpoint, extremum, and target point), its geometric attributes are extracted, including 2D coordinates and principal orientation angles, as well as a neighborhood information vector describing the texture structure of the local image region surrounding the keypoint. This keypoint and its corresponding neighborhood information vector are concatenated to form a complete texture pixel distribution vector. The neighborhood information vector is typically based on texture descriptors such as HOG, LBP, or Gabor filter responses calculated within a circular or square window centered on the keypoint and with a radius of R. This achieves a leap from discrete point detection to structured feature representation, improving recognition accuracy and environmental adaptability.

[0075] In step S340, a first palm print feature or a first vein feature is constructed based on the distribution vector of the ridge pixel points of the palm print data or vein data.

[0076] The distribution vector of ridge pixel points is normalized to eliminate scale and energy differences, and the first palm print feature or first vein feature with a uniform format and controllable dimension is output.

[0077] Therefore, the first palm print feature or the first vein feature that is finally constructed provides a solid and reliable prior foundation for subsequent deep learning feature fusion and cross-modal decision-making.

[0078] Regarding step S120, fusing the first palmprint feature and the second palmprint feature to obtain a palmprint fusion feature; fusing the first vein feature and the second vein feature to obtain a vein fusion feature, in some embodiments of this disclosure, may include, for example... Figure 4 Steps S1201 to S1202 are shown.

[0079] In step S1201, the first palm print feature or the first vein feature is transformed to determine the transformed features of the first palm print feature or the first vein feature.

[0080] The first palmprint feature or the first vein feature is subjected to dimensionality reduction, normalization, semantic alignment, or spatial projection transformation to determine the transformed features of the first palmprint feature or the first vein feature. While preserving the key discriminative information of the original features (first palmprint feature or first vein feature), the transformed features are adjusted to have a similar statistical distribution, dimensional structure, or semantic space to the deep learning features (palmprint fusion features or vein fusion features), thereby providing standardized input for subsequent feature fusion, similarity calculation, or supervised guidance.

[0081] In step S1202, the transformation features of the first palm print feature are fused with the second palm print feature to determine the palm print fusion feature, and the transformation features of the first vein feature and the second vein feature are fused to determine the vein fusion feature.

[0082] This ensures that palm prints and veins complement and enhance each other within a unified space, enabling the final palm print fusion features and vein fusion features to simultaneously possess structural interpretability, environmental robustness, and deep discriminative power.

[0083] Preferably, the transformation features of the first palm print feature or the first vein feature can be fused with the corresponding second palm print feature or the second vein feature through weighted splicing, attention mechanism or deep learning network.

[0084] Therefore, it achieves efficient collaboration between manually designed features and deep learning features in a unified semantic space, effectively bridging the expression differences between traditional manually designed features and modern deep models, and improving the system's reliability, security, and practical adaptability.

[0085] Regarding step S1202, fusing the transformed features of the first palmprint feature with the second palmprint feature to determine the palmprint fusion feature, and fusing the transformed features of the first vein feature with the second vein feature to determine the vein fusion feature, in some embodiments of this disclosure, may include, for example... Figure 5 Steps S510 to S520 are shown.

[0086] In step S510, the palm print fusion feature or vein fusion feature is separated to determine the separation feature of the palm print fusion feature or vein fusion feature.

[0087] In specific palm vein recognition scenarios, relying solely on a single fusion feature may make it difficult to pinpoint the cause of failure due to the complexity of internal information. For example, when recognition fails, it's impossible to determine whether the inaccuracy of the manually designed features or the interference with the deep learning features is the cause. Step S510, through feature separation, enables the system to diagnose the source of the problem: if the separated manually designed feature portion scores stably but the deep learning feature portion fluctuates significantly, it indicates that environmental interference has affected the deep learning model; conversely, it suggests a defect in the manually designed feature extraction process. Furthermore, in cross-device or cross-modal scenarios, the separated features can be used to individually calibrate the consistency of expression across different modalities, improving the system's generalization ability and maintainability.

[0088] Preferably, the high-dimensional palmprint fusion features or vein fusion features are decomposed using the fully connected layers or linear projection matrix of a deep learning network.

[0089] In step S520, based on palmprint fusion features or vein fusion features, and the corresponding separation features and transformation features, a loss function for the deep learning network is constructed. The loss function is used to drive the deep learning network to learn.

[0090] In palmprint or vein modalities, the second palmprint feature / second vein feature output by the deep learning network is used as the anchor point. Simultaneously, the similarity or distance loss between this feature and the transformation features of the first palmprint feature / first vein feature, as well as the separation features of the palmprint fusion feature / vein fusion feature, is calculated to form a supervised loss function. This supervised loss function not only encourages the palmprint fusion feature / vein fusion feature to approximate the real identity label, but also forces it to closely approximate the stable semantic structure represented by the first palmprint feature / first vein feature in the feature space, thereby driving the deep learning network to learn.

[0091] Therefore, by accurately locating the contributions of both human and deep components in the fusion features through feature separation, the system's fault diagnosis capability is improved. Furthermore, a multi-objective supervised loss function is constructed based on the separated and transformed features, forcing the deep learning network's output features to closely approximate the human-generated prior semantic structure. This approach maintains high generalization and robustness even under real-world challenges such as image blurring, lighting variations, or sample scarcity, thereby enhancing system availability, security, and maintainability.

[0092] Regarding step S130, the first palmprint feature and palmprint fusion feature are compared with the palmprint features and palmprint fusion features in the database to determine the palmprint comparison score; the first vein feature and vein fusion feature are compared with the vein features and vein fusion features in the database to determine the vein comparison score. In some embodiments of this disclosure, it may include, for example... Figure 6 Steps S1301 to S1303 are shown.

[0093] In step S1301, the similarity between the first palm print feature and the palm print features in the database is calculated to determine the palm print similarity; the similarity between the first vein feature and the vein features in the database is calculated to determine the vein similarity. In the palmprint modality, the similarity between the first palmprint feature and the palmprint features of all users in the database is calculated one by one, and the palmprint similarity is output as greater than or equal to the similarity threshold. Similarly, in the vein modality, the similarity between the first vein feature and the vein features of all users in the database is calculated, and the vein similarity is output as greater than or equal to the similarity threshold.

[0094] Preferably, the calculation process of palmprint similarity or vein similarity is based on fast retrieval and sorting, combined with hash index or near nearest neighbor search (ANN) to meet the millisecond-level real-time response requirements, ensuring that the initial identity screening can still be completed efficiently in high-concurrency scenarios.

[0095] In step S1302, the similarity between the palmprint fusion feature and the palmprint fusion feature in the database is calculated to determine the palmprint fusion similarity; the similarity between the vein fusion feature and the vein fusion feature in the database is calculated to determine the vein fusion similarity. In step S1303, the palmprint similarity and palmprint fusion similarity are weighted and calculated to determine the palmprint comparison score; the vein similarity and vein fusion similarity are weighted and calculated to determine the vein comparison score.

[0096] In the palmprint modality, palmprint similarity and palmprint fusion similarity are linearly weighted according to preset palmprint weights to obtain the palmprint matching score. Similarly, in the vein modality, vein similarity and vein fusion similarity are linearly weighted according to vein weights to obtain the vein matching score. This ensures that the palmprint matching score and vein matching score reflect both the basic matching ability and the gain effect of high-level feature representation.

[0097] Preferably, the palm print weight or vein weight can be a fixed weight or a dynamically adjusted weight (such as adaptively assigning weights based on image quality, device type, or historical performance).

[0098] Preferably, in the weighted calculation process for palmprint comparison score or vein comparison score, when the weighting coefficient is a non-learnable parameter, the weighting calculation is performed through a piecewise function; when the weighting coefficient is a learnable parameter, the weighting calculation is performed through a multilayer perceptron.

[0099] Furthermore, step S1302 is the same as step S1301, and will not be described in detail here.

[0100] Therefore, it can effectively cope with interference from image blur, changes in lighting, and differences in equipment. When a single feature fails, another type of feature can maintain system availability through weighted compensation, improving recognition accuracy and reducing the risk of false positives and false negatives.

[0101] Regarding step S140, based on the palmprint comparison score and vein comparison score, it is determined whether the user is a registered user in the database. In some embodiments of this disclosure, it may include, for example... Figure 7 Steps S1401 to S1402 are shown.

[0102] In step S1401, the palm print comparison score and the vein comparison score are weighted and calculated to determine the comprehensive score.

[0103] The palmprint and vein comparison scores, calculated independently for both modalities, are weighted and integrated to generate a comprehensive score, which serves as the final basis for identity verification. This approach fully leverages the complementary advantages of palmprint and vein biometrics in terms of acquisition methods, anti-forgery capabilities, and environmental adaptability, thereby achieving more accurate and reliable identity recognition in complex real-world scenarios.

[0104] Preferably, in the weighted calculation of the comprehensive score, when the weighting coefficients are non-learnable parameters, the weighting calculation is performed through a piecewise function; when the weighting coefficients are learnable parameters, the weighting calculation is performed through a multilayer perceptron.

[0105] In step S1402, if the overall score is greater than or equal to the threshold score, the user is a registered user in the database; if the overall score is less than the threshold score, the user is not a registered user in the database.

[0106] The system relies on a comprehensive score to efficiently filter palmprint and vein data from all registered users stored in the database. Only palmprint and feature data records with a comprehensive score greater than or equal to a preset threshold are retained as the final successfully matched user identities. By setting a security threshold, the system controls the risk of false identification while ensuring recognition accuracy, providing a precise and reliable basis for subsequent identity authorization, access control, or transaction execution.

[0107] This reduces the false recognition and false negative rates of palm vein recognition, providing a solid identity authentication guarantee for high-security applications.

[0108] The technical solution of this disclosure will be further explained below using a specific implementation of palm vein recognition as an example.

[0109] like Figure 8 As shown, manually designed features are extracted from palmprint data and vein data respectively to generate corresponding manually designed palmprint features (i.e., first palmprint features) and manually designed vein features (i.e., first vein features). Simultaneously, these two types of raw data are also input into corresponding adaptive feature-supervised neural networks (i.e., deep learning networks) to output palmprint deep learning features (i.e., palmprint fusion features) and vein deep learning features (vein fusion features). Subsequently, the manually designed features and deep learning features corresponding to the palmprint and vein modalities are compared with the manually designed features and deep learning features of the palmprint and vein modalities in the database. The comparison results of the two modalities are then processed through an adaptive two-level cascaded decision system to output the final recognition result.

[0110] Specifically, the artificially designed features for palm print data or vein data include texture histogram statistical vectors (i.e., image texture statistical vectors) and ridge key point distribution vectors (i.e. ridge pixel distribution vectors).

[0111] The method for calculating the statistical vector of the image texture histogram is as follows: Calculate pixel gradient information, including magnitude and angle. Divide palmprint or vein data into multiple local image regions, statistically analyze the magnitude and angle of local and global pixels, and obtain texture histogram statistical vectors according to rules.

[0112] Preferably, the mathematical expression for the pixel gradient is: , The expression for the texture histogram statistical vector is: d represents the gradient magnitude. Let dx represent the gradient angle, dy represent the gradient component in the x-direction, and dy represent the gradient component in the y-direction. Represents the image grayscale function. This represents the texture histogram statistical vector, where m represents the number of local image regions. This represents a statistical vector representing a local image region k. and This represents the gradient magnitude and gradient angle of the j-th pixel in the k-th local image region. This is the gradient histogram statistical function. n represents the number of image pixels participating in the statistics in the k-th local image region.

[0113] The calculation method for the distribution vector of key points of the ridge line is as follows: Step 1: Denoise the palmprint or vein data, enhance the ridges using directional filtering, perform ridge segmentation and binarization, and detect the intersection points and endpoint coordinates of the ridges in the thinned image by scanning the image pixels. Calculate the corresponding Gaussian response image and obtain the salient points (i.e., target points) with high extreme point responses.

[0114] Step 2: Merge the intersections, endpoints, and salient points to form a set of key point coordinates.

[0115] Step 3: Extract the neighborhood information vector of the key points in the key point coordinate set.

[0116] Step 4: Combine the coordinates, angles, and neighborhood information vectors of the key points in the key point coordinate set to form the distribution vector of the key points of the ridge line.

[0117] Preferably, the expression for the distribution vector of key points of the ridge line is: in This represents the distribution vector of key points of the ridge line. For the neighborhood information of key point i, and For the coordinates of the key points, Main direction angle, to This refers to the neighborhood information component of the key point.

[0118] Specifically, the aforementioned adaptive feature-supervised neural network is as follows: Figure 9 As shown, firstly, manually designed features are extracted from the original data (i.e., palmprint data or vein data). A signal conversion module is then used to normalize and transform these manually designed features, generating supervised features F2. Supervised features F2 are fused with the features FT (i.e., second palmprint features or second vein features) from the network extraction layer to obtain output features (i.e., palmprint fusion features or vein fusion features). The output features are then passed through a feature separation module to obtain supervised features F1. During training, the outputs of the terminal classification layer, supervised classification layer 1, and supervised classification layer 2 are simultaneously optimized to complete the adaptive supervised learning of the adaptive feature-supervised neural network.

[0119] Preferably, the formula for calculating the fused features is as follows: in The output features of the feature fusion layer (i.e., palmprint fusion features or vein fusion features). The output features of the network feature extraction layer (i.e., the second palm print feature or the second vein feature). For feature concatenation expansion operations, multidimensional vector concatenation or element-wise summation can be used. This is a commonly used feature normalization function. Features designed by humans This is a signal conversion function used to convert manually designed features into supervised features F2, while also serving as a feature balancer. and These are the learnable parameters of the model, used to adaptively adjust the weights.

[0120] Specifically, the above feature comparison and adaptive two-level cascade decision-making are as follows: Figure 10 As shown, the first-level decision-making process calculates the index results for each single modality (palmprint and vein) based on deep learning features and manually designed features, as follows: ) in, This indicates either a single-modal palmprint score (i.e., palmprint comparison score) or a single-modal vein score (i.e., vein comparison score). These represent the artificially designed features of palm prints or veins, respectively, and the palm print or vein features in the database. To design feature comparison functions manually, These represent the deep learning features of palm prints or veins, respectively, combined with palm print fusion features or vein fusion features from the database. This is a deep learning feature comparison function. Let be an adaptive score weighting function, where and For adaptive coefficients, and When set as a non-learnable parameter, This can be achieved using common piecewise functions. and When set as a learnable parameter, This can be achieved using a multilayer sensing network. Based on the above formulas, the single-modal palmprint score and the single-modal vein score are calculated respectively.

[0121] The second-level decision-making process calculates the bimodal index results based on the unimodal palmprint and vein indices, as follows: ) in, This represents the bimodal recognition score (i.e., the overall score). The first-level decision score for single-modality veins. The first-level decision score for a single-modal palmprint is determined using the same decision-making method as the formula above, where... and For adaptive coefficients, and When set as a non-learnable parameter, This can be achieved using common piecewise functions. and When set as a learnable parameter, This can be achieved using a multilayer sensing network.

[0122] It is understood that before using the technical solutions disclosed in the embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained. For example, in response to receiving a user's active request, a prompt message can be sent to the user to clearly inform the user that the requested operation will require the acquisition and use of the user's personal information. This allows the user to choose whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media that perform the operations of the technical solutions of this disclosure, based on the prompt message. As an optional but non-limiting implementation, the way to send a prompt message to the user in response to receiving a user's active request can be, for example, a pop-up window, in which the prompt message can be presented in text form. Furthermore, the pop-up window can also include a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0123] It is understood that the above notification and user authorization acquisition process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure. The data involved in the technical solution of this disclosure (including but not limited to the data itself, the acquisition or use of data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0124] Based on any of the above embodiments, this disclosure also provides a palm vein recognition device.

[0125] Figure 11 This is a schematic block diagram of the structure of a palm vein recognition device according to one embodiment of the present disclosure.

[0126] like Figure 11 As shown, the palm vein recognition device includes: The feature extraction module 1102 extracts features from the palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature. The first palmprint feature includes image texture features and image pixel position features in the palmprint data; the first vein feature includes image texture features and image pixel position features in the vein data. The feature fusion module 1104 fuses the first palmprint feature and the second palmprint feature to obtain the palmprint fusion feature. The second palmprint feature is obtained by feature extraction from the palmprint data through a deep learning network. The first vein feature and the second vein feature are fused to obtain the vein fusion feature. The second vein feature is obtained by feature extraction from the vein data through a deep learning network. The feature comparison module 1106 compares the first palmprint feature and the palmprint fusion feature with the palmprint feature and palmprint fusion feature in the database to determine the palmprint comparison score; it also compares the first vein feature and the vein fusion feature with the vein feature and vein fusion feature in the database to determine the vein comparison score. The identification module 1108 determines whether a user is a registered user in the database based on palm print comparison scores and vein comparison scores.

[0127] The aforementioned palm vein recognition device can be in the form of computer software, and each module of the aforementioned palm vein recognition device can be implemented through computer software modules.

[0128] The specific implementation process of the functions and roles of each module in the above palm vein recognition device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0129] This disclosure also provides an electronic device. Figure 12 A schematic diagram of the hardware implementation using the processing system is shown.

[0130] The hardware structure of electronic device 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc. Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one connection line is used in this figure, but this does not indicate that there is only one bus or one type of bus.

[0131] For ease of explanation, certain steps of the above method are described in relation to modules. It should be understood that the corresponding module performing one or more steps of the above method may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented by some combination thereof.

[0132] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0133] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0134] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0135] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0140] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0141] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A method for palm vein recognition, characterized in that, include: Feature extraction is performed on the user's palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature. The first palmprint feature includes image texture features and image pixel position features in the palmprint data; the first vein feature includes image texture features and image pixel position features in the vein data. The first palmprint feature and the second palmprint feature are fused to obtain a palmprint fusion feature. The second palmprint feature is obtained by feature extraction from the palmprint data through a deep learning network. The first vein feature and the second vein feature are fused to obtain a vein fusion feature. The second vein feature is obtained by feature extraction from the vein data through the deep learning network. The first palmprint feature and the palmprint fusion feature are compared with palmprint features and palmprint fusion features in the database to determine the palmprint comparison score; the first vein feature and the vein fusion feature are compared with vein features and vein fusion features in the database to determine the vein comparison score; and Based on the palm print comparison score and the vein comparison score, it is determined whether the user is a registered user in the database.

2. The palm vein recognition method as described in claim 1, characterized in that, Feature extraction is performed on the user's palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature, including: The palmprint data or vein data is segmented to determine the local image region of the palmprint data or the vein data; Calculate the pixel gradient of image pixels in a local image region of the palm print data or the vein data to determine the local pixel gradient of the local image region of the palm print data or the vein data; Statistical analysis is performed on the local pixel gradients to determine the image texture statistical vector of the palm print data or the vein data; Based on the image texture statistical vector of the palm print data or the vein data, a first palm print feature or a first vein feature is constructed.

3. The palm vein recognition method as described in claim 1, characterized in that, Feature extraction is performed on the user's palmprint data and vein data respectively to determine the first palmprint feature and the first vein feature, including: Scan the image pixels in the palm print data or vein data to determine the intersection points and endpoints of the ridges in the palm print data or vein data; Gaussian response calculation is performed on the palmprint data or the vein data to determine the extreme points and target points of the palmprint data or the vein data. The target point is an image pixel whose Gaussian response value is greater than or equal to a target threshold. The coordinates, angles, and neighborhood information vectors of the intersection point, the endpoint, the extreme point, and the target point are concatenated to determine the distribution vector of the ridge pixel points of the palm print data or the vein data. The neighborhood information vector is used to describe the texture structure of the remaining image pixels within the target range centered on the intersection point, the endpoint, the extreme point, or the target point. Based on the distribution vector of the ridge pixel points of the palm print data or the vein data, a first palm print feature or a first vein feature is constructed.

4. The palm vein recognition method as described in claim 1, characterized in that, Determine palm print fusion features and vein fusion features, including: The first palm print feature or the first vein feature is transformed to determine the transformed features of the first palm print feature or the first vein feature. The transformation features of the first palm print feature are fused with the second palm print feature to determine the palm print fusion feature, and the transformation features of the first vein feature and the second vein feature are fused to determine the vein fusion feature.

5. The palm vein recognition method as described in claim 4, characterized in that, Determining palm print fusion features and vein fusion features also includes: The palmprint fusion feature or the vein fusion feature is separated to determine the separation feature of the palmprint fusion feature or the vein fusion feature; Based on the palmprint fusion feature or the vein fusion feature, and the corresponding separation feature and transformation feature, a loss function for the deep learning network is constructed, which is used to drive the deep learning network to learn.

6. The palm vein recognition method as described in claim 1, characterized in that, Determine the palm print comparison score and vein comparison score, including: The similarity between the first palmprint feature and the palmprint features in the database is calculated to determine the palmprint similarity; the similarity between the first vein feature and the vein features in the database is calculated to determine the vein similarity. The similarity between the palmprint fusion feature and the palmprint fusion feature in the database is calculated to determine the palmprint fusion similarity; the similarity between the vein fusion feature and the vein fusion feature in the database is calculated to determine the vein fusion similarity. The palmprint similarity and the palmprint fusion similarity are weighted and calculated to determine the palmprint comparison score; the vein similarity and the vein fusion similarity are weighted and calculated to determine the vein comparison score.

7. The palm vein recognition method as described in claim 1, characterized in that, Based on the palmprint comparison score and the vein comparison score, determining whether the user is a registered user in the database includes: The palmprint comparison score and the vein comparison score are weighted and calculated to determine the comprehensive score; If the overall score is greater than or equal to the threshold score, then the user is a registered user in the database; if the overall score is less than the threshold score, then the user is not a registered user in the database.

8. The palm vein recognition method as described in claim 6 or 7, characterized in that, The weighted calculation process includes: When the weighting coefficients are non-learnable parameters, weighting is calculated using a piecewise function. When the weighting coefficients are learnable parameters, weighting calculation is performed through a multilayer perceptron.

9. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes the execution instructions stored in the memory, causing the processor to perform the palm vein recognition method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the palm vein recognition method according to any one of claims 1 to 8.

Images