Palm vein and palm print feature fused intelligent identity verification method and system

Abstract

The invention discloses a palm vein and palm print feature fusion intelligent identity verification method and system, and relates to the technical field of biological recognition, and the method comprises the steps: collecting a palm image and identity data of a to-be-authenticated user, and carrying out the identity coding of the palm image based on the identity data, and obtaining a palm identity coding image set; then, through region marking and feature extraction, multi-dimensional features of palm veins and palm prints are obtained; then, performing spatial alignment fusion on the features of the two to generate a palm code fusion feature data set; and then, through encryption processing and network training, a self-adaptive palm identity recognition network is obtained. And finally, acquiring a palm image of a target user, and performing identity verification by using the network. The technical problems of insufficient safety, low accuracy and poor anti-interference capability caused by dependence on a single biological characteristic in an existing identity authentication method are solved, and the technical effect of improving the safety, the accuracy and the anti-interference capability of identity authentication is achieved.

Description

Technical Field

[0001] The present application relates to the field of biometric identification technology, and in particular to an intelligent identity authentication method and system that integrates palm vein and palm print features. Background Art

[0002] With the continuous improvement of information security requirements, traditional authentication methods such as passwords, fingerprint recognition and facial recognition have gradually exposed problems such as insufficient security, low accuracy and poor anti-interference ability. The reliance on a single biometric feature makes these methods perform unsatisfactorily in complex environments, especially when damaged, polluted or the environment changes, the authentication process is easily affected. In order to improve the security and accuracy of identity authentication, more and more research has begun to turn to multimodal biometric technology, combining multiple biometrics for identity authentication. Palm vein recognition is widely used due to its high uniqueness and difficulty in forging, but it still faces problems with recognition rate and equipment cost. Similarly, palm print recognition technology also has certain challenges. Therefore, how to combine multiple biometrics such as palm veins and palm prints, and improve the accuracy and reliability of identity authentication through advanced algorithms, has become a key issue in the current technological development.

[0003] At the current stage, relevant technologies have technical problems in which identity authentication methods rely on a single biometric feature, resulting in insufficient security, low accuracy and poor anti-interference ability. Summary of the invention

[0004] The present application solves the technical problems that the existing identity authentication methods rely on a single biometric feature, resulting in insufficient security, low accuracy and poor anti-interference ability, by providing an intelligent identity authentication method and system that integrates palm vein and palm print features.

[0005] This application provides an intelligent identity authentication method that integrates palm vein and palm print features, including:

[0006] The palm image set and the user identity data set of the user set to be authenticated are sequentially acquired, and the palm image set is identity-encoded based on the user identity data set to obtain a palm identity encoding image set; the palm identity encoding image set is region-marked and feature-extracted to obtain a palm vein encoding multidimensional feature set and a palm print encoding multidimensional feature set; the palm vein encoding multidimensional feature set and the palm print encoding multidimensional feature set are spatially aligned and fused to obtain a palm encoding fusion feature data set; encryption processing and network training and tuning are performed based on the palm encoding fusion feature data set to obtain a palm identity recognition adaptive network; the target palm image of the target user is acquired, and the target palm image is identity-verified based on the palm identity recognition adaptive network.

[0007] This application provides an intelligent identity authentication system that integrates palm vein and palm print features, including:

[0008] A palm identity coding image set acquisition module, the palm identity coding image set acquisition module is used to sequentially acquire a palm image set and a user identity data set of a user set to be authenticated, and perform identity encoding on the palm image set based on the user identity data set to obtain a palm identity coding image set; a coding multidimensional feature set acquisition module, the coding multidimensional feature set acquisition module is used to perform region marking and feature extraction on the palm identity coding image set to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set; a spatial domain alignment and fusion module, the spatial domain alignment and fusion module is used to perform spatial domain alignment and fusion on the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain a palm coding fusion feature data set; an adaptive network acquisition module, the adaptive network acquisition module is used to perform encryption processing and network training and tuning based on the palm coding fusion feature data set to obtain a palm identity recognition adaptive network; an identity recognition and verification module, the identity recognition and verification module is used to acquire a target palm image of a target user, and perform identity recognition and verification on the target palm image based on the palm identity recognition adaptive network.

[0009] The intelligent identity authentication method and system proposed in this application that fuses palm vein and palm print features first collects the palm image and identity data of the user to be authenticated, and encodes the palm image based on the identity data to obtain a palm identity encoding image set. Then, through region marking and feature extraction, the multi-dimensional features of palm veins and palm prints are obtained. Then, the features of the two are spatially aligned and fused to generate a palm encoding fusion feature data set. Then, through encryption processing and network training, an adaptive palm identity recognition network is obtained. Finally, the palm image of the target user is collected, and the network is used for identity authentication. By fusing the palm vein and palm print features, the technical effect of improving the security, accuracy and anti-interference ability of identity authentication is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0011] Figure 1 A schematic diagram of the process of an intelligent identity authentication method for fusing palm vein and palm print features provided in an embodiment of the present application;

[0012] Figure 2A schematic diagram of the structure of an intelligent identity authentication system that integrates palm vein and palm print features provided in an embodiment of the present application.

[0013] Explanation of the reference numerals: palm identity coding image set acquisition module 10 , coding multi-dimensional feature set acquisition module 20 , spatial domain alignment and fusion module 30 , adaptive network acquisition module 40 , identity recognition and verification module 50 . DETAILED DESCRIPTION

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0015] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0016] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0017] The present application embodiment provides an intelligent identity authentication method that integrates palm vein and palm print features, such as Figure 1 As shown, the method includes:

[0018] Step S100, sequentially collect and obtain the palm image set and user identity data set of the user set to be authenticated, perform identity encoding on the palm image set based on the user identity data set, and obtain the palm identity encoding image set. Specifically, prepare an image acquisition device with high resolution and good optical performance, place it stably, ensure that the light in the acquisition environment is uniform and interference-free, and define a standard palm placement area. For the authenticated user, guide him to stretch his palm naturally and lay it flat as required, collect multiple images at a certain frame rate and select the best one, so as to obtain the palm image set. Clearly define the user identity data elements containing key information such as name and ID number, collect them through safe and compliant online or offline channels, ensure that the information is true and accurate, and obtain the user identity data set. Then, use the ID number as an associated identifier, correspond the two, build an identity encoding system based on an encryption algorithm (such as a hash function or a symmetric encryption algorithm), process the key information of the user identity data to generate a unique and irreversible (in the case of a hash function) or encrypted identity code, integrate it with the palm image, such as recording it in the image metadata or database association table, and finally generate a palm identity encoding image set, laying the foundation for the subsequent identity authentication process.

[0019] In a possible implementation, palm image sets and user identity data sets of the user set to be authenticated are collected and acquired in sequence, and the palm image sets are identity-encoded based on the user identity data sets to obtain palm identity encoding image sets. Step S100 further includes step S110, in which the user identity data sets and the palm image sets are associated and mapped to build a user identity palm image database. Specifically, a database management system is established to ensure that it has efficient data storage and retrieval capabilities. For each user to be authenticated, the user's identity information, such as name, ID number, contact information, etc., is extracted from the user identity data set, and the corresponding palm image is found from the palm image set. By designing a specific associated field, for example, using the user's unique identifier (such as ID number) as the primary key, the user's identity information and its palm image are associated and stored in the database to form a user identity palm image database. In subsequent operations, the corresponding palm image can be quickly and accurately obtained according to the user's identity information, and vice versa, which provides convenience and infrastructure for data management of the entire identity authentication system.

[0020] Step S120, extract identity elements from the user identity data set to obtain an identity key element set, and construct an identity coding system based on the identity key element set. Specifically, analyze the collected user identity data set, extract unique and representative identity elements, such as key information such as the area code, date of birth, gender code in the ID number, and surname in the name, to form an identity key element set. The elements should reflect the user's identity characteristics to a certain extent and meet the requirements of data security and privacy protection. According to the key elements, an encryption algorithm and encoding rules are used to construct an identity coding system. For example, a hash function can be used to encrypt the key elements to generate a hash value of a fixed length, and then the hash value can be further converted and combined in combination with specific encoding rules to form the final identity code. The encoding system can not only protect the user's privacy, because the original identity information will not be directly exposed, but also ensure that each user has a unique identification in the system through the uniqueness of the encoding, which is convenient for subsequent identity authentication and management operations.

[0021] Step S130, the user identity palm image database is subjected to sensitive data identification, the user identity sensitive data is obtained, and the user identity sensitive data is desensitized and encrypted to obtain an identity encrypted palm image database. Specifically, in the user identity palm image database that has been constructed, data fields that may contain sensitive information, such as user identity sensitive data such as ID card number, detailed home address, bank card number, etc., are identified through data scanning and analysis tools. For sensitive data, desensitization technology is used for processing, for example, part of the numbers in the ID card number are replaced by asterisks, only part of the information of the area code and date of birth is retained, and information such as home address and bank card number are blurred or anonymized. The desensitized data is encrypted using an encryption algorithm (such as a symmetric encryption algorithm or an asymmetric encryption algorithm) to ensure that even if the data is leaked, it is difficult for an attacker to obtain useful information. After desensitization and encryption processing, the original sensitive data in the database is replaced with the processed encrypted data, thereby obtaining an identity encrypted palm image database, which further enhances the security and privacy protection capabilities of the data.

[0022] Step S140, based on the identity coding system, the identity-encrypted palm image database is identity-encoded to obtain the palm identity-encoded image set. Specifically, based on the identity coding system constructed previously, the relevant information of each user is read from the identity-encrypted palm image database (although it is encrypted information, it can still be processed by a specific key and algorithm). According to the rules of the identity coding system, the information is encoded to generate an identity code corresponding to each user. Then, the identity code is associated with the corresponding palm image in the database, for example, a new field is added to the database to store the identity code, or the identity code is linked to the palm image by establishing an index, and finally a palm identity-encoded image set is formed. In the subsequent identity authentication process, whether in the feature extraction stage or in the network training and verification stage, the image data with the identity code can be conveniently used, while ensuring the security and integrity of the data, so that the entire identity authentication process can run efficiently and safely.

[0023] Step S200, the palm identity coding image set is subjected to region marking and feature extraction to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set. Specifically, the palm identity coding image set is subjected to image preprocessing, and noise is removed by using Gaussian filtering and median filtering, and contrast and details are highlighted by grayscale conversion and histogram equalization. Then, a semantic segmentation model based on CNN such as U-Net is selected, and a large number of labeled samples are used for training and optimization to accurately segment the palm area. The image set is input into the trained model to obtain the semantic segmentation result, and the regions of interest of the palm veins and palm prints are determined based on this, and their positions and other information are refined and annotated. In the palm vein area, after morphological operations such as expansion and corrosion, key feature points such as bifurcation points and local feature descriptors are extracted using algorithms such as SIFT to obtain a multidimensional feature set; for the palm print area, an edge image set is first generated by Canny edge detection, and then multidimensional line features and attribute information are extracted using Hough transform to obtain a palm print coding multidimensional feature set, laying the foundation for subsequent steps.

[0024] In a possible implementation, the palm identity coding image set is region marked and feature extracted to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set. Step S200 further includes step S210, performing noise identification and filtering processing on the palm identity coding image set to obtain a denoised palm identity coding image set. Specifically, the palm identity coding image set is subjected to noise identification scanning using image analysis software or an algorithm library. A variety of noise detection techniques are used, such as analyzing the distribution of image pixel values ​​to determine whether there is salt and pepper noise (pixel values ​​present extreme high or low values, and the distribution is relatively discrete), and using spectrum analysis methods to detect whether there is Gaussian noise in the image (abnormal energy in the high-frequency part of the spectrum). Once the noise type and distribution area are identified, a suitable filtering method is selected for processing. For salt and pepper noise, median filtering is usually used, that is, the pixel values ​​in the neighborhood of a pixel point in the image are sorted and the middle value is assigned to the pixel point to remove the noise point; for Gaussian noise, Gaussian filtering is used to convolve the image with the Gaussian kernel function to smooth the image and reduce the impact of noise, thereby obtaining a denoised palm identity coding image set, improving the clarity and quality of the image, and providing more reliable basic data for subsequent semantic segmentation and other operations.

[0025] Step S220, build a palm semantic segmentation model, and perform semantic segmentation on the denoised palm identity coding image set in turn based on the palm semantic segmentation model to obtain palm semantic segmentation results. Specifically, a palm semantic segmentation model is built based on a deep learning framework (such as TensorFlow, PyTorch, etc.). Common model architectures can use improved models based on convolutional neural networks (CNN), such as U-Net or SegNet, etc. The model performs well in image semantic segmentation tasks and can better handle complex image structures and details. Collect a large number of palm image samples covering different palm postures, skin colors, textures, and lighting conditions, and finely annotate the samples to mark different semantic categories such as the palm, fingers, palm side, palm vein area, and palm print area. Use the annotated sample set to train the built model. During the training process, the weight parameters of the model are continuously adjusted to enable it to learn the characteristic patterns and image expressions of different semantic areas. The denoised palm identity coding image set is input into the trained palm semantic segmentation model in sequence. The model performs pixel-level classification prediction on each image and determines the semantic category to which each pixel belongs, thereby obtaining the palm semantic segmentation result. The result is presented in the form of an image. Different colors or marks represent different semantic areas, which clearly divide the key parts of the palm and provide an accurate basis for subsequent regional anchor box marking.

[0026] Step S230, according to the palm semantic segmentation result, the denoised palm identity coding image set is marked with regional anchor frames to obtain a palm coding image region set of interest. Specifically, according to the segmentation result output by the palm semantic segmentation model, the key areas that need to be paid attention to, namely the palm vein area and the palm print area, are determined. An image annotation tool or a programming algorithm is used to draw accurate anchor frames for the areas on the denoised palm identity coding image set. For the palm vein area, the size and shape of the anchor frame are designed according to the common distribution range and morphological characteristics of the palm veins to ensure that the area where the palm veins are located can be tightly surrounded while minimizing the coverage of the surrounding irrelevant areas; for the palm print area, a suitable anchor frame is drawn according to its distribution characteristics and texture orientation to highlight the main part of the palm print. In the above manner, the palm vein and palm print areas in the palm image are marked to form a palm coding image region set of interest, so that the subsequent feature extraction operation can be concentrated on the key area, avoiding non-targeted feature extraction of the entire palm image, improving the efficiency and accuracy of feature extraction, and reducing the waste of computing resources and unnecessary feature interference.

[0027] Step S240, extracting the associated features of the palm coding image region of interest to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set. Specifically, on the marked palm vein coding image region of interest, a series of image processing operations are first performed to enhance the expressiveness of the vein features. For example, a color image is converted into a grayscale image by graying to highlight the contrast between the vein and the surrounding tissue; then the vein image is processed using morphological operations (such as expansion, corrosion, opening operation, closing operation, etc.). The expansion operation can fill in the tiny broken parts in the vein to make it more continuous. The corrosion operation can remove some tiny noise points and isolated pixels, and refine the lines of the vein. The opening operation and the closing operation further smooth the contour of the vein and remove defects such as burrs and small holes. Using feature point extraction algorithms, such as scale-invariant feature transform (SIFT), speeded up robust features (SURF) or histogram of oriented gradients (HOG), key feature points are extracted from the processed palm vein area. Feature points usually include points with obvious geometric features such as bifurcation points, intersection points, and endpoints of palm veins. Local feature descriptors around each feature point are calculated. For example, SIFT feature descriptors describe the features of the point by calculating the gradient direction and amplitude of the pixels around the feature point, thereby forming a multi-dimensional feature set for palm vein coding. The multi-dimensional features not only contain the geometric shape information of the palm vein, but also cover its local texture features. They comprehensively characterize the features of the palm vein from multiple dimensions, providing rich and distinguishing feature information for subsequent identity recognition. For the image region set of interest in palmprint coding, grayscale and filtering processing are also performed first to optimize image quality, remove noise, and enhance the clarity and contrast of the palmprint. By calculating the grayscale difference of the image, the edge information of the palmprint is determined using an edge detection algorithm (such as the Canny edge detection algorithm), and the edge points are marked to generate a palmprint coding edge image set. On this basis, we use line feature extraction algorithms, such as Hough transform, Radon transform, or chain code-based line tracking algorithms, to extract multidimensional line features such as main lines, wrinkles, and detail lines in the palmprint, as well as attribute information such as feature direction, length, curvature, and spacing, so as to obtain a palmprint encoding multidimensional feature set. Palmprint features can reflect the texture distribution and direction of the palm surface, and complement the palm vein features to form a unique biometric identification of the palm, further improving the accuracy and reliability of identity authentication. By performing the above feature extraction operations on the palm vein and palmprint areas respectively, we finally successfully obtained the palm vein encoding multidimensional feature set and the palmprint encoding multidimensional feature set, laying a solid foundation for subsequent feature fusion and identity authentication.

[0028] In a possible implementation, the palm coding image region set of interest is subjected to associated feature extraction to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set, and step S240 further includes step S241, performing grayscale image conversion and equalization processing on the palm coding image region set of interest to obtain a palm coding region grayscale image set. Specifically, for each image in the palm coding image region set of interest, a grayscale conversion algorithm (such as a weighted average method) is used to convert it from a color image to a grayscale image. The algorithm assigns different weights to the three color channels of red, green, and blue according to the sensitivity of the human eye to different colors, and obtains the grayscale value of each pixel by weighted summation, thereby removing color information and highlighting the brightness contrast of the image. The grayscale image is processed using histogram equalization technology. By counting the distribution of pixel grayscale values ​​in the image, a grayscale histogram is constructed. According to the distribution of the histogram, the grayscale values ​​of the pixels are redistributed to make the grayscale distribution more uniform, thereby enhancing the contrast and detail information of the image. In particular, for areas with originally low contrast such as palm veins and palm prints, their features can be more clearly displayed after equalization, and finally a grayscale image set of the palm coding area is obtained, providing a better quality image data basis for subsequent operations.

[0029] Step S242, morphological operations and feature point extraction are performed on the grayscale image set of the palm coding area to obtain the multi-dimensional feature set of the palm vein coding. Specifically, in the obtained grayscale image set of the palm coding area, for the regional image where the palm vein is located, morphological operations are applied to optimize the structural features of the vein. First, an erosion operation is performed, using a structural element of a specific shape and size (such as a circle or a rectangle) to traverse each pixel in the image. If the pixel value and the pixel values ​​in its neighborhood meet certain conditions (such as lower than a certain threshold), the value of the pixel is set to the central pixel value of the structural element, which can remove some small noise points and isolated pixels and refine the lines of the palm vein. Then, an expansion operation is performed. By using the same structural element, the values ​​of the neighboring pixels with the same central pixel value as the structural element are also set to the central pixel value, which helps to fill the tiny broken parts in the palm vein and make it more continuous. The contour of the palm vein is further smoothed by the opening operation (first erosion and then expansion) and the closing operation (first expansion and then erosion), the defects such as burrs and small holes are removed, and the morphological features of the palm vein are enhanced, making it easier to extract and analyze features. On the palm vein area image after morphological processing, feature point extraction algorithms such as scale-invariant feature transform (SIFT) are used. The SIFT algorithm first detects potential feature points in different scale spaces by constructing a Gaussian difference pyramid, and then screens these feature points, removes unstable points, and retains points with significant features, such as the bifurcation points, intersection points, and endpoints of the palm veins. For each retained feature point, the gradient direction and amplitude of the surrounding neighborhood pixels are calculated to construct a feature descriptor, which contains information such as the location, scale, and direction of the feature point, thereby forming a multi-dimensional feature set for palm vein encoding. The multi-dimensional features comprehensively characterize the features of the palm vein from multiple dimensions, including its geometric shape, local texture, and positional relationship in the image, providing rich and discriminative feature information for subsequent identity recognition.

[0030] Step S243, grayscale difference calculation and edge point marking are performed based on the palm coding area grayscale image set to generate a palmprint coding edge image set. Specifically, for each image in the palm coding area grayscale image set, its grayscale difference is calculated pixel by pixel. By comparing the grayscale value difference between each pixel and its neighboring pixels (such as 8 neighborhoods or 4 neighborhoods), the pixel point with a drastic grayscale change is determined, and the pixel point is often located in the edge area of ​​the palmprint. According to the preset grayscale difference threshold, when the grayscale difference between pixels exceeds the threshold, the pixel is marked as an edge point. The edge detection algorithm (such as the Canny edge detection algorithm) is used to further screen and connect the edge points initially marked, remove pseudo edge points generated due to noise and other reasons, and at the same time, through steps such as non-maximum suppression and double threshold detection, the edge of the palmprint is refined and accurately located, and finally a palmprint coding edge image set is generated, the contour and texture trend of the palmprint are depicted, and accurate basic data is provided for subsequent multi-dimensional line feature extraction.

[0031] Step S244, extract multidimensional line features from the palmprint coded edge image set in turn to obtain the palmprint coded multidimensional feature set. Specifically, a line feature extraction algorithm, such as Hough transform, is used for the generated palmprint coded edge image set. Hough transform converts the straight line detection problem in the image space into a peak detection problem in the parameter space. For each edge point in the palmprint, the corresponding curve in the Hough transform parameter space is calculated according to its coordinate value. When multiple curves intersect at a certain point in the parameter space, it indicates that there is a straight line passing through these edge points in the image space, and the parameter value of the intersection is the characteristic parameter of the straight line (such as slope and intercept). By detecting the peak value in the parameter space, the straight line features such as the main line and wrinkles in the palmprint are determined, and the attribute information such as the direction, length, curvature, and spacing of the line features are calculated. In addition, other line feature extraction methods, such as a line tracking algorithm based on a chain code, can be combined to further refine and supplement the line feature information of the palmprint, thereby obtaining a palmprint coded multidimensional feature set. Palm print features can reflect the texture distribution and direction of the palm surface, and complement the palm vein features to form a unique biometric identifier of the palm, further improving the accuracy and reliability of identity authentication.

[0032] Step S300, align and fuse the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set in the spatial domain to obtain a palm coding fusion feature data set. Specifically, feature point selection and registration are performed, and representative feature points such as bifurcation points and intersection points are selected from the palm vein coding multidimensional feature set, and their local feature descriptors are combined to ensure the discrimination; key feature points such as main line endpoints are selected from the palm print coding multidimensional feature set and the corresponding descriptors are used to enhance the recognizability, and then the feature point sets of the two are matched by using algorithms such as ICP or feature descriptor matching combined with RANSAC algorithm to determine the coding matching feature point set. The palm vein and palm print coding multidimensional feature sets are processed by dimensionality reduction algorithms such as PCA or LDA to obtain reduced dimensionality feature sets, and normalized by methods such as Min-Max or Z-Score to avoid the influence of numerical differences. Finally, a fusion algorithm is selected based on the set of encoded matching feature points and the normalized dimensionality reduction feature set, such as the weighted average method to add weights according to importance, the feature splicing method to directly splice, or the multi-core learning-based fusion algorithm to use multi-core function mapping and then fusion, so as to obtain the palm encoding fusion feature data set, which provides a strong data foundation for subsequent identity recognition.

[0033] In a possible implementation, the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set are aligned and fused in the spatial domain to obtain a palm coding fusion feature data set. Step S300 further includes step S310, respectively selecting feature points from the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain a palm vein coding feature point set and a palm print coding feature point set. Specifically, in the palm vein coding multidimensional feature set, a feature point detection algorithm is used, such as an algorithm based on scale-invariant feature transform (SIFT) or speeded up robust features (SURF). The algorithm searches for points with significant local features in different scale spaces of the image. For palm vein images, it focuses on points with obvious geometric features on the vein structure, such as bifurcation points, intersection points, and relatively clear endpoints of the vein. By calculating the gradient information and local feature descriptors of the pixels around the point, its uniqueness and stability are determined, so that points that can effectively represent the palm vein features are screened out to form a palm vein coding feature point set. The feature points not only contain the morphological information of the palm veins, but also have a certain anti-interference ability, and can maintain relative consistency under different acquisition conditions, providing a reliable basis for subsequent feature registration and fusion. For the multi-dimensional feature set of palmprint encoding, a detection method suitable for palmprint features is adopted, such as the endpoint and intersection detection algorithm based on the palmprint line features. Since the palmprint is mainly composed of main lines, wrinkles and detail lines, the endpoints, intersections and points with large curvature changes are selected as feature points. By analyzing the grayscale changes and texture direction of the palmprint image, the position of the feature points is determined, and their related attributes, such as direction, curvature, etc., are calculated as feature description information, and then a set of palmprint encoding feature points is constructed. Palmprint feature points can accurately reflect the texture direction and distribution characteristics of the palmprint, complement each other with palm vein feature points, and are used together for subsequent precise matching and fusion operations to improve the accuracy and reliability of the entire identity authentication system.

[0034] Step S320, feature registration is performed on the palm vein coding feature point set and the palm print coding feature point set to obtain a coding matching feature point set. Specifically, a feature registration algorithm, such as an iterative closest point (ICP) algorithm based on feature points or a matching algorithm based on feature descriptors (such as determining the matching relationship by calculating the Euclidean distance between feature points), is used to perform a registration operation on the palm vein coding feature point set and the palm print coding feature point set. In the two feature point sets, the initial corresponding point pairs are found, which can be achieved by comparing the local feature descriptors of the feature points or searching under certain geometric constraints. Based on the initial corresponding points, the positions and postures of the feature points in the two sets are continuously adjusted by iterative optimization to minimize the matching error between them. In the iterative process, a transformation model (such as a rigid transformation, an affine transformation or a non-rigid transformation model, depending on the actual deformation of the palm vein and palm print features) is calculated based on the matched point pairs, and the transformation is applied to the unmatched points, and the matching error is evaluated again until the preset convergence condition is met (such as the matching error is less than a certain threshold or the number of iterations reaches an upper limit). Finally, a set of accurately registered coded matching feature points is obtained. The point pairs reflect the spatial correspondence between palm vein and palm print features, providing key alignment information for subsequent feature fusion, ensuring that the fused feature data set can fully integrate the advantages of the two biometric features and improve the accuracy and stability of identity recognition.

[0035] Step S330, feature dimensionality reduction is performed on the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain the palm vein coding dimensionality reduction feature set and the palm print coding dimensionality reduction feature set. Specifically, the palm vein coding multidimensional feature set is processed by dimensionality reduction techniques such as principal component analysis (PCA) or linear discriminant analysis (LDA). The PCA method calculates the covariance matrix of the feature data, solves its eigenvalues ​​and eigenvectors, and projects the original high-dimensional feature vectors into a low-dimensional subspace composed of the main eigenvectors, thereby achieving data dimensionality reduction. In this process, the first few main eigenvectors are selected according to the size of the eigenvalues, so that the projected data can retain most of the variance information of the original data, which not only retains the main components of the palm vein features, but also reduces the dimension and computational complexity of the data. LDA is based on analyzing the category information of the data (if there is a priori category label), and finds the projection direction that can maximize the distinction between different categories. By maximizing the ratio of inter-class divergence and minimizing intra-class divergence, the optimal projection matrix is ​​determined, and the palm vein coding multidimensional feature set is projected into a low-dimensional space to obtain the palm vein coding reduced dimension feature set. While reducing the dimension, the discriminative ability of the feature is further enhanced, which is beneficial to the subsequent identity recognition task. PCA or LDA algorithm is used to reduce the dimension of the palm print coding multidimensional feature set. According to the characteristics and distribution of the palm print feature data, appropriate parameters and methods are selected for dimensionality reduction processing. For example, for the multidimensional feature set composed of palm print line features and texture features, PCA can effectively extract its main texture change mode and line feature direction information, and convert it into a low-dimensional feature representation, while LDA can highlight the differences between different individual palm print features in the case of category information, and obtain the palm print coding reduced dimension feature set, so that the palm print features after dimensionality reduction not only retain the key information, but also facilitate fusion and subsequent processing with the palm vein reduced dimension features.

[0036] Step S340, based on the set of coding matching feature points, the palm vein coding dimensionality reduction feature set and the palm print coding dimensionality reduction feature set are aligned and fused in the spatial domain to obtain the palm coding fusion feature data set. Specifically, according to the correspondence between the palm vein and palm print feature points determined by the previously obtained coding matching feature point set, the palm vein coding dimensionality reduction feature set and the palm print coding dimensionality reduction feature set are aligned and fused in the spatial domain. First, based on the coding matching feature points, the palm vein and palm print feature vectors after dimensionality reduction are adjusted and matched to ensure their consistency in spatial position. Then, a suitable fusion algorithm is selected, such as the weighted average method, the feature splicing method, etc. For example, the weighted average method assigns corresponding weights to the palm vein and palm print features according to their importance and contribution in identity recognition. For example, through experimental analysis or prior knowledge, the weight of the palm vein feature is determined to be w1, and the weight of the palm print feature is w2 (w1+w2=1). Each feature vector in the palm vein coding reduced dimension feature set and the corresponding element of the palm print coding reduced dimension feature vector are multiplied by their respective weights and then added to obtain a fused feature vector, thereby forming a palm coding fusion feature data set. This method can balance the effects of the two features to a certain extent, highlight the influence of important features, and suppress noise and unreliable feature information. The feature concatenation method concatenates the palm vein coding reduced dimension feature vector and the palm print coding reduced dimension feature vector in a certain order to form a new longer feature vector as the fused feature representation. For example, if the dimension of the palm vein coding reduced dimension feature vector is m and the dimension of the palm print coding reduced dimension feature vector is n, the dimension of the concatenated fusion feature vector is m+n. This method is simple and direct, and can retain the original information of the two features, but it may result in a higher dimension of the feature vector. It is necessary to further consider dimensionality reduction or feature selection in subsequent processing to avoid the problem of dimensionality disaster. At the same time, attention should be paid to the splicing order and dimensionality matching to ensure the accuracy and availability of the fused feature data set.

[0037] Step S400, based on the palm code fusion feature data set, encryption processing and network training and tuning are performed to obtain a palm identity recognition adaptive network. Specifically, encryption processing is performed, and an encryption algorithm (such as AES, RSA or homomorphic encryption algorithm, etc.) is selected according to security and computing resource conditions, and the feature vector of the palm code fusion feature data set is encrypted with a randomly generated symmetric key (such as AES algorithm) or a public-private key mechanism (such as RSA algorithm), and measures such as adding a message authentication code are taken to ensure data security and integrity, and a hardware security module is used to manage the key when necessary. A network architecture based on deep learning (such as GAN, including a generator and a discriminator) is built, and samples are randomly selected from the encrypted data set at a ratio of 80%, 10%, and 10% to be divided into a training set, a validation set, and a test set. When training the generator, a random noise vector is input to generate fake data, which is then input into the discriminator together with the decrypted real data. The generator adjusts parameters based on the feedback from the discriminator through back propagation and optimization algorithms (such as Adam) and loss functions (such as Wasserstein distance) to improve fidelity; when training the discriminator, the parameters are optimized based on the difference from the real label through back propagation and loss functions such as binary cross entropy to improve discrimination. During training, a network balancing mechanism is constructed, and the training process is stabilized based on methods such as WGAN gradient penalty. The network layers, number of nodes, activation functions, regularization methods, and optimizer parameters are adjusted regularly according to indicators such as accuracy and recall until the performance of the validation set is good, thereby obtaining an adaptive network for palm identity recognition.

[0038] In a possible implementation, encryption processing and network training and tuning are performed based on the palm code fusion feature data set to obtain a palm identity recognition adaptive network, and step S400 further includes step S410, adding an information encryption mask to the palm code fusion feature data set. Specifically, an information encryption mask generation algorithm is designed, and the algorithm can create a mask based on random number generation, hash function or specific encryption key stream. For example, a pseudo-random number generator is used to generate a random matrix with the same dimension as the palm code fusion feature data set as a mask, or a specific encryption key is hashed and the hash result is expanded into a mask vector matching the dimension of the data set. The generated mask is subjected to an element-by-element XOR operation with the palm code fusion feature data set, so that each eigenvalue in the data set is masked by the mask, thereby realizing encryption and hiding of the data. The purpose is to further enhance the security of the data when the data is used to train the network, prevent potential information leakage, and also add a certain degree of randomness and complexity to the subsequent network training, which helps to improve the generalization ability and anti-attack ability of the network.

[0039] Step S420, sample extraction training is performed based on the palm encoding fusion feature data set after adding the information encryption mask to obtain a palm identity recognition discriminator. Specifically, samples are extracted from the palm encoding fusion feature data set after adding the information encryption mask according to a certain sampling strategy. Random sampling, stratified sampling or cluster-based sampling and other methods can be used to ensure that the extracted samples can fully represent the feature distribution of the entire data set. For example, if the data set contains palm feature data of users of different ages, genders, races, etc., stratified sampling can be performed according to attributes so that each subgroup has an appropriate proportion in the training sample. The extracted samples are divided into a training subset and a validation subset, the training subset is used to train the discriminator, and the validation subset is used to evaluate the performance of the discriminator and prevent overfitting. Construct a discriminator model based on deep learning, for example, a multi-layer perceptron (MLP), a convolutional neural network (CNN) or other suitable classification model architecture can be used. The encrypted samples in the training subset are input into the discriminator, which extracts features and classifies each sample, and outputs a probability value, indicating the possibility that the sample belongs to real data (palm features of registered users). During the training process, according to the difference between the output of the discriminator and the real label of the sample (known real data is marked as 1, and the assumed false data is marked as 0), the gradient of the loss function (such as binary cross entropy loss function) is calculated through the back propagation algorithm, and the weight and bias parameters of the discriminator are updated using an optimization algorithm (such as stochastic gradient descent, Adagrad, Adadelta or Adam, etc.), and the decision boundary of the discriminator is continuously adjusted so that it can more accurately distinguish between real data and false data. After multiple iterations of training, until the discriminator achieves good performance on the verification subset, such as accuracy, recall rate, F1 value and other indicators meet certain threshold requirements, a palm identity recognition discriminator is obtained.

[0040] Step S430, random noise perturbation training is performed on the palm code fusion feature data set to obtain a palm identity recognition generator. Specifically, random noise perturbation is introduced on the original palm code fusion feature data set to train the generator. For each sample in the data set, a random noise vector with a specific distribution (such as Gaussian distribution, uniform distribution, etc.) is generated, and it is added to the sample feature vector or other combination operations are performed to form perturbed sample data. For example, for each feature dimension, a noise value is sampled from a Gaussian distribution according to a certain noise standard deviation, and added to the feature value of the corresponding dimension to obtain a perturbed sample. A generator model is constructed, and a neural network-based architecture can also be used, such as a deconvolutional network (DCGAN) in a generative adversarial network or a generator structure based on an MLP. The perturbed sample data is input into the generator. The goal of the generator is to generate fake sample data that is as similar as possible to the real palm code fusion feature data by learning the distribution of these noisy data to deceive the discriminator. Through the back-propagation algorithm, the gradient of the generator's loss function (such as mean square error, cross entropy, etc.) is calculated based on the difference between the output result of the fake sample generated by the generator in the discriminator (the probability of being judged as real data) and the target value (ideally, we hope that the discriminator will misjudge it as real data, that is, the probability value is close to 1), and the optimization algorithm is used to update the generator's parameters, and the generator's generation strategy is continuously adjusted to make the fake samples it generates more and more realistic. After multiple iterative training, until the generator is able to generate fake sample data with high deception in the discriminator, a palm identity recognition generator is obtained.

[0041] Step S440, constructing a network balancing mechanism, and performing alternating iterative training and verification tuning on the palm identity recognition discriminator and the palm identity recognition generator based on the network balancing mechanism to obtain a palm identity recognition adaptive network. Specifically, a network balancing mechanism is designed, the core idea of ​​which is to dynamically adjust the training intensity and optimization direction of the discriminator and the generator during the training process to ensure the confrontational balance between the two. For example, the gradient penalty mechanism in the GAN training method based on the Wasserstein distance can be adopted, and a penalty term for the discriminator gradient is added to the loss function of the discriminator to limit the decision boundary of the discriminator from being too steep or irregular, so that the generator can have enough space to learn and improve, and avoid the discriminator from having too high a discriminative ability for the false samples generated by the generator too early, causing the training to fall into a local optimal solution. During the training process, the discriminator and the generator are trained in an alternating iterative manner. The parameters of the generator are fixed, and the discriminator is trained for multiple iterations. The parameters of the discriminator are updated according to the loss function of the discriminator using real data samples and false sample data generated by the generator to improve its discrimination ability. The parameters of the discriminator are fixed, and the generator is trained iteratively. According to the loss function of the generator and the feedback of the discriminator on its generated samples, the parameters of the generator are updated so that the samples it generates can better deceive the discriminator. In each iteration, the performance of the discriminator and the generator on the validation set is evaluated, such as calculating the accuracy and recall rate of the discriminator and the deception success rate of the samples generated by the generator. According to the changes in the indicators, the training parameters such as the learning rate, the number of iterations, the weight of the loss function, etc. are dynamically adjusted to optimize the overall performance of the network. Through multiple alternating iterative training and verification tuning, the discriminator and the generator are constantly competing and co-evolving with each other, and finally a stable and powerful palm identity recognition adaptive network is obtained. The network can accurately identify whether the input palm feature data belongs to a registered user, and has good adaptability and robustness to different palm features, effectively improving the accuracy and security of identity authentication.

[0042] In a possible implementation, a network balancing mechanism is constructed, and the palm identity recognition discriminator and the palm identity recognition generator are alternately iteratively trained and verified and tuned based on the network balancing mechanism to obtain a palm identity recognition adaptive network, and step S440 further includes step S441, determining a network performance loss function and an alternating training strategy according to the network balancing mechanism. Specifically, in determining the network performance loss function, for the palm identity recognition discriminator, since it undertakes the binary classification task of distinguishing between real and generated palm encoding fusion feature data, a binary cross entropy loss function is often used, specifically, the predicted probability of the discriminator output for real samples and the real label 1 and the predicted probability of the discriminator output for false samples and the real label 0 are substituted into the formula to calculate the loss value, and the smaller the value, the more accurately the discriminator can distinguish between real and false samples. As for the palm identity generator, its goal is to generate realistic data to deceive the discriminator. The mean square error (MSE) or the loss function based on the Wasserstein distance improvement is often used. For example, MSE measures the fidelity of the generated samples by calculating the mean of the square of the difference between the probability value of the generated sample output by the discriminator and the target value 1. The smaller the MSE value, the better the performance. At the same time, according to the network balance mechanism, a gradient penalty term is added to the discriminator loss function in Wasserstein_GAN to maintain the network balance. In determining the alternating training strategy, the training round rules are first formulated, the total number of training iterations is set and divided into multiple small training cycles. In each cycle, the generator parameters are fixed first, and the discriminator is allowed to perform 1 to 5 (adjustable) training iterations to improve the discrimination ability. Then the discriminator parameters are fixed, and the generator is allowed to perform 1 to 3 (settable) training iterations to generate more realistic samples. Repeated alternating training gradually approaches the optimal state of the network within the total number of iterations, improves the overall network performance, and meets the stable and effective training effect expected by the network balance mechanism.

[0043] Step S442, based on the network performance loss function and the alternating training strategy, the palm identity recognition discriminator and the palm identity recognition generator are subjected to alternating iterative training and performance evaluation verification to obtain the discriminator loss parameters and the generator loss parameters. Specifically, the training is started according to the established alternating training strategy, and the discriminator is first trained iteratively. From the training set of real data containing encryption masks and false samples created by the generator, data is input into the discriminator in appropriate batches (such as 32 or 64, etc., depending on the hardware), which extracts features and classifies them according to the architecture of multi-layer convolutional neural network, fully connected layer, etc., and outputs the probability value of the sample being real data. Combined with the binary cross entropy loss function, the discriminator loss value of the batch is calculated according to the output probability and the real label (real is 1, false is 0), and then the back propagation algorithm is used to update its weights and biases to optimize the structure and improve the ability to distinguish true from false. After completion, the discriminator parameters are fixed according to the strategy and the generator is trained. The generator receives a random noise vector of pre-designed dimensions and distribution (such as Gaussian distribution), generates false samples through an architecture containing deconvolution layers and activation functions, and inputs it into a discriminator with fixed parameters to obtain a probability value. The generator calculates the loss according to the selected loss function (such as MSE) based on the difference between this value and the target value 1, and then uses back propagation to update the parameters to improve the generated samples to deceive the discriminator. The discriminator and generator are repeatedly trained alternately in multiple training cycles. After completing a specific number of alternating iterative trainings (such as 10 or 20 times), samples are taken from an independent validation set (including real samples and false samples at different stages of the generator) and input into the discriminator and generator respectively to evaluate the validation performance. For the discriminator, calculate its accuracy (the ratio of the sum of the number of correctly judged real and false samples to the total number of verification samples), recall (the ratio of the number of correctly judged real samples to the actual number of real samples), F1 value (calculated by the accuracy and recall formulas) and other indicators on the validation set to reflect its actual ability to distinguish between true and false; for the generator, evaluate the probability that its generated samples are misclassified as real samples by the discriminator (that is, the proportion of the discriminator output probability value close to 1). The higher the proportion, the more realistic and deceptive it is. At the same time, record the loss parameters of the discriminator and generator (such as the mean of binary cross entropy and MSE on the validation set) to provide a basis for subsequent gradient balance updates.

[0044] Step S443, based on the discriminator loss parameter and the generator loss parameter, the palm identity recognition discriminator and the palm identity recognition generator are gradient balanced and updated to obtain the target palm identity recognition discriminator and the target palm identity recognition generator. Specifically, a loss parameter difference analysis is performed, and by comparing the discriminator loss parameter and the generator loss parameter, attention is paid to the change trend and relative size of the two. When the discriminator loss parameter continues to decrease and the value is small, while the generator loss parameter decreases slowly or the value is large, it means that the discriminator learning ability is too strong, making it difficult for the generator to be effectively improved. At this time, it is necessary to appropriately adjust the learning rate of the two or limit the discriminator training intensity to balance the adversarial relationship; on the contrary, if the generator loss parameter decreases too fast, its generated samples are easy to deceive the discriminator, so that the discriminator cannot accurately grasp the difference in real sample characteristics, then the generator training must be adjusted, such as increasing the number of discriminator training times or changing the generator loss function weights. Implement the gradient balance update operation. Based on the above analysis results, use adaptive learning rate adjustment mechanisms such as those based on optimization algorithms such as Adam and Adagrad, and flexibly change the learning rates of the discriminator and generator in combination with the current loss parameter status. For example, if the discriminator performs too well, its learning rate will be appropriately reduced to slow down the pace of parameter updates, creating more learning improvement opportunities for the generator; if the fidelity of the samples generated by the generator is insufficient, its learning rate will be appropriately increased to accelerate parameter updates so that it can produce more deceptive samples faster. In addition, in addition to the learning rate adjustment, the gradients of the two may be directly constrained or corrected according to other rules in the network balance mechanism. For example, in some gradient penalty methods, if the discriminator gradient is abnormal (too large or too small) in a specific area, the corresponding penalty term is added to the gradient calculation of its loss function according to the preset gradient penalty formula, and then the gradient is corrected when the parameters are updated, making the discriminator learning process more stable and reasonable. After this series of gradient balancing update operations based on the loss parameters, the updated target palm identity discriminator and generator can be obtained. Their performance is more coordinated under the balancing mechanism, laying a good foundation for subsequent combined performance tuning.

[0045] Step S444, the target palm identity recognition discriminator and the target palm identity recognition generator are combined for performance tuning to obtain the palm identity recognition adaptive network. Specifically, the target palm identity recognition discriminator and the target palm identity recognition generator are first combined according to a predetermined network architecture. Under the framework of the generative adversarial network (GAN), the two operate in coordination through a specific input-output connection method. The samples produced by the generator are directly input into the discriminator, and the discrimination results of the discriminator are fed back to the generator to guide its parameter update, forming a closed-loop adversarial learning system, thereby constructing a complete palm identity recognition adaptive network structure. Then, the verification data set and possible test data sets are used again to conduct a comprehensive performance evaluation of the combined network, focusing on its accuracy, recall rate, F1 value and other indicators in the identity recognition task, while paying attention to the deceptiveness of the samples generated by the generator and the overall stability of the network. Subsequently, the network is finally tuned based on the evaluation results, such as fine-tuning network hyperparameters (including the number of layers of the discriminator and generator, the number of neurons in each layer, the type of activation function, etc.), further optimizing the weight of the loss function, or adjusting the training strategy (such as batch size, total number of training iterations, etc.). Continue to adjust and optimize until the network achieves ideal results in multiple performance indicators, such as an accuracy rate of more than 90% on the validation set (depending on the application scenario), the samples generated by the generator are highly deceptive and the network performs stably with different input samples. At this time, the palm identity recognition adaptive network that can be applied to actual intelligent identity authentication scenarios is successfully obtained. It can efficiently use the palm encoding fusion feature data to achieve accurate identity recognition and has good adaptability and robustness.

[0046] Step S500, collect and obtain the target palm image of the target user, and perform identity verification on the target palm image based on the palm identity recognition adaptive network. Specifically, the target palm image is collected, and an image acquisition device with the same specifications as before is selected, placed in an environment with uniform and stable light and no reflective shadows, and the acquisition angle and distance are set. After guiding the target user to place his palm in a standard posture, the image with the best quality is collected at a specific frame rate and selected, and pre-processed by grayscale conversion and filtering denoising. Then it is input into the palm identity recognition adaptive network, and the network first uses the semantic segmentation model to mark the palm vein and palm print area, and uses the morphology and feature point extraction algorithm to obtain the encoded multidimensional feature set for the palm vein, and the edge detection and line feature extraction algorithm to obtain the encoded multidimensional feature set for the palm print, and then uses the fusion algorithm during training to obtain the target palm encoded fusion feature data, which is input into the discriminator. If the output probability exceeds the preset threshold (such as 0.95), the identity authentication is passed and authorization is given, otherwise the verification fails and access is denied and an alarm is triggered, and the verification information is recorded for subsequent analysis and optimization to ensure the safety of the system and users.

[0047] The embodiment of the present application collects the palm image and identity data of the user to be authenticated, and encodes the palm image based on the identity data to obtain a palm identity encoding image set. Then, through region marking and feature extraction, the multi-dimensional features of palm veins and palm prints are obtained. Then, the features of the two are spatially aligned and fused to generate a palm encoding fusion feature data set. Then, through encryption processing and network training, an adaptive palm identity recognition network is obtained. Finally, the palm image of the target user is collected, and the network is used to authenticate the identity. By fusing the palm vein and palm print features, the technical effect of improving the security, accuracy and anti-interference ability of identity authentication is achieved.

[0048] In the above, refer to Figure 1 The intelligent identity authentication method based on the fusion of palm vein and palm print features according to an embodiment of the present invention is described in detail. Figure 2 An intelligent identity authentication system that integrates palm vein and palm print features according to an embodiment of the present invention is described.

[0049] The intelligent identity authentication system integrating palm vein and palm print features according to the embodiment of the present invention is used to solve the technical problems that the existing identity authentication method relies on a single biometric feature, resulting in insufficient security, low accuracy and poor anti-interference ability. By integrating palm vein and palm print features, the technical effect of improving the security, accuracy and anti-interference ability of identity authentication is achieved. The intelligent identity authentication system integrating palm vein and palm print features includes: a palm identity coding image set acquisition module 10, a coding multi-dimensional feature set acquisition module 20, a spatial domain alignment fusion module 30, an adaptive network acquisition module 40, and an identity recognition and verification module 50.

[0050] The palm identity coding image set acquisition module 10 is used to sequentially acquire a palm image set and a user identity data set of a user set to be authenticated, and perform identity coding on the palm image set based on the user identity data set to obtain a palm identity coding image set.

[0051] The coding multi-dimensional feature set acquisition module 20 is used to perform region marking and feature extraction on the palm identity coding image set to obtain a palm vein coding multi-dimensional feature set and a palm print coding multi-dimensional feature set.

[0052] The spatial domain alignment and fusion module 30 is used to perform spatial domain alignment and fusion on the palm vein code multidimensional feature set and the palm print code multidimensional feature set to obtain a palm code fusion feature data set.

[0053] The adaptive network acquisition module 40 is used to perform encryption processing and network training and optimization based on the palm encoding fusion feature data set to obtain a palm identity recognition adaptive network.

[0054] The identity recognition and verification module 50 is used to collect and obtain a target palm image of a target user, and perform identity recognition and verification on the target palm image based on the palm identity recognition adaptive network.

[0055] The specific configuration of the palm identity coding image set acquisition module 10 will be described in detail below. As described above, the palm image set and the user identity data set of the user set to be authenticated are collected and acquired in sequence, and the palm image set is identity-encoded based on the user identity data set to obtain the palm identity coding image set. The palm identity coding image set acquisition module 10 further includes: an association mapping unit, the association mapping unit is used to associate and map the user identity data set with the palm image set to build a user identity palm image database; an identity element extraction unit, the identity element extraction unit is used to extract identity elements from the user identity data set to obtain an identity key element set, and construct an identity coding system based on the identity key element set; a sensitive data identification unit, the sensitive data identification unit is used to identify sensitive data of the user identity palm image database to obtain user identity sensitive data, and desensitize and encrypt the user identity sensitive data to obtain an identity encrypted palm image database; an identity coding unit, the identity coding unit is used to perform identity coding on the identity encrypted palm image database based on the identity coding system to obtain the palm identity coding image set.

[0056] Next, the specific configuration of the encoding multi-dimensional feature set acquisition module 20 will be described in detail. As described above, the palm identity coding image set is subjected to region marking and feature extraction to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set. The coding multidimensional feature set acquisition module 20 further includes: a denoised palm identity coding image set acquisition unit, the denoised palm identity coding image set acquisition unit is used to perform noise recognition and filtering processing on the palm identity coding image set to obtain a denoised palm identity coding image set; a segmentation model building unit, the segmentation model building unit is used to build a palm semantic segmentation model, and based on the palm semantic segmentation model, the denoised palm identity coding image set is sequentially semantically segmented to obtain a palm semantic segmentation result; a regional anchor frame marking unit, the regional anchor frame marking unit is used to perform regional anchor frame marking on the denoised palm identity coding image set according to the palm semantic segmentation result to obtain a palm coding image region set of interest; and an associated feature extraction unit, the associated feature extraction unit is used to perform associated feature extraction on the palm coding image region set of interest to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set.

[0057] Among them, the palm coding image region set of interest is subjected to associated feature extraction to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set, and the associated feature extraction unit further includes: a grayscale image set acquisition subunit, the grayscale image set acquisition subunit is used to perform grayscale image conversion and equalization processing on the palm coding image region set of interest to obtain a palm coding region grayscale image set; a palm vein coding multidimensional feature set acquisition subunit, the palm vein coding multidimensional feature set acquisition subunit is used to perform morphological operations and feature point extraction on the palm coding region grayscale image set respectively to obtain the palm vein coding multidimensional feature set; an edge image set generation subunit, the edge image set generation subunit is used to perform grayscale difference calculation and edge point marking based on the palm coding region grayscale image set to generate a palm print coding edge image set; a multidimensional line feature extraction subunit, the multidimensional line feature extraction subunit is used to perform multidimensional line feature extraction on the palm print coding edge image set in sequence to obtain the palm print coding multidimensional feature set.

[0058] Next, the specific configuration of the spatial domain alignment and fusion module 30 will be described in detail. As described above, the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set are spatially aligned and fused to obtain a palm coding fusion feature data set. The spatial domain alignment and fusion module 30 further includes: a feature point selection unit, the feature point selection unit is used to respectively select feature points from the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain a palm vein coding feature point set and a palm print coding feature point set; a feature registration unit, the feature registration unit is used to perform feature registration on the palm vein coding feature point set and the palm print coding feature point set to obtain a coding matching feature point set; a feature dimension reduction unit, the feature dimension reduction unit is used to perform feature dimension reduction on the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain a palm vein coding reduced dimension feature set and a palm print coding reduced dimension feature set; a spatial domain alignment and fusion unit, the spatial domain alignment and fusion unit is used to perform spatial domain alignment and fusion on the palm vein coding reduced dimension feature set and the palm print coding reduced dimension feature set based on the coding matching feature point set to obtain the palm coding fusion feature data set.

[0059] The specific configuration of the adaptive network acquisition module 40 will be described in detail below. As described above, based on the palm encoding fusion feature data set, encryption processing and network training and tuning are performed to obtain a palm identity recognition adaptive network. The adaptive network acquisition module 40 further includes: an information encryption mask adding unit, the information encryption mask adding unit is used to add an information encryption mask to the palm encoding fusion feature data set; a sample extraction training unit, the sample extraction training unit is used to perform sample extraction training based on the palm encoding fusion feature data set after adding the information encryption mask, to obtain a palm identity recognition discriminator; a random noise perturbation training unit, the random noise perturbation training unit is used to perform random noise perturbation training on the palm encoding fusion feature data set to obtain a palm identity recognition generator; a network balance mechanism construction unit, the network balance mechanism construction unit is used to construct a network balance mechanism, based on the network balance mechanism, the palm identity recognition discriminator and the palm identity recognition generator are alternately iterated trained and verified and tuned to obtain a palm identity recognition adaptive network.

[0060] Among them, a network balancing mechanism is constructed, and the palm identity recognition discriminator and the palm identity recognition generator are alternately iteratively trained and verified and tuned based on the network balancing mechanism to obtain a palm identity recognition adaptive network. The network balancing mechanism construction unit further includes: a strategy determination subunit, which is used to determine a network performance loss function and an alternating training strategy according to the network balancing mechanism; a loss parameter acquisition subunit, which is used to perform alternating iterative training on the palm identity recognition discriminator and the palm identity recognition generator based on the network performance loss function and the alternating training strategy. Alternate iterative training and performance evaluation verification to obtain discriminator loss parameters and generator loss parameters; a gradient balance update subunit, the gradient balance update subunit is used to perform gradient balance update on the palm identity recognition discriminator and the palm identity recognition generator based on the discriminator loss parameters and the generator loss parameters to obtain a target palm identity recognition discriminator and a target palm identity recognition generator; a combined performance tuning subunit, the combined performance tuning subunit is used to perform combined performance tuning on the target palm identity recognition discriminator and the target palm identity recognition generator to obtain the palm identity recognition adaptive network.

[0061] The intelligent identity authentication system for fusing palm vein and palm print features provided in the embodiment of the present invention can execute the intelligent identity authentication method for fusing palm vein and palm print features provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0062] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0063] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An intelligent identity authentication method based on the fusion of palm vein and palm print features, characterized in that: The method comprises: Sequentially acquiring a palm image set and a user identity data set of a user set to be authenticated, and performing identity encoding on the palm image set based on the user identity data set to obtain a palm identity encoding image set; Performing region marking and feature extraction on the palm identity coding image set to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set; Aligning and fusing the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set in the spatial domain to obtain a palm coding fusion feature data set; Perform encryption processing and network training optimization based on the palm encoding fusion feature data set to obtain a palm identity recognition adaptive network; A target palm image of a target user is acquired, and identity recognition and verification is performed on the target palm image based on the palm identity recognition adaptive network.

2. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 1, characterized in that: The step of obtaining a palm identity coding image set comprises: Associatively mapping the user identity data set and the palm image set to build a user identity palm image database; Extracting identity elements from the user identity data set to obtain an identity key element set, and constructing an identity coding system based on the identity key element set; Performing sensitive data identification on the user identity palm image database to obtain user identity sensitive data, and desensitizing and encrypting the user identity sensitive data to obtain an identity encrypted palm image database; The identity-encrypted palm image database is identity-encoded based on the identity coding system to obtain the palm identity-encoded image set.

3. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 1, characterized in that: The step of obtaining a palm vein code multidimensional feature set and a palm print code multidimensional feature set includes: Performing noise identification and filtering processing on the palm identity coding image set to obtain a denoised palm identity coding image set; Building a palm semantic segmentation model, and performing semantic segmentation on the denoised palm identity coding image set in sequence based on the palm semantic segmentation model to obtain a palm semantic segmentation result; According to the palm semantic segmentation result, the denoised palm identity coding image set is marked with regional anchor frames to obtain a palm coding image region set of interest; The associated features are extracted from the palm coding image region of interest set to obtain a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set.

4. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 3, characterized in that: The method of obtaining a palm vein coding multidimensional feature set and a palm print coding multidimensional feature set comprises: Performing grayscale image conversion and equalization processing on the palm coding image region set of interest to obtain a palm coding region grayscale image set; Performing morphological operations and feature point extraction on the grayscale image set of the palm coding area respectively to obtain the multi-dimensional feature set of the palm vein coding; Perform grayscale difference calculation and edge point marking based on the palm coding area grayscale image set to generate a palmprint coding edge image set; Multi-dimensional line feature extraction is performed on the palmprint code edge image set in sequence to obtain the palmprint code multi-dimensional feature set.

5. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 1, characterized in that: The palm encoding fusion feature dataset is obtained, including: Selecting feature points from the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set respectively to obtain a palm vein coding feature point set and a palm print coding feature point set; Performing feature registration on the palm vein coding feature point set and the palm print coding feature point set to obtain a coding matching feature point set; Performing feature dimensionality reduction on the palm vein code multidimensional feature set and the palm print code multidimensional feature set to obtain a palm vein code dimensionality reduction feature set and a palm print code dimensionality reduction feature set; Based on the encoding matching feature point set, the palm vein encoding dimensionality reduction feature set and the palm print encoding dimensionality reduction feature set are spatially aligned and fused to obtain the palm encoding fusion feature data set.

6. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 1, characterized in that: The method of obtaining a palm identity recognition adaptive network comprises: Adding an information encryption mask to the palm encoding fusion feature data set; Performing sample extraction training based on the palm encoding fusion feature data set after adding the information encryption mask to obtain a palm identity recognition discriminator; Performing random noise perturbation training on the palm encoding fusion feature data set to obtain a palm identity recognition generator; A network balancing mechanism is constructed, and based on the network balancing mechanism, the palm identity recognition discriminator and the palm identity recognition generator are alternately iteratively trained and verified and tuned to obtain a palm identity recognition adaptive network.

7. The intelligent identity authentication method of palm vein and palm print feature fusion as claimed in claim 6, characterized in that: The method of obtaining a palm identity recognition adaptive network comprises: Determining a network performance loss function and an alternating training strategy according to the network balancing mechanism; Based on the network performance loss function and the alternating training strategy, the palm identity recognition discriminator and the palm identity recognition generator are subjected to alternating iterative training and performance evaluation verification to obtain a discriminator loss parameter and a generator loss parameter; Based on the discriminator loss parameter and the generator loss parameter, performing gradient balancing update on the palm identity recognition discriminator and the palm identity recognition generator to obtain a target palm identity recognition discriminator and a target palm identity recognition generator; The target palm identity recognition discriminator and the target palm identity recognition generator are combined for performance tuning to obtain the palm identity recognition adaptive network.

8. An intelligent identity authentication system integrating palm vein and palm print features, characterized in that: The system is used to implement the intelligent identity authentication method for fusing palm vein and palm print features as described in any one of claims 1 to 7, and the system comprises: A palm identity coding image set acquisition module, the palm identity coding image set acquisition module is used to sequentially acquire a palm image set and a user identity data set of a user set to be authenticated, and perform identity coding on the palm image set based on the user identity data set to obtain a palm identity coding image set; A coding multi-dimensional feature set acquisition module, which is used to perform region marking and feature extraction on the palm identity coding image set to obtain a palm vein coding multi-dimensional feature set and a palm print coding multi-dimensional feature set; A spatial domain alignment and fusion module, wherein the spatial domain alignment and fusion module is used to perform spatial domain alignment and fusion on the palm vein coding multidimensional feature set and the palm print coding multidimensional feature set to obtain a palm coding fusion feature data set; An adaptive network acquisition module, the adaptive network acquisition module is used to perform encryption processing and network training and optimization based on the palm encoding fusion feature data set to obtain a palm identity recognition adaptive network; The identity recognition and verification module is used to collect and obtain a target palm image of a target user, and perform identity recognition and verification on the target palm image based on the palm identity recognition adaptive network.

Images