Multi-dimensional biometric identification and identity authentication method and system for digital payment

Abstract

The application provides a multi-dimensional biological feature recognition and identity authentication method and system for digital payment, relates to the technical field of payment security, and comprises the following steps: collecting first biological feature data of a user and extracting a feature vector to generate a multi-dimensional feature set; selecting a corresponding number of biological features according to a payment amount to generate a to-be-authenticated feature set; obtaining second biological feature data of the user and extracting an authentication feature vector; calculating an environment credibility according to an identity authentication environment parameter and dynamically adjusting a similarity threshold; and performing feature matching to determine an authentication result.The application can flexibly select biological features and authentication strictness according to the payment amount, and improves payment security and user experience.

Description

Technical Field

[0001] This invention relates to the field of payment security technology, and in particular to a multi-dimensional biometric identification and authentication method and system for digital payments. Background Technology

[0002] With the rapid development of information technology and e-commerce, digital payment has become an indispensable part of people's daily lives. Traditional payment authentication methods mainly rely on passwords, PIN codes, or physical bank cards, but these methods have security vulnerabilities and are easily stolen or forged. In recent years, biometric identification technology, due to its uniqueness and non-replicability, has been widely used in the field of identity authentication, playing an increasingly important role, especially in digital payment security.

[0003] Biometric identification technologies mainly include fingerprint recognition, facial recognition, iris recognition, and voiceprint recognition. These technologies collect users' biometric data, extract feature vectors, and perform matching and verification to confirm user identity. In digital payment scenarios, biometric identification technology can effectively improve payment security, simplify user operation processes, and enhance user experience.

[0004] However, existing biometric identification technologies in digital payments still have the following defects and shortcomings:

[0005] Most existing technologies rely on a single biometric for identity authentication, such as fingerprint or facial recognition. This single-dimensional authentication method has low security and is easily cracked or deceived by specific technical means, failing to meet the differentiated security requirements of different payment amounts.

[0006] Existing biometric authentication typically uses a fixed similarity threshold for identity verification, without considering the security differences in the user's environment. Under different environmental conditions (such as lighting, noise, network security status, etc.), fixed threshold authentication strategies struggle to balance security and convenience, easily leading to false rejections or false acceptances. Summary of the Invention

[0007] This invention provides a multi-dimensional biometric identification and authentication method and system for digital payments, which can solve the problems in the prior art.

[0008] A first aspect of this invention provides a multi-dimensional biometric identification and authentication method for digital payments, comprising:

[0009] Collect the user's first biometric data, and based on the first biometric data, extract the corresponding feature vector through the biometric processing unit to generate a multidimensional feature set;

[0010] The corresponding number of biometric features are selected from the multidimensional feature set according to the distribution of payment amount within a preset amount range for authentication, thus obtaining the feature set to be authenticated;

[0011] The user's second biometric data is obtained based on the feature set to be authenticated, and the corresponding authentication feature vector is extracted based on the second biometric data.

[0012] Obtain the user's identity authentication environment parameters, calculate the environment credibility based on the identity authentication environment parameters, and dynamically adjust the similarity threshold of feature matching based on the environment credibility to obtain the target similarity threshold;

[0013] The authentication feature vector is matched with the pre-stored registration feature vector, and the identity authentication is determined based on the similarity of the feature matching and the target similarity threshold.

[0014] Once identity authentication is confirmed, a success message is sent to the payment system, triggering the execution of the payment process.

[0015] Collect the user's first biometric data, and based on the first biometric data, extract the corresponding feature vector through the biometric processing unit to generate a multi-dimensional feature set, including:

[0016] Image preprocessing is performed on the first biometric data to obtain preprocessed biometric data;

[0017] A multi-dimensional image quality assessment is performed on the preprocessed biometric data, and a comprehensive image quality score is calculated based on the assessment results.

[0018] Based on the comprehensive image quality score, the preprocessed biometric data is subjected to block-based adaptive enhancement processing. Specifically, image blocks with a comprehensive image quality score less than a preset score threshold are enhanced with image enhancement parameters of greater than a preset strength, while image blocks with a comprehensive image quality score greater than the preset score threshold are enhanced with image enhancement parameters of less than a preset strength, thereby obtaining enhanced biometric data.

[0019] Based on the comprehensive image quality score, feature completion and feature enhancement are performed on the enhanced biometric data to obtain optimized biometric data. Key feature points are extracted from the optimized biometric data to obtain a feature point set.

[0020] Based on the feature point set, various biological feature vectors are constructed respectively, and feature fusion processing is performed on the various biological feature vectors to generate a multi-dimensional feature set.

[0021] From the multidimensional feature set, a corresponding number of biometric features are selected for authentication based on the distribution of payment amounts within a preset amount range, resulting in a feature set to be authenticated, including:

[0022] Construct a payment amount range distribution table, which records the correspondence between payment amount ranges and the number of biometric authentications;

[0023] Obtain the user's historical transaction data and historical biometric authentication records, calculate the user's payment habit parameters based on the historical transaction data, and obtain a payment risk assessment value based on the payment habit parameters;

[0024] The amount ranges in the payment amount range distribution table are adjusted based on the payment risk assessment value to obtain the adjusted amount ranges. The higher the payment risk assessment value, the smaller the adjusted amount range.

[0025] The credibility score of the biometrics is calculated based on the historical biometric authentication records and the payment risk assessment value. The biometrics in the multidimensional feature set are dynamically weighted based on the credibility score to obtain a weighted multidimensional feature set.

[0026] Determine the position of the current payment amount within the adjusted amount range, and obtain the number of biometric features requiring authentication from the payment amount range distribution table based on the position of the range;

[0027] Based on the number of biometric features, the biometric feature with the largest weight is selected from the weighted multidimensional feature set to obtain the feature set to be authenticated.

[0028] Based on the historical biometric authentication records and the payment risk assessment value, a credibility score for the biometrics is calculated. Then, based on this credibility score, the biometrics in the multidimensional feature set are dynamically weighted to obtain a weighted multidimensional feature set, including:

[0029] Based on the historical biometric authentication records, a time decay coefficient for each historical biometric is calculated. The time decay coefficient decreases as the authentication time interval increases, thus obtaining a time-weighted coefficient.

[0030] A risk weighting coefficient is determined based on the payment risk assessment value, and the risk weighting coefficient increases as the payment risk assessment value increases;

[0031] The historical biometric authentication records are classified according to the authentication scenario, and the authentication accuracy of biometrics in each authentication scenario is calculated.

[0032] Based on the authentication accuracy, a scene similarity matrix is ​​constructed. The current payment scene is matched with the scene similarity matrix. The biometric authentication success probability of the authentication scene with the highest matching degree is extracted. The scene weighting coefficient is calculated based on the biometric authentication success probability.

[0033] The time-weighted coefficient, the risk-weighted coefficient, and the scenario-weighted coefficient are combined to calculate the credibility score of the biometric feature.

[0034] The biometric features in the multidimensional feature set are dynamically weighted based on the credibility score to obtain the weighted multidimensional feature set.

[0035] Obtain the user's identity authentication environment parameters, calculate the environment credibility based on the identity authentication environment parameters, and dynamically adjust the similarity threshold for feature matching based on the environment credibility to obtain the target similarity threshold, including:

[0036] Obtain the user's identity authentication environment parameters and historical authentication environment parameters, perform stratified processing on the historical authentication environment parameters according to the authentication success rate, and construct an authentication environment baseline value for each layer of the historical authentication environment parameters.

[0037] The identity authentication environment parameters are matched and calculated with the authentication environment benchmark value in multiple dimensions. The validity of the identity authentication environment parameters is verified based on the matching results to obtain the environment parameter matching degree.

[0038] Obtain the user's historical authentication behavior data, extract the user's behavior trajectory during the authentication process from the historical authentication behavior data, construct an authentication behavior feature curve based on the user's behavior trajectory, and extract the user's operation behavior feature value from the authentication behavior feature curve;

[0039] The user operation behavior feature values ​​are divided into multiple feature segments according to the time window, and the fluctuation trend of the feature segments is calculated to obtain the user behavior stability index. The environmental credibility is determined based on the user behavior stability index.

[0040] A threshold adjustment coefficient is generated based on the environmental credibility, and the baseline similarity threshold for feature matching is dynamically adjusted according to the threshold adjustment coefficient to obtain the target similarity threshold.

[0041] The user operation behavior feature values ​​are divided into multiple feature segments according to a time window. The fluctuation trend of the feature segments is calculated to obtain a user behavior stability index. The environmental credibility is determined based on the user behavior stability index, including:

[0042] A benchmark library of authentication behavior features is constructed. Authentication behavior trajectories are extracted from historical successful authentication records. Data cleaning is performed on the authentication behavior trajectories to generate benchmark values ​​of behavior features.

[0043] The user operation behavior feature values ​​are segmented and normalized according to the time window size to generate multiple normalized feature fragments;

[0044] Based on the behavioral feature benchmark value, a reference weight is calculated for each normalized feature segment, and the normalized feature segments are reconstructed according to the reference weight to obtain a reconstructed feature sequence.

[0045] The reconstructed feature sequences are arranged in chronological order to construct a feature fluctuation matrix, and the difference between adjacent feature segments is calculated to obtain the fluctuation subsequence.

[0046] The fluctuation subsequence is recursively decomposed to extract fluctuation trend features and generate a fluctuation trend function;

[0047] The user behavior stability index is calculated based on the fluctuation trend function, and the environmental credibility is determined according to the degree of deviation between the user behavior stability index and the behavioral feature benchmark value.

[0048] The authentication feature vector is matched with a pre-stored registration feature vector. Based on the similarity of the feature matches and the target similarity threshold, the authentication is determined to be successful, including:

[0049] The authentication feature vector is decomposed to extract key feature points and generate an authentication feature point set.

[0050] The registration feature vector is subjected to feature decomposition to extract key feature points and generate a registration feature point set.

[0051] The authentication feature point set and the registration feature point set are matched point-to-point to obtain feature point matching pairs. The local similarity of each feature point in the feature point matching pair is calculated, and the weight coefficient of each feature point is determined based on the local similarity.

[0052] The feature point matching pairs are weighted according to the weight coefficients to generate a global similarity for feature matching.

[0053] The global similarity is compared with the target similarity threshold. When the global similarity is greater than the target similarity threshold, the identity authentication is determined to be successful. When the global similarity is less than or equal to the target similarity threshold, the identity authentication is determined to be unsuccessful.

[0054] A second aspect of this invention provides a multi-dimensional biometric identification and authentication system for digital payments, comprising:

[0055] The first unit is used to collect the user's first biometric data, and based on the first biometric data, the corresponding feature vector is extracted by the biometric processing unit to generate a multidimensional feature set.

[0056] The second unit is used to select a corresponding number of biometric features from the multidimensional feature set according to the distribution of payment amount within a preset amount range for authentication, thereby obtaining a feature set to be authenticated;

[0057] The third unit is used to obtain the user's second biometric data based on the feature set to be authenticated, and to extract the corresponding authentication feature vector based on the second biometric data.

[0058] The fourth unit is used to obtain the user's identity authentication environment parameters, calculate the environment credibility based on the identity authentication environment parameters, and dynamically adjust the similarity threshold of feature matching based on the environment credibility to obtain the target similarity threshold.

[0059] The fifth unit is used to perform feature matching between the authentication feature vector and the pre-stored registration feature vector, and determine whether the identity authentication is successful based on the similarity of the feature matching and the target similarity threshold.

[0060] The sixth unit is used to send authentication success information to the payment system when identity authentication is confirmed to be successful, thereby triggering the execution of the payment process.

[0061] A third aspect of the present invention,

[0062] An electronic device is provided, comprising:

[0063] processor;

[0064] Memory used to store processor-executable instructions;

[0065] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0066] Fourth aspect of the embodiments of the present invention,

[0067] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0068] The beneficial effects of this application are as follows:

[0069] This invention employs a multidimensional biometric recognition and authentication method, which can automatically select an appropriate number of biometric features for authentication based on the payment amount. This ensures security while avoiding unnecessary authentication costs, thereby improving user experience and payment efficiency.

[0070] This invention calculates the credibility of the environment by acquiring user identity authentication environment parameters and dynamically adjusts the similarity threshold of feature matching, so that the authentication mechanism can be flexibly adjusted according to different environmental conditions, which not only ensures the security in high-risk environments, but also improves the convenience of authentication in normal environments.

[0071] This invention links multidimensional biometric recognition with payment amount ranges to form a tiered security authentication mechanism. This mechanism can meet the high security requirements of large payments while ensuring the convenience of small payments, effectively balancing the contradiction between payment security and user experience, and adapting to the differentiated needs of various payment scenarios. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating the multi-dimensional biometric recognition and authentication method for digital payments according to an embodiment of the present invention.

[0073] Figure 2 This is a schematic diagram of the process for determining whether identity authentication passes based on the similarity between feature matching and the target similarity threshold. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0076] Figure 1 This is a flowchart illustrating the multi-dimensional biometric recognition and authentication method for digital payments according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0077] Collect the user's first biometric data, and based on the first biometric data, extract the corresponding feature vector through the biometric processing unit to generate a multidimensional feature set;

[0078] The corresponding number of biometric features are selected from the multidimensional feature set according to the distribution of payment amount within a preset amount range for authentication, thus obtaining the feature set to be authenticated;

[0079] The user's second biometric data is obtained based on the feature set to be authenticated, and the corresponding authentication feature vector is extracted based on the second biometric data.

[0080] Obtain the user's identity authentication environment parameters, calculate the environment credibility based on the identity authentication environment parameters, and dynamically adjust the similarity threshold of feature matching based on the environment credibility to obtain the target similarity threshold;

[0081] The authentication feature vector is matched with the pre-stored registration feature vector, and the identity authentication is determined based on the similarity of the feature matching and the target similarity threshold.

[0082] Once identity authentication is confirmed, a success message is sent to the payment system, triggering the execution of the payment process.

[0083] In one optional implementation, first biometric data of the user is collected, and based on the first biometric data, corresponding feature vectors are extracted by a biometric processing unit to generate a multidimensional feature set, including:

[0084] Image preprocessing is performed on the first biometric data to obtain preprocessed biometric data;

[0085] A multi-dimensional image quality assessment is performed on the preprocessed biometric data, and a comprehensive image quality score is calculated based on the assessment results.

[0086] Based on the comprehensive image quality score, the preprocessed biometric data is subjected to block-based adaptive enhancement processing. Specifically, image blocks with a comprehensive image quality score less than a preset score threshold are enhanced with image enhancement parameters of greater than a preset strength, while image blocks with a comprehensive image quality score greater than the preset score threshold are enhanced with image enhancement parameters of less than a preset strength, thereby obtaining enhanced biometric data.

[0087] Based on the comprehensive image quality score, feature completion and feature enhancement are performed on the enhanced biometric data to obtain optimized biometric data. Key feature points are extracted from the optimized biometric data to obtain a feature point set.

[0088] Based on the feature point set, various biological feature vectors are constructed respectively, and feature fusion processing is performed on the various biological feature vectors to generate a multi-dimensional feature set.

[0089] During the biometric data acquisition phase, the system obtains the user's initial biometric data through biometric acquisition devices, such as fingerprint images captured by optical sensors, facial images captured by cameras, or iris images captured by iris scanners. The raw biometric data obtained may contain issues such as noise, blurriness, and insufficient contrast, requiring subsequent processing to improve the accuracy of feature extraction.

[0090] The acquired first biometric data undergoes image preprocessing, including grayscale conversion, denoising, and standardization. Grayscale conversion transforms the color image into a grayscale image, simplifying subsequent processing. Denoising uses a Gaussian filter with a kernel size of 5×5 pixels and a variance of 1.2, effectively eliminating high-frequency noise in the image. Standardization adjusts the image pixel values ​​to the range of 0-255 and performs histogram equalization to improve image contrast. For example, for a raw fingerprint image, grayscale conversion converts the RGB three-channel image into a single channel, then a denoising algorithm is applied to remove random noise generated during acquisition, and finally standardization makes the fingerprint texture clearer and more discernible.

[0091] Multi-dimensional image quality assessment is performed on preprocessed biometric data, including sharpness assessment, contrast assessment, signal-to-noise ratio (SNR) assessment, and integrity assessment. Sharpness assessment measures edge sharpness by calculating the Laplacian operator response value of the image; contrast assessment quantifies the image by calculating the variance of the gray-level distribution; SNR assessment uses the peak signal-to-noise ratio (PSNR) metric; and integrity assessment is determined by the proportion of detected effective feature regions. The assessment results of each dimension are combined according to weights to obtain a comprehensive image quality score, with weights of: sharpness 0.35, contrast 0.25, SNR 0.2, and integrity 0.2. For example, a fingerprint image with a sharpness score of 75, a contrast score of 80, an SNR score of 70, and an integrity score of 85 has a comprehensive quality score of 77.

[0092] Based on the overall image quality score, the preprocessed biometric data undergoes adaptive block enhancement. The image is uniformly divided into 8×8 small blocks, and each block is individually quality-assessed. A preset score threshold of 70 is set. For blocks with a quality score below 70, adaptive histogram equalization with an enhancement intensity parameter of 1.5 is applied; for blocks with a quality score above 70, a mild enhancement with an enhancement intensity parameter of 0.8 is applied. The enhancement intensity parameter is controlled by limiting the contrast threshold in the adaptive histogram equalization algorithm; a higher enhancement intensity parameter results in a more significant contrast improvement. For example, for the edge region of a fingerprint image with a quality score of 55, the system applies an enhancement intensity parameter of 1.5, significantly improving the contrast of that region; while for the relatively high-quality central region with a score of 85, a mild enhancement with an enhancement intensity parameter of 0.8 is used to avoid information distortion caused by over-enhancement.

[0093] The enhanced biometric data is augmented and its features are completed based on the overall image quality score. For regions with a quality score below 60, the system uses a feature inpainting algorithm based on local similarity to learn texture features from adjacent high-quality regions and fill in the low-quality regions. For regions with a quality score between 60 and 80, a directional filter is used to enhance texture features. The filter direction is consistent with the local texture direction, with a length of 7 pixels and a width of 3 pixels. For regions with a quality score above 80, the original features are preserved. This adaptive processing strategy yields optimized biometric data. Key feature points are extracted from the optimized biometric data, such as minutiae (terminal points and bifurcation points) in fingerprint images and facial key points in face images, resulting in a feature point set. For example, 128 minutiae are extracted from an optimized fingerprint image, each containing attributes such as location coordinates, orientation angle, and minutiae type.

[0094] Multiple biometric feature vectors are constructed based on the extracted feature point set. For fingerprint features, a local feature vector based on minutiae (dimensional 256) and a global feature vector based on overall texture (dimensional 512) are constructed simultaneously. For facial features, a structural feature vector based on geometric relationships (dimensional 128) and an appearance feature vector based on texture (dimensional 256) are constructed. These different types of feature vectors are then fused using a combination of feature concatenation and feature transformation to generate a feature set containing multidimensional information. During feature fusion, each feature vector is first normalized to ensure its component values ​​are evenly distributed within the 0-1 range. Then, different weights are assigned based on each vector's discriminative ability in identity recognition, for example, a weight of 0.6 for local features and 0.4 for global features. Finally, a weighted combination is used to generate the final multidimensional feature set. The generated multidimensional feature set contains both local detail information and global structural information of biometric features, improving the robustness and accuracy of biometric recognition.

[0095] Through the above implementation methods, the present invention achieves high-quality processing and feature extraction of biometric data. The generated multidimensional feature set can be effectively applied to scenarios such as biometric recognition and identity verification, thereby improving system performance and user experience.

[0096] In one optional implementation, a corresponding number of biometric features are selected from the multidimensional feature set for authentication based on the distribution of payment amounts within a preset amount range, resulting in a feature set to be authenticated, including:

[0097] Construct a payment amount range distribution table, which records the correspondence between payment amount ranges and the number of biometric authentications;

[0098] Obtain the user's historical transaction data and historical biometric authentication records, calculate the user's payment habit parameters based on the historical transaction data, and obtain a payment risk assessment value based on the payment habit parameters;

[0099] The amount ranges in the payment amount range distribution table are adjusted based on the payment risk assessment value to obtain the adjusted amount ranges. The higher the payment risk assessment value, the smaller the adjusted amount range.

[0100] The credibility score of the biometrics is calculated based on the historical biometric authentication records and the payment risk assessment value. The biometrics in the multidimensional feature set are dynamically weighted based on the credibility score to obtain a weighted multidimensional feature set.

[0101] Determine the position of the current payment amount within the adjusted amount range, and obtain the number of biometric features requiring authentication from the payment amount range distribution table based on the position of the range;

[0102] Based on the number of biometric features, the biometric feature with the largest weight is selected from the weighted multidimensional feature set to obtain the feature set to be authenticated.

[0103] This embodiment provides a method for selecting a corresponding number of biometric features from a multidimensional feature set based on the distribution of payment amounts within a preset amount range for authentication. This method can dynamically adjust the number and type of biometric features required for authentication based on the user's payment habits and risk level, improving payment security while ensuring a good user experience.

[0104] In this implementation, the system pre-constructs a payment amount range distribution table, recording the correspondence between different amount ranges and the required number of biometric authentications. For example, 0-100 yuan requires 1 biometric authentication, 100-1000 yuan requires 2 biometric authentications, 1000-10000 yuan requires 3 biometric authentications, and over 10000 yuan requires 4 biometric authentications. This design follows the principle of commensurate risk and authentication strength; the larger the amount, the stricter the authentication requirements.

[0105] The system acquires users' historical transaction data and historical biometric authentication records. Historical transaction data includes information such as transaction time, transaction amount, transaction frequency, and transaction location. For example, user A made 50 payments in the past 3 months, with an average payment amount of 200 yuan and a maximum payment amount of 2000 yuan. 90% of the payments were below 500 yuan, and most transactions occurred in a few fixed cities. Historical biometric authentication records include the frequency and success rate of various biometric methods, such as fingerprint recognition (80% frequency, 95% success rate); facial recognition (15% frequency, 90% success rate); and voiceprint recognition (5% frequency, 85% success rate).

[0106] Based on historical transaction data, the system calculates users' payment habit parameters. These parameters include average payment amount, standard deviation of payment amount, payment frequency, and frequently used payment locations. For user A, the system calculates an average payment amount of 200 yuan, a standard deviation of 150 yuan, an average of one payment every two days, and 95% of payments occurring at three fixed locations.

[0107] The system calculates a payment risk assessment value based on payment habit parameters. The calculation of this value considers factors such as the deviation between the current payment amount and the user's average payment amount, whether the payment location is a frequently used location, and whether the payment frequency is abnormal. For example, if user A makes a payment of 2000 yuan, which is significantly higher than their average payment amount and is made in an uncommon location, the system assigns a high risk value of 0.8 (risk value ranges from 0 to 1, with higher values ​​indicating higher risk).

[0108] Based on the payment risk assessment value, the system adjusts the payment amount range distribution table. The higher the risk assessment value, the smaller the adjusted amount range, meaning more biometric authentication is required for the same amount. For user A, due to the high risk value (0.8), the original amount ranges are adjusted as follows: 0-50 yuan requires 1 authentication, 50-500 yuan requires 2 authentications, 500-5000 yuan requires 3 authentications, and over 5000 yuan requires 4 authentications.

[0109] The system calculates a credibility score for each biometric feature based on historical biometric authentication records and payment risk assessment values. The credibility score considers the historical success rate, frequency of use, and security level of the biometric feature. For user A, the system calculates a credibility score of 0.85 for fingerprint recognition, 0.75 for facial recognition, 0.65 for voiceprint recognition, and 0.95 for iris recognition.

[0110] Based on the confidence score, the system dynamically weights the biometric features in the multidimensional feature set to obtain a weighted multidimensional feature set. The weighting process considers the confidence score and the current payment environment, such as lighting conditions and noise levels. For example, under the current environment, fingerprint recognition is weighted at 0.9, facial recognition at 0.65 due to poor lighting, voiceprint recognition at 0.6 due to environmental noise, and iris recognition remains at 0.95.

[0111] The system determines the position of the current payment amount within the adjusted amount range and obtains the number of biometric features required for authentication. User A's payment of 2,000 yuan falls within the adjusted "500-5,000 yuan" range. According to the payment amount range distribution table, 3 biometric features are required for authentication.

[0112] The system selects the biometric feature with the highest weight from the weighted multidimensional feature set based on the required number of biometric features, thus obtaining the feature set to be authenticated. For user A, the system selects the three features with the highest weights: iris recognition (0.95), fingerprint recognition (0.9), and facial recognition (0.65) as the feature set to be authenticated.

[0113] During the actual payment process, the system will display the required biometric authentication steps to the user. After the user completes iris recognition, fingerprint recognition, and facial recognition in sequence, the system will verify the authentication results. Only if all authentications are successful will the high-risk payment transaction be completed. If the user fails in any authentication step, the system will provide a retry opportunity or suggest using other verification methods.

[0114] This method allows the system to dynamically adjust authentication requirements based on users' actual payment behavior and risk profile, providing a smooth user experience while ensuring security. Low-risk transactions require only minimal biometric authentication, while high-risk transactions demand a more diverse combination of biometric authentication to effectively prevent fraud.

[0115] In one optional implementation, a biometric credibility score is calculated based on the historical biometric authentication records and the payment risk assessment value. The biometric features in the multidimensional feature set are then dynamically weighted based on the credibility score to obtain a weighted multidimensional feature set, including:

[0116] Based on the historical biometric authentication records, a time decay coefficient for each historical biometric is calculated. The time decay coefficient decreases as the authentication time interval increases, thus obtaining a time-weighted coefficient.

[0117] A risk weighting coefficient is determined based on the payment risk assessment value, and the risk weighting coefficient increases as the payment risk assessment value increases;

[0118] The historical biometric authentication records are classified according to the authentication scenario, and the authentication accuracy of biometrics in each authentication scenario is calculated.

[0119] Based on the authentication accuracy, a scene similarity matrix is ​​constructed. The current payment scene is matched with the scene similarity matrix. The biometric authentication success probability of the authentication scene with the highest matching degree is extracted. The scene weighting coefficient is calculated based on the biometric authentication success probability.

[0120] The time-weighted coefficient, the risk-weighted coefficient, and the scenario-weighted coefficient are combined to calculate the credibility score of the biometric feature.

[0121] The biometric features in the multidimensional feature set are dynamically weighted based on the credibility score to obtain the weighted multidimensional feature set.

[0122] In this embodiment, a payment risk control method based on biometric authentication is provided. This method improves payment security by calculating the credibility score of biometric features and dynamically weighting a multi-dimensional feature set.

[0123] When implementing this method, the system first obtains the user's historical biometric authentication records and the risk assessment value for the current payment scenario. The historical biometric authentication records include information such as the time, scenario, and success rate of the user's past authentication using biometrics such as fingerprints, faces, and irises. The payment risk assessment value is a risk indicator based on a comprehensive evaluation of factors such as the current payment amount, frequency, and location.

[0124] When calculating the time decay coefficient for each historical biometric authentication record, the system uses a time decay function to handle the authentication time interval. For example, for fingerprint authentication records, if the most recent successful authentication was 1 day ago, the time decay coefficient can be set to 0.95; if it was 7 days ago, the decay coefficient can be set to 0.85; and if it was 30 days ago, the decay coefficient can be set to 0.60. The longer the time interval, the smaller the decay coefficient, indicating that the reference value of the historical record decreases. The system can calculate the time decay coefficient separately for each biometric feature (such as fingerprint, face, and iris) to obtain the corresponding time-weighted coefficient.

[0125] When determining the risk weighting coefficient based on the payment risk assessment value, the system quantifies the payment risk into a value between 0 and 1. For example, when the payment risk assessment value is 0.2 (low risk), the risk weighting coefficient can be set to 0.6; when the risk assessment value is 0.5 (medium risk), the risk weighting coefficient can be set to 0.8; and when the risk assessment value is 0.9 (high risk), the risk weighting coefficient can be set to 0.95. The higher the risk assessment value, the larger the risk weighting coefficient, indicating that the system has stricter requirements for biometric authentication in high-risk transactions.

[0126] When categorizing historical biometric authentication records according to authentication scenarios, the system can divide scenarios into multiple categories, such as ordinary online shopping, high-value online shopping, offline physical store payments, and cross-border payments. The system then calculates the authentication accuracy rate for various biometrics in each scenario. For example, in ordinary online shopping scenarios, the accuracy rate for fingerprint authentication is 0.98, and the accuracy rate for facial recognition is 0.97; in high-value online shopping scenarios, the accuracy rate for fingerprint authentication is 0.95, and the accuracy rate for facial recognition is 0.96.

[0127] When constructing a scene similarity matrix based on authentication accuracy, the system calculates the similarity between different scenes. For example, the similarity between ordinary online shopping and large-value online shopping is 0.85, and the similarity between ordinary online shopping and offline physical store payment is 0.70. When a user performs an operation in the current payment scenario (such as large-value online shopping), the system matches the current scenario with the scene similarity matrix to find the scenario with the highest similarity. If the large-value online shopping scenario exists in the historical records, the system directly extracts the biometric authentication success probability under that scenario; otherwise, it extracts the biometric authentication success probability from the scenario with the highest similarity (such as ordinary online shopping, similarity 0.85) and multiplies it by the scene similarity as an adjustment. Based on the extracted biometric authentication success probability, a scene weighting coefficient is calculated. For example, if the success probability of fingerprint authentication in the matching scenario is 0.95, the scene weighting coefficient can be set to 0.92.

[0128] When combining time-weighted, risk-weighted, and scenario-weighted coefficients, the system can use a weighted average method. Assuming a user's fingerprint authentication time-weighted coefficient is 0.85, risk-weighted coefficient is 0.80, and scenario-weighted coefficient is 0.92, the system can calculate a weighted average of these three coefficients in a 4:3:3 ratio, resulting in a fingerprint biometric confidence score of 0.857. The same method can be used to calculate the confidence scores for other biometric features such as face and iris scans.

[0129] When dynamically weighting biometric features in a multi-dimensional feature set based on their credibility scores, the system adjusts the weight of each biometric feature in the authentication process according to its credibility score. For example, if the credibility score for fingerprints is 0.857, for faces it's 0.823, and for iris scans it's 0.912, then when comprehensively evaluating a user's identity, the weight of the fingerprint feature is 0.857 / (0.857+0.823+0.912)=0.331, the weight of the face feature is 0.318, and the weight of the iris scan is 0.351. Through this dynamic weighting method, the system places greater trust in biometric features with high credibility scores, improving the overall authentication accuracy.

[0130] In practical applications, when a user makes an online transfer of 10,000 yuan, the system assesses the payment risk as 0.75 (medium-high risk). The user's most recent fingerprint authentication was 2 days ago (time-weighted coefficient 0.9), and facial authentication was 10 days ago (time-weighted coefficient 0.8). The system calculates a credibility score of 0.88 for the fingerprint and 0.82 for the facial recognition. Based on this, the system assigns a weight of 0.52 to the fingerprint feature and a weight of 0.48 to the facial feature in the multidimensional feature set, thereby generating a weighted multidimensional feature set for subsequent identity authentication and risk control.

[0131] Through the aforementioned technical means, the system can dynamically adjust the weight of biometric features based on historical authentication records, payment risks, and scenario similarity, thereby improving authentication accuracy and payment security.

[0132] In one optional implementation, the user's identity authentication environment parameters are obtained, an environment credibility is calculated based on the identity authentication environment parameters, and a target similarity threshold is obtained by dynamically adjusting the similarity threshold of feature matching based on the environment credibility, including:

[0133] Obtain the user's identity authentication environment parameters and historical authentication environment parameters, perform stratified processing on the historical authentication environment parameters according to the authentication success rate, and construct an authentication environment baseline value for each layer of the historical authentication environment parameters.

[0134] The identity authentication environment parameters are matched and calculated with the authentication environment benchmark value in multiple dimensions. The validity of the identity authentication environment parameters is verified based on the matching results to obtain the environment parameter matching degree.

[0135] Obtain the user's historical authentication behavior data, extract the user's behavior trajectory during the authentication process from the historical authentication behavior data, construct an authentication behavior feature curve based on the user's behavior trajectory, and extract the user's operation behavior feature value from the authentication behavior feature curve;

[0136] The user operation behavior feature values ​​are divided into multiple feature segments according to the time window, and the fluctuation trend of the feature segments is calculated to obtain the user behavior stability index. The environmental credibility is determined based on the user behavior stability index.

[0137] A threshold adjustment coefficient is generated based on the environmental credibility, and the baseline similarity threshold for feature matching is dynamically adjusted according to the threshold adjustment coefficient to obtain the target similarity threshold.

[0138] Identity authentication environment parameters include device information, network information, geographical location information, and time information. Device information may include device model, operating system version, MAC address, and unique device identifier; network information may include IP address, network type (e.g., 4G, 5G, WiFi), and network provider; geographical location information may include GPS coordinates and city / region; and time information may include the authentication time period and whether it is operating hours.

[0139] Historical authentication environment parameters refer to records of environmental parameters used during a user's past successful authentications. These records are stratified according to authentication success rate. For example, historical authentication environment parameters are divided into three tiers based on authentication success rate, from highest to lowest: High Trust Layer (success rate ≥ 95%), Medium Trust Layer (85% ≤ success rate < 95%), and Low Trust Layer (success rate < 85%). For each tier, corresponding authentication environment baseline values ​​are constructed. For instance, the baseline values ​​for the High Trust Layer include: frequently used home Wi-Fi networks, working hours (9:00-18:00), and frequently used office areas; the Medium Trust Layer includes: frequently used mobile networks and non-working hours; and the Low Trust Layer includes: infrequently used areas and infrequently used networks.

[0140] The current identity authentication environment parameters are matched against the aforementioned authentication environment baseline values ​​using a multi-dimensional matching calculation. Different weights are assigned to parameters in different dimensions during the matching calculation; for example, device information has a weight of 0.4, network information 0.3, geographic location information 0.2, and time information 0.1. The calculated matching degree is the weighted sum of the matching results for each dimension. Assuming the current user is using a frequently used device (matching degree 0.9), connected to a frequently used WiFi network (matching degree 0.8), located in a frequently used office area (matching degree 0.95), and within a working time period (matching degree 0.9), then the environmental parameter matching degree is 0.9 × 0.4 + 0.8 × 0.3 + 0.95 × 0.2 + 0.9 × 0.1 = 0.875.

[0141] Historical authentication behavior data includes user actions during the authentication process, such as changes in pressure applied to the screen, finger swipe speed, and time intervals between password inputs. User behavior trajectories are extracted from this data to construct authentication behavior feature curves. For example, for fingerprint authentication, pressure variation curves can be extracted; for facial authentication, facial angle variation curves can be extracted; and for password input, input rhythm variation curves can be extracted.

[0142] User behavior feature values ​​are extracted from the authentication behavior feature curve, including but not limited to operation speed, operation pressure, and operation angle. These feature values ​​are divided into multiple feature segments according to a time window, which can be set to 200 milliseconds. For example, the facial angle change curve during a face authentication process can be divided into multiple feature segments with each 200 millisecond window as a unit. The fluctuation trend of each feature segment is calculated to obtain a user behavior stability index. The fluctuation trend can be represented by calculating the standard deviation of the feature values ​​within each time window. A smaller standard deviation indicates stable user behavior; a larger standard deviation indicates greater fluctuation in user behavior, which may indicate anomalies.

[0143] Environmental credibility is determined by combining user behavior stability metrics with environmental parameter matching scores. Environmental credibility is calculated using a weighted average, with environmental parameter matching scores having a weight of 0.7 and user behavior stability metrics having a weight of 0.3. For example, assuming an environmental parameter matching score of 0.875 and a user behavior stability metric of 0.92 (small standard deviation, indicating stable behavior), the environmental credibility would be 0.875 × 0.7 + 0.92 × 0.3 = 0.889.

[0144] Based on the calculated environmental credibility, a threshold adjustment coefficient is generated. The threshold adjustment coefficient is inversely proportional to the environmental credibility; the higher the environmental credibility, the smaller the threshold adjustment coefficient, meaning the similarity threshold requirement is relaxed; conversely, the lower the environmental credibility, the larger the threshold adjustment coefficient, meaning the similarity threshold requirement is increased. The formula for calculating the threshold adjustment coefficient can be set as: Threshold adjustment coefficient = 1 - Environmental credibility. In the example above, the threshold adjustment coefficient = 1 - 0.889 = 0.111.

[0145] The baseline similarity threshold for feature matching is dynamically adjusted based on a threshold adjustment coefficient to obtain the target similarity threshold. The baseline similarity threshold is typically set to 0.75, meaning that a match is considered successful when the feature matching similarity reaches 0.75 or higher. The formula for calculating the target similarity threshold is: Target similarity threshold = Baseline similarity threshold + Threshold adjustment coefficient × (1 - Baseline similarity threshold). In the example above, the target similarity threshold = 0.75 + 0.111 × (1 - 0.75) = 0.778. This means that in the current environment, the feature matching similarity must reach 0.778 or higher to be considered a successful match.

[0146] Through the above steps, this invention implements a method for dynamically adjusting the feature matching similarity threshold based on user authentication environment parameters, thereby improving the security and accuracy of identity authentication. This method is applicable to various biometric recognition scenarios, such as face recognition, fingerprint recognition, and voiceprint recognition, and can effectively prevent spoofing attacks and deception.

[0147] In one optional implementation, the user operation behavior feature values ​​are divided into multiple feature segments according to a time window, the fluctuation trend of the feature segments is calculated to obtain a user behavior stability index, and the environmental credibility is determined based on the user behavior stability index, including:

[0148] A benchmark library of authentication behavior features is constructed. Authentication behavior trajectories are extracted from historical successful authentication records. Data cleaning is performed on the authentication behavior trajectories to generate benchmark values ​​of behavior features.

[0149] The user operation behavior feature values ​​are segmented and normalized according to the time window size to generate multiple normalized feature fragments;

[0150] Based on the behavioral feature benchmark value, a reference weight is calculated for each normalized feature segment, and the normalized feature segments are reconstructed according to the reference weight to obtain a reconstructed feature sequence.

[0151] The reconstructed feature sequences are arranged in chronological order to construct a feature fluctuation matrix, and the difference between adjacent feature segments is calculated to obtain the fluctuation subsequence.

[0152] The fluctuation subsequence is recursively decomposed to extract fluctuation trend features and generate a fluctuation trend function;

[0153] The user behavior stability index is calculated based on the fluctuation trend function, and the environmental credibility is determined according to the degree of deviation between the user behavior stability index and the behavioral feature benchmark value.

[0154] In one specific embodiment of the present invention, by dividing the user operation behavior feature values ​​into time windows and analyzing the fluctuation trend, a user behavior stability index is calculated, and the environmental credibility is determined accordingly.

[0155] The system extracts authentication behavior patterns from historical successful authentication records, such as user behavior characteristics of successful logins within the past 30 days, including keystroke intervals, mouse movement speed, and touch screen pressure. These authentication behavior patterns undergo data cleaning to remove outliers; for example, data points with keystroke intervals exceeding 10 seconds may indicate that the user has temporarily left the system and should be removed. After data cleaning, the mean and standard deviation of each feature are calculated to generate baseline values ​​for the behavioral characteristics. For example, the mean keystroke interval for a normal user is 0.8 seconds, with a standard deviation of 0.15 seconds; the mean mouse movement speed is 20 pixels / second, with a standard deviation of 5 pixels / second.

[0156] User behavior feature values ​​are segmented and normalized according to the time window size. Specifically, an appropriate time window size is selected, such as 60 seconds, and continuously collected user behavior data is divided into multiple feature segments according to this window size. For each feature segment, the feature data is normalized so that its mean is 0 and its standard deviation is 1. Normalization can be achieved by subtracting the feature mean and dividing by the feature standard deviation. For example, the keystroke interval sequence [0.7, 0.9, 0.8, 0.6, 1.0] within a time window may become [-0.67, 0.67, 0, -1.33, 1.33] after normalization.

[0157] The reference weight of each normalized feature segment is calculated based on the behavioral feature baseline value. Specifically, the similarity between the normalized feature segment and the behavioral feature baseline value is calculated; the higher the similarity, the greater the reference weight. Similarity can be calculated using the cosine similarity of the feature vectors or the reciprocal of the Euclidean distance. For example, if the Euclidean distance between a normalized feature segment and the baseline value is 2.5, its reference weight can be set to 1 / 2.5 = 0.4. The normalized feature segments are reconstructed based on the calculated reference weights, retaining feature components with higher weights and suppressing feature components with lower weights to obtain the reconstructed feature sequence. The reconstruction process can be achieved by multiplying the weight by the original feature value. For example, if the original feature value is [-0.67, 0.67, 0, -1.33, 1.33] and the reference weight is 0.4, the reconstructed feature value will be [-0.268, 0.268, 0, -0.532, 0.532].

[0158] The reconstructed feature sequences are arranged in chronological order to construct a feature fluctuation matrix. Assuming there are n time windows, and each window extracts m features, an n×m matrix is ​​constructed, where each row represents the feature vector of a time window. The difference between adjacent feature segments is calculated using metrics such as Euclidean distance, Manhattan distance, or Chebyshev distance. For example, the feature vector of time window 1 is [0.1, 0.2, 0.3], and the feature vector of time window 2 is [0.15, 0.25, 0.28], with a Euclidean distance of 0.0714 between them. The difference between all adjacent time windows is arranged sequentially to obtain the fluctuation subsequence [0.0714, 0.0856, 0.1125, ...].

[0159] Recursive decomposition can be achieved through methods such as empirical mode decomposition or wavelet transform. The volatile subsequence is decomposed into a trend term, a periodic term, and a residual term. The trend term reflects long-term trends, the periodic term reflects periodic fluctuation patterns, and the residual term reflects random noise. By analyzing these components, a volatility trend function is generated, which can describe the changes in user behavior patterns over time.

[0160] User behavior stability index is calculated based on a fluctuation trend function. The stability index can be obtained by calculating the mean absolute value of the first derivative of the fluctuation trend function; the smaller the value, the more stable the behavior. For example, if the first derivative sequence of the fluctuation trend function is [0.02, 0.03, 0.01, 0.02, 0.04], then the stability index is 0.024. The credibility of the environment is determined by the deviation between the user behavior stability index and the behavioral feature benchmark value. A threshold range can be set; when the stability index is within a certain range near the benchmark value, it is judged as a high-credibility environment; when the stability index deviates significantly from the benchmark value, it is judged as a low-credibility environment. Specifically, if the stability index of the behavioral feature benchmark value is 0.025, and the current stability index is 0.024, the difference is 0.001, which is less than the preset threshold of 0.01, then the credibility of the current environment is determined to be 95%.

[0161] Furthermore, environment trustworthiness can be combined with other security measures. For example, when the environment trustworthiness is below 80%, the system can require the user to perform two-factor authentication; when the environment trustworthiness is below 60%, the system can temporarily lock the account and notify the security administrator. In this way, the present invention can automatically determine the environment trustworthiness based on the stability of user behavior, improving system security while maintaining a good user experience.

[0162] In one optional implementation, the authentication feature vector is matched with a pre-stored registration feature vector, and the identity authentication is determined based on the similarity of the feature matches and the target similarity threshold, including:

[0163] The authentication feature vector is decomposed to extract key feature points and generate an authentication feature point set.

[0164] The registration feature vector is subjected to feature decomposition to extract key feature points and generate a registration feature point set.

[0165] The authentication feature point set and the registration feature point set are matched point-to-point to obtain feature point matching pairs. The local similarity of each feature point in the feature point matching pair is calculated, and the weight coefficient of each feature point is determined based on the local similarity.

[0166] The feature point matching pairs are weighted according to the weight coefficients to generate a global similarity for feature matching.

[0167] The global similarity is compared with the target similarity threshold. When the global similarity is greater than the target similarity threshold, the identity authentication is determined to be successful. When the global similarity is less than or equal to the target similarity threshold, the identity authentication is determined to be unsuccessful.

[0168] Figure 2 This is a flowchart illustrating the process of determining whether identity authentication passes based on the similarity of feature matching and the target similarity threshold. At the start of the authentication process, the system acquires the biometric data of the user to be authenticated, such as a fingerprint image, face image, or iris image. The acquired biometric data undergoes preprocessing, including noise removal, contrast enhancement, and feature standardization. An authentication feature vector is then extracted from the preprocessed biometric data using a feature extraction algorithm. For example, for a fingerprint image, ridge feature points can be extracted; for a face image, facial key point features can be extracted.

[0169] Feature decomposition is performed on the authentication feature vector to extract key feature points. Principal component analysis (PCA) is used to map the high-dimensional feature vector to a low-dimensional space. For facial features, 68 key feature points, such as the corners of the eyes, nose, and mouth, can be extracted; for fingerprint features, approximately 30-40 feature points can be extracted, including endpoints and bifurcation points. The extracted key feature points form the authentication feature point set.

[0170] The system reads pre-stored registration feature vectors from the database. It then performs feature decomposition on these vectors to extract key feature points, generating a registration feature point set. For example, the fingerprint feature vector captured during registration can be decomposed into 35 key feature points, forming the registration feature point set.

[0171] During feature point matching, the system performs point-to-point matching between the authenticated feature point set and the registered feature point set. The matching process employs a nearest neighbor search strategy, searching for the feature point with the smallest Euclidean distance in the registered feature point set for each authenticated feature point. For example, if the authenticated feature point set has 32 points and the registered feature point set has 35 points, the matching may result in 30 feature point matching pairs, with 2 authenticated feature points failing to find a suitable match.

[0172] For each feature point pair, local similarity is calculated. Local similarity is based on the distance between feature points in the feature space; the closer the distance, the higher the similarity. Specifically, for a pair (pi, qi), where pi is the certified feature point and qi is the corresponding registered feature point, the Euclidean distance d(pi, qi) is calculated. This distance is then mapped to the range 0-1 using the transformation d_norm = 1 / (1+d(pi, qi)) to represent the local similarity. For example, if the Euclidean distance of a pair is 0.2, its local similarity is 1 / (1+0.2) = 0.833.

[0173] The weight coefficient of each feature point is determined based on local similarity. The weight coefficient reflects the importance of the feature point in the overall matching. Feature points with high local similarity receive larger weights, while feature points with low local similarity receive smaller weights. In practice, the local similarity can be normalized to obtain the weight coefficient. For example, for 30 matching pairs, the calculated local similarities are [0.833, 0.762, 0.915, ... etc. 30 values], and after normalization, the weight coefficients are [0.032, 0.029, 0.035, ... etc. 30 values], with the sum of all weight coefficients being 1.

[0174] The feature point matching pairs are weighted according to weight coefficients to generate a global similarity. The global similarity is calculated as a weighted sum of the local similarities of each matching pair. For example, if the local similarities of 30 matching pairs are [0.833, 0.762, 0.915, ... etc. 30 values], and the corresponding weight coefficients are [0.032, 0.029, 0.035, ... etc. 30 values], then the global similarity = 0.833×0.032 + 0.762×0.029 + 0.915×0.035 + ... = 0.875.

[0175] The calculated global similarity is compared with a preset target similarity threshold. The target similarity threshold is set according to security requirements, typically between 0.7 and 0.9. For example, if the target similarity threshold is set to 0.8 and the global similarity is 0.875, the system determines that authentication has passed since 0.875 > 0.8. If the global similarity is 0.765, the system determines that authentication has failed since 0.765 < 0.8.

[0176] In practical applications, the target similarity threshold can be dynamically adjusted according to different security levels. For scenarios with high security requirements, such as financial payments, the threshold can be set to 0.85; for scenarios with general security requirements, such as app unlocking, the threshold can be set to 0.75.

[0177] The system can also set different target similarity thresholds for different biometric types. For example, the threshold for fingerprint authentication can be set to 0.82, the threshold for face authentication can be set to 0.78, and the threshold for iris authentication can be set to 0.88.

[0178] When partial mismatches occur in feature matching, the system will focus on analyzing the matching of key feature points. For example, for facial features, if the feature points of key areas such as the eyes and nose have a high matching degree, while the feature points of facial edges have a low matching degree, authentication may still be successful. This weighted strategy increases the system's tolerance for slight changes while maintaining strict requirements for key features.

[0179] This method effectively improves the accuracy and robustness of feature matching through feature decomposition and local weight calculation, and can adapt to various biometric recognition scenarios, providing reliable technical support for identity authentication.

[0180] This invention provides a multi-dimensional biometric recognition and authentication system for digital payments, comprising:

[0181] The first unit is used to collect the user's first biometric data, and based on the first biometric data, the corresponding feature vector is extracted by the biometric processing unit to generate a multidimensional feature set.

[0182] The second unit is used to select a corresponding number of biometric features from the multidimensional feature set according to the distribution of payment amount within a preset amount range for authentication, thereby obtaining a feature set to be authenticated;

[0183] The third unit is used to obtain the user's second biometric data based on the feature set to be authenticated, and to extract the corresponding authentication feature vector based on the second biometric data.

[0184] The fourth unit is used to obtain the user's identity authentication environment parameters, calculate the environment credibility based on the identity authentication environment parameters, and dynamically adjust the similarity threshold of feature matching based on the environment credibility to obtain the target similarity threshold.

[0185] The fifth unit is used to perform feature matching between the authentication feature vector and the pre-stored registration feature vector, and determine whether the identity authentication is successful based on the similarity of the feature matching and the target similarity threshold.

[0186] The sixth unit is used to send authentication success information to the payment system when identity authentication is confirmed to be successful, thereby triggering the execution of the payment process.

[0187] A third aspect of the present invention provides an electronic device, comprising:

[0188] processor;

[0189] Memory used to store processor-executable instructions;

[0190] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0191] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0192] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for multi-dimensional biometric identification and identity authentication for digital payment, characterized in that, The method comprises the following steps: Collecting first biometric data of a user, extracting corresponding feature vectors based on the first biometric data through a biometric processing unit, and generating a multi-dimensional feature set; Selecting a corresponding number of biometrics for authentication from the multi-dimensional feature set according to the distribution of payment amounts in a preset amount interval to obtain a feature set to be authenticated; Obtaining second biometric data of the user according to the feature set to be authenticated, and extracting corresponding authentication feature vectors based on the second biometric data; Obtaining identity authentication environment parameters of the user, calculating an environment trustworthiness according to the identity authentication environment parameters, and dynamically adjusting a similarity threshold of feature matching based on the environment trustworthiness to obtain a target similarity threshold; Performing feature matching on the authentication feature vectors and pre-stored registration feature vectors, determining whether the identity authentication passes based on the similarity of feature matching and the target similarity threshold; When it is confirmed that the identity authentication passes, sending authentication success information to a payment system to trigger the execution of a payment process; Based on the first biometric data, the biometric processing unit extracts corresponding feature vectors to generate a multi-dimensional feature set, including: Performing image preprocessing on the first biometric data to obtain preprocessed biometric data; Performing multi-dimensional image quality assessment on the preprocessed biometric data, and calculating an image quality comprehensive score according to the evaluation results of the multi-dimensional image quality assessment; Based on the image quality comprehensive score, performing partition block adaptive enhancement processing on the preprocessed biometric data, wherein an image enhancement parameter greater than a preset intensity is used for an image block with an image quality comprehensive score less than a preset score threshold, and an image enhancement parameter less than a preset intensity is used for an image block with an image quality comprehensive score greater than the preset score threshold, to obtain enhanced biometric data; Based on the image quality comprehensive score, performing feature completion and feature enhancement on the enhanced biometric data to obtain optimized biometric data, extracting key feature points in the optimized biometric data to obtain a feature point set; Based on the feature point set, a plurality of biometric feature vectors are constructed, feature fusion processing is performed on the plurality of biometric feature vectors, and a multi-dimensional feature set is generated.

2. The method of claim 1, wherein, From the multi-dimensional feature set, a corresponding number of biometrics are selected for authentication according to the distribution of payment amounts in a preset amount interval to obtain a feature set to be authenticated, including: Constructing a payment amount interval distribution table, which records the corresponding relationship between the amount interval and the number of biometric authentication; Obtaining historical transaction data of the user and historical biometric authentication records of the user, calculating payment habit parameters of the user according to the historical transaction data, and obtaining a payment risk assessment value based on the payment habit parameters; Based on the payment risk assessment value, the amount interval in the payment amount interval distribution table is adjusted to obtain an adjusted amount interval. The higher the payment risk assessment value, the smaller the adjusted amount interval. According to the historical biometric authentication record and the payment risk assessment value, a credibility score of a biometric feature is calculated, and the biometric features in the multi-dimensional feature set are dynamically weighted based on the credibility score, to obtain a weighted multi-dimensional feature set; An interval position of the current payment amount in the adjusted amount interval is determined, and the number of biometric features that need to be authenticated is obtained from the payment amount interval distribution table according to the interval position; The biometric feature with the largest weight is selected from the weighted multi-dimensional feature set according to the number of biometric features, to obtain a feature set to be authenticated.

3. The method of claim 2, wherein, According to the historical biometric authentication record and the payment risk assessment value, a credibility score of a biometric feature is calculated, and the biometric features in the multi-dimensional feature set are dynamically weighted based on the credibility score, to obtain a weighted multi-dimensional feature set, comprising: A time decay coefficient of each historical biometric feature is calculated based on the historical biometric authentication record, and the time decay coefficient decreases as the authentication time interval increases, to obtain a time weighting coefficient; A risk weighting coefficient is determined according to the payment risk assessment value, and the risk weighting coefficient increases as the payment risk assessment value increases; The historical biometric authentication record is classified according to authentication scenarios, and the authentication accuracy of biometric features under each authentication scenario is calculated; A scenario similarity matrix is constructed based on the authentication accuracy, the current payment scenario is matched with the scenario similarity matrix, the biometric feature authentication success probability under the authentication scenario with the highest matching degree is extracted, and a scenario weighting coefficient is calculated according to the biometric feature authentication success probability; The time weighting coefficient, the risk weighting coefficient and the scenario weighting coefficient are combined to calculate a credibility score of a biometric feature; Based on the credibility score, the biometric features in the multi-dimensional feature set are dynamically weighted to obtain a weighted multi-dimensional feature set.

4. The method of claim 1, wherein, An identity authentication environment parameter of a user is obtained, an environment credibility is calculated according to the identity authentication environment parameter, and a target similarity threshold is obtained by dynamically adjusting a similarity threshold of feature matching based on the environment credibility, comprising: An identity authentication environment parameter and a historical authentication environment parameter of a user are obtained, the historical authentication environment parameter is hierarchically processed according to an authentication success rate, and an authentication environment reference value is constructed for each layer of the historical authentication environment parameter; The identity authentication environment parameter and the authentication environment reference value are multi-dimensionally matched and calculated, the identity authentication environment parameter is verified for effectiveness according to the matching result, and an environment parameter matching degree is obtained; Historical authentication behavior data of a user is obtained, a user behavior trajectory in an authentication process is extracted from the historical authentication behavior data, an authentication behavior feature curve is constructed based on the user behavior trajectory, and a user operation behavior feature value is extracted from the authentication behavior feature curve; The user operation behavior feature value is divided into a plurality of feature segments according to a time window, a fluctuation trend of the feature segments is calculated to obtain a user behavior stability index, and an environment credibility is determined according to the user behavior stability index; Generate a threshold adjustment coefficient based on the environment credibility, dynamically adjust the reference similarity threshold of feature matching according to the threshold adjustment coefficient, and obtain a target similarity threshold.

5. The method of claim 4, wherein, The user operation behavior feature value is divided into multiple feature segments according to a time window, the fluctuation trend of the feature segments is calculated to obtain a user behavior stability index, and the environment credibility is determined according to the user behavior stability index, including: A behavior feature reference library is constructed, authentication behavior trajectories are extracted from historical authentication success records, the authentication behavior trajectories are data cleaned, and behavior feature reference values are generated; The user operation behavior feature value is segmented and normalized according to the time window size to generate multiple normalized feature segments; The reference weight of each normalized feature segment is calculated based on the behavior feature reference value, the normalized feature segments are reconstructed according to the reference weight, and a reconstructed feature sequence is obtained; The reconstructed feature sequence is constructed into a feature fluctuation matrix in chronological order, and the difference between adjacent feature segments is calculated to obtain a fluctuation subsequence; The fluctuation subsequence is recursively decomposed to extract the fluctuation trend feature, and a fluctuation trend function is generated; The user behavior stability index is calculated based on the fluctuation trend function, and the environment credibility is determined according to the deviation degree of the user behavior stability index and the behavior feature reference value.

6. The method of claim 1, wherein, The authentication feature vector and the pre-stored registration feature vector are matched, and based on the similarity of feature matching and the target similarity threshold, it is determined whether the identity authentication is passed, including: The authentication feature vector is decomposed to extract key feature points of the authentication feature vector, and an authentication feature point set is generated; The authentication feature vector is decomposed to extract key feature points of the authentication feature vector, and an authentication feature point set is generated; The authentication feature point set and the registration feature point set are matched point by point to obtain a feature point matching pair, the local similarity of each feature point in the feature point matching pair is calculated, and the weight coefficient of each feature point is determined based on the local similarity; The feature point matching pair is calculated according to the weight coefficient to generate a global similarity of feature matching; The global similarity is compared with the target similarity threshold, when the global similarity is greater than the target similarity threshold, it is determined that the identity authentication is passed, and when the global similarity is less than or equal to the target similarity threshold, it is determined that the identity authentication is not passed.

7. A multi-dimensional biometric identification and identity authentication system for digital payment, for implementing the method according to any one of claims 1-6, characterized in that, Including: The first unit is used for collecting the first biological feature data of the user, extracting the corresponding feature vector through the biological feature processing unit based on the first biological feature data, and generating a multi-dimensional feature set; The second unit is used for selecting a corresponding number of biological features for authentication from the multi-dimensional feature set according to the distribution of the payment amount in the preset amount interval to obtain a to-be-authenticated feature set; The third unit is used for obtaining the second biological feature data of the user according to the to-be-authenticated feature set, and extracting the corresponding authentication feature vector based on the second biological feature data; The fourth unit is configured to acquire an identity authentication environment parameter of a user, calculate an environment credibility according to the identity authentication environment parameter, and dynamically adjust a similarity threshold of feature matching based on the environment credibility to obtain a target similarity threshold. The fifth unit is configured to perform feature matching on the authentication feature vector and a pre-stored registration feature vector, and determine whether the identity authentication passes based on a similarity of the feature matching and the target similarity threshold. The sixth unit is configured to send authentication success information to a payment system to trigger execution of a payment process when it is confirmed that the identity authentication passes. The first unit is configured to: perform image preprocessing on the first biological feature data to obtain preprocessed biological feature data; perform multi-dimensional image quality evaluation on the preprocessed biological feature data, and calculate an image quality comprehensive score according to an evaluation result of the multi-dimensional image quality evaluation; perform block-adaptive enhancement processing on the preprocessed biological feature data based on the image quality comprehensive score, wherein an image enhancement parameter greater than a preset intensity is used for an image block with an image quality comprehensive score less than a preset score threshold, and an image enhancement parameter less than the preset intensity is used for an image block with an image quality comprehensive score greater than the preset score threshold, to obtain enhanced biological feature data; perform feature completion and feature enhancement on the enhanced biological feature data based on the image quality comprehensive score to obtain optimized biological feature data, extract key feature points in the optimized biological feature data to obtain a feature point set, and construct a plurality of biological feature vectors based on the feature point set. The computer program instructions are executed by the processor to implement the method of any one of claims 1-6.

8. An electronic device, comprising: The computer program instructions are executed by the processor to implement the method of any one of claims 1-6. The computer program instructions are executed by the processor to implement the method of any one of claims 1-6. ​ ​ 9. A computer-readable storage medium having stored thereon computer program instructions, wherein, ​

Images