Biological recognition method and device, storage medium and computer equipment

Abstract

According to the biological recognition method and device, the storage medium and the computer equipment, the hardware parameters and the sample feature sets of the new equipment and the old equipment are collected, and accurate conversion from the old template features to the new equipment features is achieved. The conversion process not only considers hardware differences, but also combines multiple groups of sample features, and ensures that the converted new template features can fully adapt to the hardware environment of new equipment. Meanwhile, local fine tuning is performed on the feature extraction model in the new equipment by using the feature image set of a small number of users and the corresponding new template features, so that the recognition accuracy of the model on the new equipment is further improved. The local fine tuning mode avoids the problems of large data transmission quantity, privacy leakage risk, high operation and maintenance cost, equipment elimination and the like caused by full data synchronization and unified algorithm framework forced upgrading. In addition, the method is also helpful for promoting the popularization and deepening of a biological recognition technology in more fields and scenes.

Description

Technical Field

[0001] This application relates to the field of biometric technology, and particularly to a biometric method, device, storage medium, and computer device. Background Art

[0002] In existing biometric systems, due to hardware parameter differences (including image sensor resolution, light source spectral range, and acquisition angle deviation) and algorithm version differences (such as iterations of feature extraction models and adjustments of matching thresholds) between different devices (such as panel machine A and Android machine B), the biometric template data registered by old users on panel machine A cannot be effectively recognized directly on Android machine B. Users need to re-collect and register on Android machine B, which not only reduces the user experience but also increases the device management cost and data redundancy risk in enterprise-level scenarios.

[0003] For the above technical problems, traditional solutions include: 1) Full data synchronization: Synchronize the complete templates and algorithm parameters of panel machine A to Android machine B; 2) Forced upgrade of the unified algorithm framework: Require all devices to use the same version of the algorithm. However, the full data synchronization method is limited by hardware compatibility. When the hardware differences are large, the recognition accuracy will drop significantly, and the data transmission volume is large, there is a risk of privacy leakage, and large-scale firmware updates are required; while the method of forced upgrade of the unified algorithm framework has high operation and maintenance costs, and may also lead to device elimination due to insufficient hardware performance of old devices. Therefore, neither of these two methods can achieve a balance among ensuring recognition accuracy, reducing user operation costs, and controlling operation and maintenance costs, and it is difficult to meet the requirements of multi-device linkage scenarios. Summary of the Invention

[0004] The purpose of this application is to at least solve one of the above technical defects, especially the technical defect that when replacing old and new devices in the prior art, it is impossible to achieve a balance among ensuring recognition accuracy, reducing user operation costs, and controlling operation and maintenance costs, and it is difficult to meet the requirements of multi-device linkage scenarios.

[0005] This application provides a biometric method, and the method includes:

[0006] When replacing old and new devices, respectively collect the new hardware parameters of the new device and the old hardware parameters of the old device, and use the new device and the old device to synchronously collect new sample feature sets and old sample feature sets corresponding to multiple groups of samples;

[0007] According to the new hardware parameters, the old hardware parameters, the new sample feature sets, and the old sample feature sets, convert the old template features of all users in the target user group saved by the old device into new template features that conform to the hardware characteristics of the new device;

[0008] The new device is used to collect a set of feature images corresponding to a small number of users in the target user group, and the feature extraction model in the new device is locally fine-tuned using the set of feature images and the new template features corresponding to the small number of users.

[0009] When any user in the target user group is identified in the new device, the fine-tuned feature extraction model is used to extract features from the user, and the extracted features are matched with the new template features corresponding to the user. The identification is then determined based on the matching results.

[0010] Optionally, the step of converting the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, includes:

[0011] Based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, determine the feature transformation function that converts the old template features of the old device into the new template features of the new device;

[0012] Using the feature transformation function, the old template features of all users in the target user group stored in the old device are transformed into new template features that conform to the hardware characteristics of the new device.

[0013] Optionally, determining the feature transformation function for converting the old template features of the old device into the new template features of the new device based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set includes:

[0014] Determine the device feature mapping model, and the feature difference vector set between the new sample feature set and the old sample feature set;

[0015] The new hardware parameters, the old hardware parameters, and the old sample feature set are used as the training sample set, and the feature difference vector set is used as the sample label set.

[0016] Each training sample in the training sample set is sequentially input into the device feature mapping model to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model.

[0017] With the goal of making each predicted feature difference approach each feature difference vector in the sample label set, the device feature mapping model is iteratively trained until the preset training target is reached, thus obtaining the trained device feature mapping model.

[0018] The trained device feature mapping model is used as a feature transformation function to convert the old template features of the old device into the new template features of the new device.

[0019] Optionally, the device feature mapping model is an ensemble model composed of multiple regression trees;

[0020] The step of sequentially inputting each training sample in the training sample set into the device feature mapping model to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model includes:

[0021] For each training sample in the training sample set:

[0022] The new hardware parameters, old hardware parameters, and old sample features from the training sample are input into each regression tree to obtain the residual correction value output by each regression tree.

[0023] The weighted summation of the residual correction values ​​output by each regression tree yields the predicted feature difference between the old sample and the corresponding new sample.

[0024] Optionally, the feature extraction model in the new device includes a feature fusion layer, a classification layer, and a feature extraction layer;

[0025] The step of locally fine-tuning the feature extraction model in the new device using the feature image set and the new template features corresponding to the small number of users includes:

[0026] The parameters in the feature extraction layer are fixed, and the parameters of the feature fusion layer and the classification layer are fine-tuned using the feature image set and the new template features corresponding to the small number of users, and a mini-batch gradient descent algorithm is used.

[0027] Optionally, determining whether the identification is successful based on the matching result includes:

[0028] When any user is a user being identified for the first time on the new device, the initial matching threshold of the new device is obtained, as well as the matching result obtained by matching the extracted features with the new template features corresponding to the any user;

[0029] The initial matching threshold is dynamically adjusted based on the matching results, and the recognition of any user is determined based on the adjusted initial matching threshold.

[0030] Optionally, the method further includes:

[0031] Obtain the matching threshold when any user is successfully identified multiple times consecutively;

[0032] The optimal matching threshold for any user is determined based on the number of times the user is identified and the matching threshold at each successful identification.

[0033] The optimal matching threshold is used to confirm whether the identification is successful when the user subsequently performs identification on the new device.

[0034] This application also provides a biometric device, including:

[0035] The information acquisition module is used to collect the new hardware parameters of the new device and the old hardware parameters of the old device when replacing old devices with new ones, and to simultaneously collect new sample feature sets and old sample feature sets corresponding to multiple sets of samples using the new device and the old device.

[0036] The feature conversion module is used to convert the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set.

[0037] The local fine-tuning module is used to collect a set of feature images corresponding to a small number of users in the target user group using the new device, and to perform local fine-tuning on the feature extraction model in the new device using the set of feature images and the new template features corresponding to the small number of users.

[0038] The biometric module is used to extract features from any user in the target user group when the new device is used for identification, using a fine-tuned feature extraction model, matching the extracted features with the new template features corresponding to the user, and determining whether the identification is successful based on the matching result.

[0039] This application also provides a computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the biometric method as described in any of the above embodiments.

[0040] This application also provides a computer device, including: one or more processors, and memory;

[0041] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the biometric method as described in any of the above embodiments.

[0042] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0043] The biometric identification method, device, storage medium, and computer equipment provided in this application achieve accurate conversion from old template features to new device characteristics by collecting hardware parameters and sample feature sets from both old and new devices. This conversion process not only considers hardware differences but also incorporates multiple sets of sample features, ensuring that the converted new template features can fully adapt to the hardware environment of the new device. Simultaneously, this application uses a small set of user feature images and corresponding new template features to locally fine-tune the feature extraction model in the new device, further improving the model's recognition accuracy on the new device. This local fine-tuning method avoids the problems of large data transmission volumes, privacy leakage risks, high maintenance costs, and device obsolescence caused by full data synchronization and forced upgrades to a unified algorithm framework. When any user in the target user group is identified on the new device, the finely tuned feature extraction model can be used to extract features from any user, and the extracted features can be matched with the corresponding new template features for any user. The matching result can then be used to determine whether the identification is successful.

[0044] In practical applications, this application fully considers the hardware differences between different devices and uses multiple sets of sample features for accurate conversion, enabling new devices to quickly integrate into existing biometric systems without requiring large-scale hardware modifications or software upgrades. Furthermore, by fine-tuning the feature extraction model locally, this application not only improves recognition accuracy but also reduces reliance on full data, effectively protecting user privacy while reducing data transmission and maintenance costs. This method provides a new approach for the cross-device application of biometric technology, helping to promote the popularization and deepening of biometric technology in more fields and scenarios. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a biometric identification method provided in an embodiment of this application;

[0047] Figure 2 This is a schematic diagram illustrating the process of converting old template features according to an embodiment of this application;

[0048] Figure 3 This is a schematic diagram illustrating the process of determining whether a user has been successfully identified based on the matching results, provided in an embodiment of this application.

[0049] Figure 4This is a schematic diagram of the structure of a biometric device provided in an embodiment of this application;

[0050] Figure 5 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] In one embodiment, such as Figure 1 As shown, Figure 1 This is a flowchart illustrating a biometric identification method provided in an embodiment of this application; this application provides a biometric identification method, which may include:

[0053] S110: When replacing old equipment with new equipment, collect the new hardware parameters of the new equipment and the old hardware parameters of the old equipment respectively, and use the new equipment and the old equipment to simultaneously collect the new sample feature set and the old sample feature set corresponding to multiple sets of samples.

[0054] In this step, when replacing old equipment with new equipment, we can collect the new hardware parameters of the new equipment and the old hardware parameters of the old equipment. The new equipment refers to the new biometric device that will soon be put into use. Its hardware parameters may include key indicators such as image sensor pixel size, dynamic range, light source wavelength distribution, default focal length and angle, etc. These parameters directly affect the quality of biometric feature acquisition. The old equipment refers to the biometric device that is currently in use and will be replaced, which also has its specific hardware parameters. By simultaneously collecting multiple sets of samples, we can obtain new sample feature sets and old sample feature sets. These feature sets contain biometric information collected under different hardware environments, providing a data foundation for subsequent feature conversion.

[0055] It should be noted that during the data collection process, this application ensures that the collection conditions of the new and old devices are as consistent as possible to reduce feature deviations caused by environmental differences. For example, this application can simultaneously collect 10-20 sets of standard palm samples on two devices. These standard palm samples need to cover different palm sizes (15-20cm palm length) and different skin types (dry / oily / normal), and the collection environment (light intensity 500-800 lux, collection distance 10-15cm) for each set of samples should remain consistent when using the new and old devices. This step breaks through the limitations of traditional methods that only collect parameters from a single device. By collecting the same sample across devices, a direct correspondence between hardware parameters and feature differences is established, laying the foundation for subsequent accurate mapping. Furthermore, by collecting simultaneously, this application can ensure the temporal consistency of the feature sets of the new and old samples, thereby more accurately reflecting the differences between the new and old devices.

[0056] S120: Based on the new hardware parameters, old hardware parameters, new sample feature set, and old sample feature set, convert the old template features of all users in the target user group saved by the old device into new template features that conform to the hardware characteristics of the new device.

[0057] In this step, the new hardware parameters of the new device and the old hardware parameters of the old device are collected by S110 respectively. After the new and old devices are used to collect new sample feature sets and old sample feature sets corresponding to multiple sets of samples simultaneously, this application can convert the old template features of all users in the target user group saved by the old device into new template features that conform to the hardware characteristics of the new device based on the new hardware parameters, old hardware parameters, new sample feature sets and old sample feature sets.

[0058] It is understood that the old devices in this application store old template features of various users in the target user group. These old template features are generated based on the hardware parameters and acquisition conditions of the old devices. The target user group here refers to the set of all users who use the old devices for biometric identification and have stored template features. These users may include different groups such as corporate employees, community residents, and visitors to specific locations. When replacing devices, if the old template features are directly used for identification on the new devices, the identification accuracy will often decrease due to hardware differences.

[0059] Therefore, this application constructs a mapping relationship between hardware parameters and feature differences by collecting new hardware parameters of the new device and old hardware parameters of the old device, and combining them with new and old sample feature sets corresponding to multiple sets of samples collected simultaneously. This mapping relationship can accurately reflect the differences in biometrics under different hardware environments, thereby converting old template features into new template features that conform to the hardware characteristics of the new device. This conversion process not only considers the differences in hardware parameters but also combines multiple sets of sample features, making the converted new template features closer to the actual collection situation of the new device, thus improving the accuracy and reliability of recognition. Furthermore, this conversion method does not require large-scale hardware modifications or software upgrades to the old device; it can be achieved simply by collecting relevant parameters and sample feature sets, reducing upgrade costs and maintenance difficulties.

[0060] S130: Use the new device to collect a set of feature images corresponding to a small number of users in the target user group, and use the set of feature images and the new template features corresponding to the small number of users to make local fine-tuning of the feature extraction model in the new device.

[0061] In this step, after converting the old template features of all users in the target user group saved by the old device into new template features that conform to the hardware characteristics of the new device through S120, this application can also use the new device to collect feature image sets corresponding to a small number of users in the target user group, and use the feature image sets and the new template features corresponding to a small number of users to locally fine-tune the feature extraction model in the new device. In this way, the fine-tuned feature extraction model can be used to extract the user's biometric features.

[0062] Understandably, in traditional technologies, there are usually two methods to enable new devices to recognize features from old templates: one is to completely replace the algorithm model (which will increase the computational burden on the device), and the other is to add a large number of samples for retraining (resulting in high user cooperation costs). However, this application, by adopting a design scheme of hierarchical fine-tuning and pairing with a small number of samples, can significantly reduce the dependence on computing power and data while ensuring the adaptation effect.

[0063] During local fine-tuning, this application can use new equipment to collect biometric images of a small number of users in the target user group (e.g., 10-20 users, the specific number can be flexibly adjusted according to the size of the target user group and actual needs). These images should cover different biometric types (such as fingerprints, palm prints, facial features, etc.) and different acquisition conditions (such as different light intensities, angles, etc.). Then, these feature image sets are compared with the corresponding new template features to analyze the accuracy and bias of the model in extracting these features. Next, based on the comparison results, the relevant parameters in the feature extraction model (such as convolution kernel size, stride, activation function, etc.) are adjusted to enable the model to extract the biometric features of these users more accurately.

[0064] This local fine-tuning approach has several advantages. First, it only requires a small set of feature images from a few users, eliminating the need for full data collection and processing from all users, thus significantly reducing data transmission volume and processing costs. Second, since the fine-tuning process is performed on new devices, it better adapts to the hardware characteristics of the new devices, improving the model's recognition accuracy on them. Furthermore, local fine-tuning avoids the privacy risks and increased maintenance costs associated with full data synchronization and forced upgrades to a unified algorithm framework.

[0065] S140: When any user in the target user group is identified in the new device, the fine-tuned feature extraction model is used to extract features from any user, and the extracted features are matched with the new template features corresponding to any user. The identification is then determined based on the matching results.

[0066] In this step, after collecting a set of feature images of a small number of users in the target user group using the new device via S130 and fine-tuning the feature extraction model locally, the recognition process can be started when any user in the target user group is identified on the new device.

[0067] Specifically, this application first utilizes a finely tuned feature extraction model to extract biometric features from any user. This model, after local fine-tuning, is better adapted to the hardware characteristics of the new device and extracts the user's biometric features more accurately. Then, the extracted features are matched with a new template feature corresponding to the user. This new template feature, derived from the hardware parameters and sample feature sets of both the old and new devices, fully reflects the user's biometric performance on the new device. Finally, the matching result determines whether the identification is successful. If the matching degree reaches a preset threshold, the identification is successful, allowing the user to proceed with subsequent operations; if the matching degree does not reach the preset threshold, the identification fails, denying the user access or prompting the user to re-identify. This process not only improves the accuracy of identification but also ensures the security and reliability of the biometric system.

[0068] In the above embodiments, by collecting hardware parameters and sample feature sets from both old and new devices, a precise conversion from old template features to new device characteristics is achieved. This conversion process not only considers hardware differences but also incorporates multiple sets of sample features, ensuring that the converted new template features can fully adapt to the hardware environment of the new device. Simultaneously, this application utilizes a small set of user feature images and corresponding new template features to locally fine-tune the feature extraction model in the new device, further improving the model's recognition accuracy on the new device. This local fine-tuning method avoids the problems of large data transmission volumes, privacy leakage risks, high maintenance costs, and device obsolescence caused by full data synchronization and forced upgrades of a unified algorithm framework. When any user in the target user group is identified on the new device, the finely tuned feature extraction model can be used to extract features from any user, and the extracted features can be matched with the corresponding new template features of any user. This allows the determination of whether the identification is successful based on the matching results.

[0069] In practical applications, this application fully considers the hardware differences between different devices and uses multiple sets of sample features for accurate conversion, enabling new devices to quickly integrate into existing biometric systems without requiring large-scale hardware modifications or software upgrades. Furthermore, by fine-tuning the feature extraction model locally, this application not only improves recognition accuracy but also reduces reliance on full data, effectively protecting user privacy while reducing data transmission and maintenance costs. This method provides a new approach for the cross-device application of biometric technology, helping to promote the popularization and deepening of biometric technology in more fields and scenarios.

[0070] In one embodiment, such as Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the process of converting old template features according to an embodiment of this application; S120, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, converts the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, which may include:

[0071] S121: Based on the new hardware parameters, old hardware parameters, new sample feature set, and old sample feature set, determine the feature transformation function that converts the old template features of the old device into the new template features of the new device.

[0072] S122: Using a feature transformation function, convert the old template features of all users in the target user group saved on the old device into new template features that conform to the hardware characteristics of the new device.

[0073] In this embodiment, when converting the old template features of all users in the target user group saved on the old device into new template features that conform to the hardware characteristics of the new device, this application first determines a feature transformation function that can convert the old template features of the old device into the new template features of the new device based on the new hardware parameters, old hardware parameters, new sample feature set, and old sample feature set. This feature transformation function is the core of the entire transformation process. It comprehensively considers the differences in hardware parameters between the old and new devices as well as the differences in sample feature sets. Through complex algorithms and model construction, it ensures that the old template features can be accurately mapped to the feature space corresponding to the new device.

[0074] Once the feature transformation function is determined, this application can utilize this function to transform the old template features of all users in the target user group stored on the old device. This process is performed in batches, eliminating the need for individual processing for each user, thus significantly improving transformation efficiency. Through the feature transformation function, the old template features are accurately transformed into new template features that conform to the hardware characteristics of the new device. These new template features are better adapted to the acquisition environment and recognition requirements of the new device, providing a reliable data foundation for subsequent biometric identification processes.

[0075] This conversion method not only considers differences in hardware parameters but also incorporates features from multiple sets of samples, making the converted template features more closely resemble the actual acquisition conditions of the new device. Compared to traditional methods of directly replacing or simply adjusting template features, the method presented in this application offers higher accuracy and reliability, effectively avoiding recognition errors or decreased recognition rates caused by hardware differences. Furthermore, since the conversion process is based on a feature transformation function, it also possesses a degree of universality and scalability, making it applicable to template feature conversion between different types and specifications of biometric devices.

[0076] In one embodiment, determining a feature transformation function in S121 to convert the old template features of the old device into the new template features of the new device based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set may include:

[0077] S1211: Determine the device feature mapping model, and the feature difference vector set between the new sample feature set and the old sample feature set.

[0078] S1212: Use the new hardware parameters, the old hardware parameters, and the old sample feature set as the training sample set, and use the feature difference vector set as the sample label set.

[0079] S1213: Input each training sample in the training sample set into the device feature mapping model in sequence to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model.

[0080] S1214: With the goal of making each predicted feature difference approach each feature difference vector in the sample label set, the device feature mapping model is iteratively trained until the preset training target is reached, and the trained device feature mapping model is obtained.

[0081] S1215: Use the trained device feature mapping model as a feature transformation function to convert the old template features of the old device into the new template features of the new device.

[0082] In this embodiment, when determining the feature transformation function to convert the old template features of the old device into the new template features of the new device, this application first determines the device feature mapping model. This model is the basic framework for constructing the feature transformation function, and its design must fully consider the hardware characteristics of the old and new devices, as well as the complexity and diversity of sample features. Simultaneously, this application can calculate the feature difference vector set between the new sample feature set and the old sample feature set. This vector set can intuitively reflect the degree of deviation of sample features under different device acquisition conditions, providing crucial guidance for subsequent model training.

[0083] For example, when calculating the feature difference vector set between the new and old sample feature sets, this application can calculate the Euclidean distance, cosine similarity, and other metrics between the feature vectors acquired on the new and old devices for each sample pair, thereby obtaining a set of feature difference vectors. These feature difference vectors include, but are not limited to, contour feature deviation values, vein texture contrast differences, and feature point coordinate offsets. These vectors not only contain differences in feature values ​​but also implicitly reflect the influence of factors such as the acquisition environment and hardware performance of different devices on feature performance. In this way, this application can more comprehensively capture the feature differences between new and old devices, providing rich and accurate data support for subsequent device feature mapping model training.

[0084] After determining the device feature mapping model and the feature difference vector set, this application can use the new hardware parameters, old hardware parameters, and old sample feature sets as the training sample set, and the feature difference vector set as the sample label set. This design enables the model to simultaneously learn the impact of hardware parameter differences and sample feature differences on feature transformation during training, thereby constructing a more accurate and robust feature transformation function.

[0085] Furthermore, this application can further decompose the feature difference vector into three quantifiable sub-feature dimensions (a total of 28 dimensions) as input features for the model. Taking palm features as an example, the sub-dimensions include: contour feature deviation (8 dimensions): including the offset of the vertex coordinates of the palm's outer contour polygon and the deviation of the palm area scaling ratio; vein texture feature difference (12 dimensions): including the contrast deviation of the main vein texture, the difference in texture continuity score, and the offset of the bifurcation point coordinates; and image attribute difference (8 dimensions): including the deviation of the mean image brightness, the difference in local sharpness, and the difference in noise variance. This detailed dimensional decomposition enables the model to more accurately capture the difference information at different feature levels, thereby improving the adaptability of the feature transformation function to complex hardware differences and changes in sample features.

[0086] During training, the device feature mapping model can continuously adjust its internal parameters based on the input training sample set (new hardware parameters, old hardware parameters, and old sample feature set) and sample label set (feature difference vector set) through backpropagation. Specifically, the model can sequentially input training samples into the network layers, calculate the loss value between the predicted feature difference and the true feature difference vector, and use gradient descent to update the convolutional kernel weights, fully connected layer parameters, and activation function thresholds, so that the error between the predicted feature difference and the sample label gradually decreases after each iteration. For example, for the conversion from panel device A (cool white light source) to Android device B (warm yellow light source), the model will learn the rule that the contrast of vein texture under cool white light is 15%-20% higher than that under warm yellow light, and automatically reduce the value of the corresponding texture dimension during the conversion to match the light source characteristics of the target device. In this way, when the loss function value converges to the preset threshold or reaches the maximum number of iterations, the training process terminates, and the resulting device feature mapping model has the ability to map the old device template features to the new device feature space.

[0087] After iterative training, this application obtains a trained device feature mapping model. This model has fully learned the differences in hardware parameters and sample features between old and new devices, and can accurately convert the old template features of the old device into the new template features of the new device. Therefore, this application can use this trained device feature mapping model as a feature transformation function for subsequent template feature transformation processes.

[0088] In this way, this application can construct an accurate and robust feature transformation function, providing strong technical support for the cross-device application of biometric devices. This feature transformation function not only considers the differences in hardware parameters but also combines multiple sets of sample features, making the transformed new template features closer to the actual acquisition conditions of the new device, thus improving the accuracy and reliability of recognition.

[0089] In one embodiment, the device feature mapping model is an ensemble model composed of multiple regression trees.

[0090] In step S1213, each training sample in the training sample set is sequentially input into the device feature mapping model to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model. This can include:

[0091] S12131: For each training sample in the training sample set: input the new hardware parameters, old hardware parameters and old sample features in the training sample into each regression tree respectively to obtain the residual correction value output by each regression tree.

[0092] S12132: After weighted summation of the residual correction values ​​output by each regression tree, the predicted feature difference between the old sample and the corresponding new sample is obtained.

[0093] In this embodiment, when the device feature mapping model adopts an ensemble model composed of multiple regression trees, this application performs detailed processing on each training sample in the training sample set during the calculation of predicted feature differences. Specifically, for each training sample, this application can input its new hardware parameters, old hardware parameters, and old sample features into each regression tree in the ensemble model. Each regression tree can output a residual correction value for that training sample based on its own node splitting rules and leaf node values. This residual correction value reflects the degree of difference that may exist between the old sample features and the new sample features under the current regression tree structure.

[0094] Subsequently, this application can perform a weighted summation operation on the residual correction values ​​output by each regression tree. The weights here can be set according to the importance or performance of each regression tree in the ensemble model, or they can be processed using equal weights. Through weighted summation, this application can synthesize the prediction results of each regression tree to obtain a more accurate and comprehensive predicted feature difference between the old sample and the corresponding new sample features. The resulting feature transformation function can apply differentiated correction strategies for different feature dimensions, as shown in the following example:

[0095] Contour features: Calculate the scaling factor based on the difference in device focal length (e.g., 20mm focal length for panel camera → 15mm focal length for Android camera, multiply the contour coordinates by a scaling factor of 0.75).

[0096] Vein texture: Adjust texture contrast based on differences in light source spectrum (e.g., cool white light → warm yellow light, texture intensity value multiplied by a correction factor of 0.85).

[0097] Image attributes: Correct the average brightness value according to the sensor's dynamic range (e.g., dynamic range 8bit → 10bit, brightness value mapped to 0-1023 range).

[0098] This conversion function allows old device template data to be accurately converted into a format that conforms to the hardware characteristics of new devices.

[0099] In one embodiment, the feature extraction model in the new device may include a feature fusion layer, a classification layer, and a feature extraction layer.

[0100] S130, utilizing the feature image set and the new template features corresponding to the small number of users, performs local fine-tuning of the feature extraction model in the new device, which may include:

[0101] S131: Fix the parameters in the feature extraction layer, use the feature image set and the new template features corresponding to the small number of users, and use the mini-batch gradient descent algorithm to fine-tune the parameters of the feature fusion layer and the classification layer.

[0102] In this embodiment, when fine-tuning the feature extraction model in a new device using a feature image set and new template features corresponding to a small number of users, this application adopts a strategy of fixing the parameters of the feature extraction layer. This is because the feature extraction layer is mainly responsible for extracting basic biometric information from the original image. This information has a certain degree of universality across different devices, so frequent adjustments are unnecessary. By fixing the parameters of the feature extraction layer, this application can ensure that the model maintains its basic ability to recognize biometric features during fine-tuning, while avoiding performance degradation caused by excessive parameter adjustments.

[0103] Next, this application utilizes the feature image set and a small number of new template features corresponding to users to fine-tune the parameters of the feature fusion layer and classification layer using a mini-batch gradient descent algorithm. Mini-batch gradient descent is a commonly used optimization algorithm that effectively avoids getting trapped in local optima while ensuring training efficiency. During fine-tuning, this application divides the feature image set into multiple mini-batches, using only one mini-batch of data for training in each iteration. Furthermore, this application can set the learning rate to 0.001-0.005 and the number of iterations to 20-30 rounds, thus ensuring lightweight design and avoiding overloading the computing power of new devices, allowing fine-tuning to be completed within 1-2 hours. Compared to traditional full-scale training requiring several days and high-performance GPU support, this application improves fine-tuning efficiency by more than 10 times and is compatible with the computing power levels of ordinary Android devices.

[0104] Furthermore, the feature fusion layer in this application plays the role of fusing feature information from different levels within the model. By fine-tuning its parameters, this application enables the model to better capture the correlation and complementarity between different features, thereby improving recognition accuracy. The classification layer, on the other hand, is responsible for mapping the fused feature information to specific category labels. By fine-tuning its parameters, this application can optimize the model's classification boundary, enabling the model to make more accurate judgments when faced with biometric features on new devices.

[0105] After fine-tuning, the feature extraction model in the new device can better adapt to the hardware environment and acquisition conditions of the new device, while maintaining a high accuracy rate in recognizing biometric features. This local fine-tuning approach not only avoids many problems caused by full data synchronization and forced upgrades to a unified algorithm framework, but also provides a more flexible and efficient solution for cross-device applications of biometric technology.

[0106] In one embodiment, such as Figure 3 As shown, Figure 3 This is a schematic diagram illustrating the process of determining whether a user has passed identification based on the matching result in an embodiment of this application; S140, determining whether identification has passed based on the matching result, may include:

[0107] S141: When any user is the first user to be identified in a new device, obtain the initial matching threshold of the new device, and the matching result obtained by matching the extracted features with the new template features corresponding to any user.

[0108] S142: Dynamically adjust the initial matching threshold based on the matching results, and determine whether any user is successfully identified based on the adjusted initial matching threshold.

[0109] In this embodiment, when encountering a user being identified for the first time on a new device, this application first obtains an initial matching threshold preset by the new device. This threshold is set based on a large amount of experimental data and statistical laws, aiming to provide a reasonable judgment benchmark for initial identification. Subsequently, this application matches the extracted user biometric features with the new template features corresponding to the user to obtain a specific matching result. This matching result reflects the degree of similarity between the extracted features and the template features.

[0110] Next, this application can dynamically adjust the initial matching threshold based on the matching results. If the matching result value is high, it indicates that the extracted features are highly similar to the template features. In this case, the initial matching threshold can be appropriately reduced to make the subsequent recognition process more lenient and reduce recognition failures caused by subtle differences. Conversely, if the matching result value is low, it indicates that there are certain differences between the extracted features and the template features. In this case, the initial matching threshold is increased to raise the recognition standard and ensure that only highly similar features can pass the recognition.

[0111] After dynamic adjustment, an adjusted initial matching threshold is obtained. This application can compare the adjusted threshold with the matching result value. If the matching result value is greater than or equal to the adjusted threshold, it is determined that the user identification is successful, and the user is allowed to use the new device for subsequent operations; if the matching result value is less than the adjusted threshold, it is determined that the identification is unsuccessful, and the user is refused to use the new device or required to re-collect and identify biometric features.

[0112] This dynamic adjustment of matching thresholds fully considers the differences in biometric characteristics among different users and the actual data collection situation of new devices. It can flexibly adjust the recognition criteria according to the actual situation, improve the accuracy and reliability of recognition, and provide a more intelligent and user-friendly solution for the cross-device application of biometric technology.

[0113] In one embodiment, the method may further include:

[0114] S151: Obtain the matching threshold when any user is successfully identified multiple times consecutively.

[0115] S152: Determine the optimal matching threshold for any user based on the number of times the user is identified and the matching threshold when the user is successfully identified each time.

[0116] The optimal matching threshold is used to confirm whether the identification is successful when the user subsequently performs identification on the new device.

[0117] In this embodiment, in the practical application of biometric technology, for users who frequently use new devices, the collection of their biometric features on the new devices will gradually stabilize. At this time, if the initial matching threshold or the dynamically adjusted threshold is still used for identification, the optimal identification effect may not be achieved. Therefore, this application proposes a step of obtaining the matching threshold for any user to successfully identify multiple times consecutively.

[0118] Specifically, this application can record the matching thresholds corresponding to multiple consecutive successful identifications of any user on a new device. These matching thresholds reflect the stability and consistency of the user's biometric data collection on the new device. Subsequently, this application can perform comprehensive analysis and calculation based on the number of identifications of the arbitrary user and the matching thresholds at each successful identification to determine the optimal matching threshold for the arbitrary user.

[0119] This application can employ various methods to determine the optimal matching threshold. For example, the average of the matching thresholds from multiple consecutive successful recognitions can be calculated and used as the optimal matching threshold; alternatively, different weights can be assigned to the matching thresholds for different recognition counts, and then a weighted sum can be performed to obtain the optimal matching threshold. These methods can flexibly determine the optimal matching threshold based on the user's actual recognition performance.

[0120] After obtaining the optimal matching threshold, this application stores it in the new device and uses it to confirm whether the identification is successful when any user subsequently performs identification on the new device. In this way, for users who frequently use new devices, the identification of their biometrics on new devices will be more accurate and reliable, improving the user experience and the practicality of biometric technology.

[0121] The biometric device provided in the embodiments of this application is described below. The biometric device described below can be referred to in correspondence with the biometric method described above.

[0122] In one embodiment, such as Figure 4 As shown, Figure 4 This is a schematic diagram of a biometric device provided in an embodiment of this application. This application also provides a biometric device, which may include an information acquisition module 210, a feature conversion module 220, a local fine-tuning module 230, and a biometric module 240, specifically including the following:

[0123] The information acquisition module 210 is used to acquire the new hardware parameters of the new device and the old hardware parameters of the old device when replacing old devices with new ones, and to simultaneously acquire new sample feature sets and old sample feature sets corresponding to multiple sets of samples using the new device and the old device.

[0124] The feature conversion module 220 is used to convert the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set.

[0125] The local fine-tuning module 230 is used to collect a set of feature images corresponding to a small number of users in the target user group using the new device, and to perform local fine-tuning on the feature extraction model in the new device using the set of feature images and the new template features corresponding to the small number of users.

[0126] The biometric module 240 is used to extract features of any user in the target user group when the new device is used for identification, using a fine-tuned feature extraction model, matching the extracted features with the new template features corresponding to the user, and determining whether the identification is successful based on the matching result.

[0127] In the above embodiments, by collecting hardware parameters and sample feature sets from both old and new devices, a precise conversion from old template features to new device characteristics is achieved. This conversion process not only considers hardware differences but also incorporates multiple sets of sample features, ensuring that the converted new template features can fully adapt to the hardware environment of the new device. Simultaneously, this application utilizes a small set of user feature images and corresponding new template features to locally fine-tune the feature extraction model in the new device, further improving the model's recognition accuracy on the new device. This local fine-tuning method avoids the problems of large data transmission volumes, privacy leakage risks, high maintenance costs, and device obsolescence caused by full data synchronization and forced upgrades of a unified algorithm framework. When any user in the target user group is identified on the new device, the finely tuned feature extraction model can be used to extract features from any user, and the extracted features can be matched with the corresponding new template features of any user. This allows the determination of whether the identification is successful based on the matching results.

[0128] In practical applications, this application fully considers the hardware differences between different devices and uses multiple sets of sample features for accurate conversion, enabling new devices to quickly integrate into existing biometric systems without requiring large-scale hardware modifications or software upgrades. Furthermore, by fine-tuning the feature extraction model locally, this application not only improves recognition accuracy but also reduces reliance on full data, effectively protecting user privacy while reducing data transmission and maintenance costs. This method provides a new approach for the cross-device application of biometric technology, helping to promote the popularization and deepening of biometric technology in more fields and scenarios.

[0129] In one embodiment, this application also provides a computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the biometric method as described in any of the above embodiments.

[0130] In one embodiment, this application also provides a computer device, including: one or more processors, and memory.

[0131] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the biometric method as described in any of the above embodiments.

[0132] Indicatively, such as Figure 5 As shown, Figure 5 This is a schematic diagram of the internal structure of a computer device 300 provided in an embodiment of this application. The computer device 300 can be provided as a server. (Refer to...) Figure 5 The computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions, such as application programs, that can be executed by the processing component 302. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 302 is configured to execute instructions to perform the biometric methods of any of the above embodiments.

[0133] The computer device 300 may also include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate on an operating system stored in memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0134] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0135] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0136] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0137] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A biometric identification method, characterized in that, The method includes: When replacing old equipment with new equipment, the new hardware parameters of the new equipment and the old hardware parameters of the old equipment are collected respectively, and multiple sets of new sample feature sets and old sample feature sets corresponding to the new equipment and the old equipment are collected simultaneously. Based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, the old template features of all users in the target user group stored in the old device are converted into new template features that conform to the hardware characteristics of the new device. The new device is used to collect a set of feature images corresponding to a small number of users in the target user group, and the feature extraction model in the new device is locally fine-tuned using the set of feature images and the new template features corresponding to the small number of users. When any user in the target user group is identified in the new device, the fine-tuned feature extraction model is used to extract features from the user, the extracted features are matched with the new template features corresponding to the user, and the identification is determined based on the matching result.

2. The biometric identification method according to claim 1, characterized in that, The step of converting the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, includes: Based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set, determine the feature transformation function that converts the old template features of the old device into the new template features of the new device; Using the feature transformation function, the old template features of all users in the target user group stored in the old device are transformed into new template features that conform to the hardware characteristics of the new device.

3. The biometric identification method according to claim 2, characterized in that, The step of determining the feature transformation function to convert the old template features of the old device into the new template features of the new device based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set includes: Determine the device feature mapping model, and the feature difference vector set between the new sample feature set and the old sample feature set; The new hardware parameters, the old hardware parameters, and the old sample feature set are used as the training sample set, and the feature difference vector set is used as the sample label set. Each training sample in the training sample set is sequentially input into the device feature mapping model to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model. With the goal of making each predicted feature difference approach each feature difference vector in the sample label set, the device feature mapping model is iteratively trained until the preset training target is reached, thus obtaining the trained device feature mapping model. The trained device feature mapping model is used as a feature transformation function to convert the old template features of the old device into the new template features of the new device.

4. The biometric identification method according to claim 3, characterized in that, The device feature mapping model is an ensemble model composed of multiple regression trees; The step of sequentially inputting each training sample in the training sample set into the device feature mapping model to obtain the predicted feature difference between the old sample features and the corresponding new sample features in each training sample output by the device feature mapping model includes: For each training sample in the training sample set: The new hardware parameters, old hardware parameters, and old sample features from the training sample are input into each regression tree to obtain the residual correction value output by each regression tree. The weighted summation of the residual correction values ​​output by each regression tree yields the predicted feature difference between the old sample and the corresponding new sample.

5. The biometric identification method according to claim 1, characterized in that, The feature extraction model in the new device includes a feature fusion layer, a classification layer, and a feature extraction layer; The step of locally fine-tuning the feature extraction model in the new device using the feature image set and the new template features corresponding to the small number of users includes: The parameters in the feature extraction layer are fixed, and the parameters of the feature fusion layer and the classification layer are fine-tuned using the feature image set and the new template features corresponding to the small number of users, and a mini-batch gradient descent algorithm is used.

6. The biometric identification method according to claim 1, characterized in that, The step of determining whether the identification is successful based on the matching results includes: When any user is a user being identified for the first time on the new device, the initial matching threshold of the new device is obtained, as well as the matching result obtained by matching the extracted features with the new template features corresponding to the any user; The initial matching threshold is dynamically adjusted based on the matching results, and the recognition of any user is determined based on the adjusted initial matching threshold.

7. The biometric identification method according to claim 6, characterized in that, The method further includes: Obtain the matching threshold when any user is successfully identified multiple times consecutively; The optimal matching threshold for any user is determined based on the number of times the user is identified and the matching threshold at each successful identification. The optimal matching threshold is used to confirm whether the identification is successful when the user subsequently performs identification on the new device.

8. A biometric identification device, characterized in that, include: The information acquisition module is used to collect the new hardware parameters of the new device and the old hardware parameters of the old device when replacing old devices with new ones, and to simultaneously collect new sample feature sets and old sample feature sets corresponding to multiple sets of samples using the new device and the old device. The feature conversion module is used to convert the old template features of all users in the target user group stored in the old device into new template features that conform to the hardware characteristics of the new device, based on the new hardware parameters, the old hardware parameters, the new sample feature set, and the old sample feature set. The local fine-tuning module is used to collect a set of feature images corresponding to a small number of users in the target user group using the new device, and to perform local fine-tuning on the feature extraction model in the new device using the set of feature images and the new template features corresponding to the small number of users. The biometric module is used to extract features from any user in the target user group when the new device is used for identification, using a fine-tuned feature extraction model, matching the extracted features with the new template features corresponding to the user, and determining whether the identification is successful based on the matching result.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the biometric method as described in any one of claims 1 to 7.

10. A computer device, characterized in that, include: One or more processors, and memory; The memory stores computer-readable instructions that, when executed by the one or more processors, perform the steps of the biometric method as described in any one of claims 1 to 7.

Images