Fingerprint unified identity authentication platform

Abstract

The invention relates to the technical field of biological feature recognition, and discloses a fingerprint unified identity authentication platform which comprises a fingerprint registration module, a template synthesis module, a data storage module, a fingerprint comparison module and a dynamic threshold adjustment module. According to the unified fingerprint identity authentication platform, the fingerprint templates are stored in the database in a centralized manner in a hexadecimal data format through the unified fingerprint identity authentication platform, multi-system redundant storage is eliminated, the management complexity is reduced, unified encryption strategy and authority level-to-level management are supported, the risk of data leakage or tampering is prevented, and the user experience is improved. A user does not need to repeatedly register fingerprints in a plurality of systems, the operation process is simplified, the fingerprints are loaded to the memory space of the server in a binary data form based on multithreading, concurrent loading and comparison operation, and the fingerprint data are processed in parallel, so that the recognition efficiency and accuracy are improved.

Description

Technical Field

[0001] The present application relates to the field of biometric identification technology, and specifically to a fingerprint unified identity authentication platform. Background Art

[0002] In the field of biometric recognition, fingerprint recognition technology is widely used in identity authentication, security systems, financial payments and other scenarios due to its practicality and reliability.

[0003] The traditional method requires storing user fingerprint templates in the database user table of the corresponding business system, which depends on the database storage of each business system, resulting in multiple redundant storage of the same user's fingerprint data in multiple business systems. This leads to high management complexity (multiple independent fingerprint databases need to be maintained), high security risks (distributed storage is prone to data leakage or tampering risks) and poor user experience (users need to repeatedly register fingerprints in different systems). In addition, the fingerprint database needs to be deployed independently, and each application system needs to uniformly call the interface of the fingerprint identity authentication platform.

[0004] Therefore, there is an urgent need for a unified fingerprint identity authentication platform to achieve centralized storage and standardized access of fingerprint data. Summary of the Invention

[0005] The purpose of this application is to provide a fingerprint unified identity authentication platform to solve the technical problems raised in the above background technology.

[0006] To achieve the above objectives, this application discloses the following technical solutions: a fingerprint unified identity authentication platform, comprising:

[0007] The fingerprint registration module is configured to generate three fingerprint data respectively when receiving three presses of the user to collect the fingerprint image;

[0008] Template synthesis module: configured to perform template synthesis on the three fingerprint data through the template synthesis interface of the fingerprint algorithm, and if the synthesis is successful, enter the data saving step, otherwise prompt the user to re-register;

[0009] Data saving module: saves the successfully synthesized fingerprint template to the database in hexadecimal data format;

[0010] A fingerprint comparison module configured to include a first comparison and recognition mode and a second comparison and recognition mode, wherein in the first comparison and recognition mode, the module is configured to receive a user name input by a user and a real-time fingerprint pressed by the user, and in the second comparison and recognition mode, the module is configured to receive a real-time fingerprint pressed by the user;

[0011] Dynamic threshold adjustment module: configured to dynamically adjust the comparison threshold based on the real-time false positive rate and false rejection rate, maintaining a false rejection rate of ≤0.05%;

[0012] The fingerprint algorithm includes: a dynamic topology fusion mechanism that integrates local detail features and global topology features of fingerprints, a mixed precision quantization model that uses FP16-INT8 mixed quantization to store feature vectors, and a heterogeneous computing accelerator through CPU / GPU collaborative computing.

[0013] Preferably, the first comparison mode includes:

[0014] Receive the user name and real-time fingerprint input by the user, extract fingerprint features through the feature extraction interface, and return to re-collect if the extraction fails; after successful extraction, compare the features with the fingerprint template of the corresponding user in the database, and return the user ID, finger position and comparison score if the comparison is successful; otherwise, return failure and prompt to re-collect;

[0015] The second comparison mode includes:

[0016] After receiving the real-time fingerprint pressed by the user, the feature is extracted through the feature extraction interface. If the extraction fails, it returns to re-collect. After successful extraction, the feature is compared with all fingerprint templates in the database one by one. If there is a match, the recognition success result is returned, otherwise it prompts to re-collect.

[0017] Preferably, the dynamic topology fusion mechanism specifically includes:

[0018] D1: Perform multi-scale feature extraction on the three fingerprint data to obtain local detail feature vectors and global topology feature vectors respectively;

[0019] D2: The local detail feature vector and the global topology feature vector are fused by a weighted fusion algorithm to generate a fused multi-scale feature vector, wherein the weighted fusion algorithm dynamically adjusts the weight coefficient based on the fingerprint image quality.

[0020] Preferably, the mixed precision quantization model specifically includes:

[0021] Store local feature vectors in FP16 format and global feature vectors in INT8 format;

[0022] The weights of local eigenvectors and global eigenvectors are dynamically balanced through the mixing coefficient.

[0023] Preferably, the mixing coefficient is determined by the following formula:

[0024] Among them, α0 is the initial mixing coefficient, ε is the current feature error, and ε thresholdis the preset error threshold, and FRR is the true rejection rate calculated in real time.

[0025] Preferably, the dynamic adjustment of the comparison threshold according to the real-time false recognition rate and false rejection rate specifically includes:

[0026] Based on the deviation between the real-time error recognition rate and the target error recognition rate, the comparison threshold is dynamically adjusted through an adaptive threshold attenuation mechanism;

[0027] When the false rejection rate is close to the preset false rejection rate threshold, the threshold is increased by the compensation coefficient.

[0028] Preferably, the comparison threshold is adjusted by the following formula:

[0029] θ=θ0+λ×(FAR-FAR target )

[0030] Among them, θ is the comparison threshold after real-time adjustment, θ0 is the initial threshold, FAR is the real-time false recognition rate, λ is the attenuation coefficient, λ = 0.1 + 0.2 × EWMA (FAR), where EWMA (FAR) is the exponentially weighted moving average of the real-time false recognition rate; FAR target is the target false recognition rate, FRR is the real-time false rejection rate, and when FRR ≥ 0.05%, the following is enforced: Among them, β compensation coefficient, θ prev is the comparison threshold of the previous moment, FAR threshold is the preset false rejection rate threshold.

[0031] Preferably, the heterogeneous computing accelerator specifically includes:

[0032] Y1: Divide the fingerprint matching task into subtasks that are processed collaboratively by the CPU and GPU;

[0033] Y2: Balance the CPU and GPU loads based on heterogeneous computing acceleration algorithms to reduce task duration.

[0034] Preferably, the heterogeneous computing acceleration algorithm is used to balance the CPU and GPU loads to reduce task time consumption, specifically comprising the following steps:

[0035] P1 - Resource Allocation: Select hardware based on the task type. For example, the Manhattan distance calculation of the 12-dimensional feature vector is assigned to the GPU, while the CPU is responsible for task scheduling and result verification.

[0036] P2-Balancing load: Calculate the CPU execution time T CPU When the CPU execution time is too high, increase the number of CPU cores to reduce the CPU execution time. N is the size of the fingerprint database, H is the number of CPU cores, wd is the memory bandwidth, and Freq is the CPU frequency;

[0037] Calculate GPU execution time T GPU When the GPU execution time is too high, increase the number of GPU cores, optimize SIMD instruction width, or optimize data locality to reduce the GPU execution time. M is the feature vector dimension, SIMD width is the SIMD instruction width, GPU cores is the number of GPU cores, FLOPS is the floating-point computing capability of GPU, and Locality is the data locality coefficient;

[0038] P3-communication optimization: calculate the total task time T, T = max(T CPU ,T GPU )+τ overlap When the total time of real-time computing tasks is too high, the total time of tasks can be reduced by optimizing the communication between CPU and GPU to reduce the overlapping execution time of tasks, τ overlap T is the task overlapping execution time; CPU T is the CPU execution time; GPU The GPU execution time.

[0039] Preferably, when executing data saving, the data saving module includes the following steps:

[0040] C1: Arrange the data to be written into the buffer;

[0041] C2: Copy the buffer data to a preset memory target address at one time.

[0042] Beneficial effects: The fingerprint unified identity authentication platform of the present application stores fingerprint templates in a hexadecimal data format in a database through the fingerprint unified identity authentication platform, eliminating redundant storage of multiple systems and reducing management complexity. In addition, it supports unified encryption strategies and hierarchical authority management to prevent data leakage or tampering risks. Users do not need to repeatedly register fingerprints in multiple systems, simplifying the operation process. Secondly, it adopts multi-threading, concurrent loading and comparison operations, loads the data into the server memory space in the form of binary data, and processes the fingerprint data in parallel. From feature extraction to comparison and matching, each link runs at high speed under the drive of multi-threading. In this way, in the fingerprint feature extraction stage, different threads analyze different areas of the fingerprint image at the same time, greatly shortening the extraction time; and in the comparison process, multi-threading can simultaneously match the fingerprint to be compared with multiple groups of fingerprints in the database, quickly screening out potential matching objects, and the accuracy far exceeds the traditional single-threaded mode of writing to files, and the data reading and writing processing speed is fast and convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A structural block diagram of the fingerprint unified identity authentication platform provided in the embodiment of the present application;

[0045] Figure 2 The figure is a flowchart of an authentication method implemented by the fingerprint unified identity authentication platform based on this embodiment. DETAILED DESCRIPTION

[0046] The following is a clear and complete description of the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] In this document, the term "comprising" is intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0048] This embodiment provides a Figure 1 The fingerprint unified identity authentication platform shown includes:

[0049] Fingerprint registration module: configured to generate three fingerprint data respectively when receiving three presses of the user to collect the fingerprint image;

[0050] Template synthesis module: configured to perform template synthesis on the three fingerprint data through the template synthesis interface of the fingerprint algorithm, and if the synthesis is successful, enter the data saving step, otherwise prompt the user to re-register;

[0051] Data saving module: saves the successfully synthesized fingerprint template in a hexadecimal data format to the database; it is feasible that the data saving module includes the following steps when executing data saving: C1: arranging the data to be written into the buffer; C2: copying the buffer data to a pre-set memory target address at one time;

[0052] The fingerprint comparison module is configured to include a first comparison and recognition mode and a second comparison and recognition mode; wherein in the first comparison and recognition mode, it is configured to receive the user name entered by the user and the real-time fingerprint pressed by the user, and in the second comparison and recognition mode, it is configured to receive the real-time fingerprint pressed by the user. That is, the first comparison mode (1:1 comparison) is: the user enters the user name and presses the fingerprint, and the fingerprint features are extracted through the feature extraction interface. If the extraction fails, it returns to re-collection; after successful extraction, the features are compared with the fingerprint template of the corresponding user in the database, and the user ID, finger position and comparison score are returned if the comparison is successful, otherwise it returns failure and prompts to re-collect; the second comparison mode (1:N comparison) is that the user directly presses the fingerprint, and the features are extracted through the feature extraction interface. If the extraction fails, it returns to re-collection; after successful extraction, the features are compared with all fingerprint templates in the database one by one. If there is a match, the recognition success result is returned, otherwise it prompts to re-collect;

[0053] Dynamic threshold adjustment module: Dynamically adjusts the comparison threshold based on the real-time false acceptance rate (FAR) and false rejection rate (FRR), reducing the false acceptance rate by an order of magnitude while maintaining the false rejection rate ≤ 0.05%;

[0054] The fingerprint algorithm includes: a dynamic topological fusion mechanism that integrates local detail features (such as minutiae) of fingerprints with global topological features (such as ridge direction and bifurcation point distribution), a mixed precision quantization model that uses FP16-INT8 mixed quantization to store feature vectors, and a heterogeneous computing accelerator through CPU / GPU collaborative computing.

[0055] In this embodiment, the dynamic topology fusion mechanism specifically includes:

[0056] D1: Perform multi-scale feature extraction on the three fingerprint data to obtain the local detail feature vectors F local and the global topological eigenvector F global ;

[0057] D2: The local detail feature vector and the global topology feature vector are fused using a weighted fusion algorithm to generate a fused multi-scale feature vector. The weighted fusion algorithm dynamically adjusts the weight coefficient based on the fingerprint image quality. It should be noted that the weighted fusion algorithm can be any one of the existing technologies. An example of a feasible algorithm is as follows: The expression of the weighted fusion algorithm is:

[0058] W=f(Q,C)×(F local +F global )

[0059] Where W is the fused multi-scale feature vector (128-byte aligned), f(Q,C) is the weight coefficient function, f(Q,C) = Q×(1+0.3×log2(SNR)×C), SNR is the fingerprint image signal-to-noise ratio (dB); Q is the fingerprint image quality score (0≤Q≤1), calculated based on the signal-to-noise ratio (SNR) and clarity. The higher the quality of the image, the higher the weight is, and high-quality features are retained first. C is the local feature confidence (the value range of C is 0≤C≤1), and the reliability of the local feature is measured by the Minutiae positioning accuracy. The weight of the low-confidence area (such as the fingerprint folds) is reduced to avoid noise interference. local is the local detail feature vector (FP16 format, including Minutiae coordinates and direction angles); F global is the global topological feature vector (INT8 format, including ridge density and bifurcation point distribution). In the formula, log2(SNR) performs logarithmic transformation on SNR to reduce the sensitivity of extreme high / low signal-to-noise ratio to weights and make weight distribution smoother.

[0060] The design of the dynamic topology fusion mechanism described above improves the robustness of low-quality fingerprints. In low-quality fingerprint scenarios (such as wet and dry fingers), the recognition rate is improved by dynamically suppressing the weight of noisy areas (approximately 15% higher than the fixed weight method). Secondly, by combining global topological feature vectors with local detail feature vectors, the local details (such as bifurcation points) and global structure (such as ridge direction) of the fingerprint are retained, reducing the false recognition rate.

[0061] In this embodiment, the mixed precision quantization model specifically includes:

[0062] Storing local feature vectors in FP16 format and global feature vectors in INT8 format. This hybrid storage method reduces storage space while maintaining recognition accuracy.

[0063] The weights of local eigenvectors and global eigenvectors are dynamically balanced through the mixing coefficient.

[0064] Furthermore, the mixing coefficient is determined by the following formula:

[0065] Among them, α0 is the initial mixing coefficient, α0 = 0.8; ε is the current feature error, ε threshold is the preset error threshold (e.g. 0.01), and FRR is the real-time calculated false rejection rate, obtained by comparing the results of the sliding window statistics. Indicates that when the feature error exceeds the threshold ε thresholdWhen the error is high, the weight of α is dynamically increased, prioritizing the retention of feature layers with larger errors to avoid precision loss due to over-quantization. Secondly, α is adjusted based on the real-time calculated false rejection rate. When the FRR increases, the weight of global features is reduced, focusing on local detail features to improve the recognition pass rate, allowing the system to maintain stable performance even in low-quality fingerprint scenarios.

[0066] In this embodiment, the dynamic adjustment of the comparison threshold according to the real-time false recognition rate and false rejection rate specifically includes:

[0067] Based on the deviation between the real-time error recognition rate and the target error recognition rate, the comparison threshold is dynamically adjusted through an adaptive threshold attenuation mechanism;

[0068] When the false rejection rate approaches the preset false rejection rate threshold, the threshold is increased by the compensation coefficient to maintain system stability.

[0069] Furthermore, the comparison threshold is adjusted by the following formula:

[0070] θ=θ0+λ×(FAR-FAR target )

[0071] Wherein, θ is the comparison threshold after real-time adjustment, θ0 is the initial threshold (default is 0.7); FAR is the real-time false recognition rate, λ is the attenuation coefficient (0.1≤α≤0.3); λ=0.1+0.2×EWMA(FAR), where EWMA(FAR) is the exponentially weighted moving average of the real-time false recognition rate. The λ coefficient can limit the adjustment range to avoid system oscillation caused by sudden changes in the threshold. For example, if λ=0.1, the threshold adjustment range will not exceed 10% of the current deviation each time; FAR target is the target false recognition rate (e.g. 0.01%); FRR is the real-time false rejection rate. When FRR ≥ 0.05%, the following is enforced: Among them, β compensation coefficient, β = 0.05; θ prev is the comparison threshold of the previous moment, FAR threshold is the preset false rejection rate threshold. In the formula, by calculating the deviation between the current FAR and the target value, the threshold θ is dynamically adjusted to make FAR towards FAR target Convergence. For example, if the current FAR is higher than the target value, the threshold θ is increased to reduce the false recognition rate; otherwise, the threshold θ is decreased to reduce the false rejection rate.

[0072] The aforementioned design for dynamic threshold adjustment reduces FAR fluctuations, maintaining a consistent FRR below 0.05%, improving user authentication pass rates. Combined with an adaptive threshold attenuation mechanism, this reduces threshold adjustment latency, improving fingerprint matching request efficiency, and, in low-quality fingerprints (SNR < 20dB), reduces false positive rates with minimal increases in false rejection rates. For example, in financial payment applications, strict control of FAR < 0.01% and FRR < 0.05% ensures transaction security. In access control systems, compensation mechanisms reduce false rejection rates in low-light or humid environments, enhancing user experience. In mobile device biometrics applications, real-time dynamic adjustment is supported to adapt to varying fingerprint quality across users. Consequently, precise control and robustness of dynamic threshold adjustment in fingerprint authentication systems are achieved, striking a balance between security and user experience.

[0073] In this embodiment, the heterogeneous computing accelerator specifically includes:

[0074] Y1: Divide the fingerprint matching task into subtasks that are processed collaboratively by the CPU and GPU;

[0075] Y2: Balance the CPU and GPU loads based on heterogeneous computing acceleration algorithms to reduce task duration.

[0076] Furthermore, the heterogeneous computing acceleration algorithm is used to balance the CPU and GPU loads to reduce task time consumption, specifically including the following steps:

[0077] P1 - Resource Allocation: Hardware is selected based on the task type. For example, the Manhattan distance calculation of 12-dimensional feature vectors is assigned to the GPU, while the CPU is responsible for task scheduling and result verification. The hardware resource requirements are as follows: the GPU is dominated by Manhattan distance calculations and requires high parallelism; the CPU is dominated by logic verification and requires low-latency memory access.

[0078] P2-Balancing load: Calculate the CPU execution time T CPU When the CPU execution time is too high, increasing the number of CPU cores can reduce the CPU execution time. However, it is necessary to ensure that the allocated CPU resources do not exceed the available number. N is the size of the fingerprint database, H is the number of CPU cores, wd is the memory bandwidth (GB / s), and Freq is the CPU frequency (GHz);

[0079] Calculate GPU execution time T GPU When the GPU execution time is too high, increase the number of GPU cores, optimize SIMD instruction width, or optimize data locality (improve locality) to reduce the GPU execution time. However, it is necessary to ensure that the allocated GPU resources do not exceed the available number. M is the feature vector dimension, SIMD width SIMD instruction width (such as 512 bits for AVX-512), GPU cores is the number of GPU cores, FLOPS is the GPU floating-point computing power, and Locality is the data locality coefficient. 0.8≤Locality≤1.0 is based on the memory access efficiency after optimization by Morton coding. Morton coding is an existing technology that is used to optimize memory access patterns through spatial locality. The Locality coefficient improves the GPU cache hit rate and reduces memory latency.

[0080] P3-communication optimization: calculate the total task time T, T = max(T CPU ,T GPU )+τ overlap When the total time of real-time computing tasks is too high, the total time of tasks can be reduced by optimizing the communication between CPU and GPU (such as asynchronous transmission and pipeline technology) to reduce the overlapping execution time of tasks, τ overlap The CPU is responsible for task scheduling, and the GPU performs computationally intensive tasks (such as Manhattan distance), thus reducing the total execution time by overlapping execution. CPU T is the CPU execution time; GPU The GPU execution time.

[0081] Based on the design of the above heterogeneous computing accelerator, a quantitative basis is provided (such as the calculation of the total task time) to guide the optimization direction of resource allocation. The resource allocation is coordinated through hardware, software, and algorithms to minimize the total task time, ultimately achieving efficient resource allocation, improving the throughput in the 1:N comparison recognition mode, and optimizing T CPU With T GPU The balance is achieved to achieve a comparison throughput of 1.2 million times per second, and the energy efficiency is improved by using GPU parallel design.

[0082] In summary, the fingerprint unified identity authentication platform of this embodiment centrally stores fingerprint templates in a hexadecimal data format in a database, eliminating redundant storage across multiple systems and reducing management complexity. Furthermore, it supports unified encryption policies and hierarchical permission management to mitigate the risk of data leakage or tampering. Users no longer need to repeatedly register fingerprints across multiple systems, streamlining the operational process. Secondly, it employs multi-threaded, concurrent loading and matching operations, loading binary data into server memory and processing fingerprint data in parallel. From feature extraction to matching and matching, each step operates at high speed driven by multi-threading. Thus, during the fingerprint feature extraction phase, different threads simultaneously analyze different regions of the fingerprint image, significantly reducing extraction time. During the matching process, multi-threading can simultaneously match the fingerprint to be compared with multiple sets of fingerprints in the database, rapidly screening potential matches with far greater accuracy than traditional single-threaded methods for writing to files. This makes data reading and writing fast and convenient.

[0083] In one possible implementation, Figure 2 As shown, the fingerprint unified identity authentication method based on the fingerprint unified identity authentication platform of this embodiment specifically includes the following steps:

[0084] A. Fingerprint registration: The user collects the fingerprint by pressing it three times for fingerprint template synthesis;

[0085] B. Fingerprint template synthesis: The three collected fingerprint data are synthesized into templates through the fingerprint algorithm's fingerprint template synthesis interface. If the synthesis is successful, the data will be saved. If the synthesis fails, a re-registration prompt will be prompted.

[0086] C. Save fingerprint template data: save the successfully synthesized fingerprint template to the database in hexadecimal data format;

[0087] D. Fingerprint 1:1 matching: In the 1:1 matching application scenario, the user enters the username and presses to collect the fingerprint for fingerprint feature extraction;

[0088] E. Fingerprint feature extraction: Use the fingerprint feature extraction interface of the fingerprint algorithm to perform fingerprint comparison on the successfully extracted features. If the extraction fails, return to step D and repeat the process.

[0089] F. Fingerprint comparison: The system compares the extracted features with the fingerprint template of the corresponding user in the system. If the comparison is successful, a successful result will be returned. If the comparison fails, it is necessary to return to step D and repeat the process.

[0090] G. Return the result of successful comparison: If the comparison is successful, the user ID, user location, and comparison score of the current successful comparison will be returned;

[0091] H. Fingerprint 1:N recognition: In 1:N recognition application scenarios, users do not need to enter a username and can directly press the fingerprint to collect fingerprint features for extraction;

[0092] I. Fingerprint feature extraction: Use the fingerprint algorithm's fingerprint feature extraction interface to perform fingerprint recognition on the successfully extracted features. If the extraction fails, return to step H and repeat the process.

[0093] J. Fingerprint recognition: The system compares the extracted features with all fingerprint templates in the system one by one. After each comparison is completed, the system returns the comparison results. If there is a successful fingerprint, the recognition is successful. Otherwise, the comparison fails and the system needs to return to step H and repeat the process.

[0094] K. Return the result of successful recognition: If the recognition is successful, the user ID, user location and comparison score of the current successful recognition will be returned.

[0095] In the embodiments provided herein, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication media include any medium that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a computer can access. The computer-readable storage medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0096] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A fingerprint unified identity authentication platform, characterized by: include: The fingerprint registration module is configured to generate three fingerprint data respectively when receiving three presses of the user to collect the fingerprint image; Template synthesis module: configured to perform template synthesis on the three fingerprint data through the template synthesis interface of the fingerprint algorithm, and if the synthesis is successful, enter the data saving step, otherwise prompt the user to re-register; Data saving module: saves the successfully synthesized fingerprint template to the database in hexadecimal data format; A fingerprint comparison module configured to include a first comparison and recognition mode and a second comparison and recognition mode, wherein in the first comparison and recognition mode, the module is configured to receive a user name input by a user and a real-time fingerprint pressed by the user, and in the second comparison and recognition mode, the module is configured to receive a real-time fingerprint pressed by the user; Dynamic threshold adjustment module: configured to dynamically adjust the comparison threshold based on the real-time false positive rate and false rejection rate, maintaining a false rejection rate of ≤0.05%; The fingerprint algorithm includes: a dynamic topology fusion mechanism that integrates local detail features and global topology features of fingerprints, a mixed precision quantization model that uses FP16-INT8 mixed quantization to store feature vectors, and a heterogeneous computing accelerator through CPU / GPU collaborative computing.

2. The fingerprint unified identity authentication platform according to claim 1, It is characterized in that in, The first comparison mode includes: Receive the user name and real-time fingerprint input by the user, extract fingerprint features through the feature extraction interface, and return to re-collect if the extraction fails; after successful extraction, compare the features with the fingerprint template of the corresponding user in the database, and return the user ID, finger position and comparison score if the comparison is successful; otherwise, return failure and prompt to re-collect; The second comparison mode includes: After receiving the real-time fingerprint pressed by the user, the feature is extracted through the feature extraction interface. If the extraction fails, it returns to re-collect. After successful extraction, the feature is compared with all fingerprint templates in the database one by one. If there is a match, the recognition success result is returned, otherwise it prompts to re-collect.

3. The fingerprint unified identity authentication platform according to claim 1, characterized in that: The dynamic topology fusion mechanism specifically includes: D1: Perform multi-scale feature extraction on the three fingerprint data to obtain local detail feature vectors and global topology feature vectors respectively; D2: The local detail feature vector and the global topology feature vector are fused by a weighted fusion algorithm to generate a fused multi-scale feature vector, wherein the weighted fusion algorithm dynamically adjusts the weight coefficient based on the fingerprint image quality.

4. The fingerprint unified identity authentication platform according to claim 1, characterized in that: The mixed precision quantization model specifically includes: Store local feature vectors in FP16 format and global feature vectors in INT8 format; The weights of local eigenvectors and global eigenvectors are dynamically balanced through the mixing coefficient.

5. The fingerprint unified identity authentication platform according to claim 4, characterized in that: The mixing coefficient is determined by the following formula: Among them, α0 is the initial mixing coefficient, ε is the current feature error, and ε threshold is the preset error threshold, and FRR is the false rejection rate calculated in real time.

6. The fingerprint unified identity authentication platform according to claim 1, characterized in that: The dynamic adjustment of the comparison threshold according to the real-time false recognition rate and false rejection rate specifically includes: Based on the deviation between the real-time error recognition rate and the target error recognition rate, the comparison threshold is dynamically adjusted through an adaptive threshold attenuation mechanism; When the false rejection rate is close to the preset false rejection rate threshold, the threshold is increased by the compensation coefficient.

7. The fingerprint unified identity authentication platform according to claim 6, characterized in that: The comparison threshold is adjusted by the following formula: θ=θ0+λ×(FAR-FAR target ) Among them, θ is the comparison threshold after real-time adjustment, θ0 is the initial threshold, FAR is the real-time false recognition rate, λ is the attenuation coefficient, λ = 0.1 + 0.2 × EWMA (FAR), where EWMA (FAR) is the exponentially weighted moving average of the real-time false recognition rate; FAR target is the target false recognition rate, FRR is the real-time false rejection rate, and when FRR ≥ 0.05%, the following is enforced: Among them, β compensation coefficient, θ prev is the comparison threshold of the previous moment, FAR threshold is the preset false rejection rate threshold.

8. The fingerprint unified identity authentication platform according to claim 1, characterized in that: The heterogeneous computing accelerator specifically includes: Y1: Divide the fingerprint matching task into subtasks that are processed collaboratively by the CPU and GPU; Y2: Balance the CPU and GPU loads based on heterogeneous computing acceleration algorithms to reduce task duration.

9. The fingerprint unified identity authentication platform according to claim 8, characterized in that: The heterogeneous computing acceleration algorithm is used to balance the CPU and GPU loads to reduce task time consumption, specifically including the following steps: P1 - Resource Allocation: Select hardware based on the task type. For example, the Manhattan distance calculation of the 12-dimensional feature vector is assigned to the GPU, while the CPU is responsible for task scheduling and result verification. P2-Balancing load: Calculate the CPU execution time T CPU When the CPU execution time is too high, increase the number of CPU cores to reduce the CPU execution time. N is the size of the fingerprint database, H is the number of CPU cores, wd is the memory bandwidth, and Freq is the CPU frequency; Calculate GPU execution time T GPU When the GPU execution time is too high, increase the number of GPU cores, optimize SIMD instruction width, or optimize data locality to reduce the GPU execution time. M is the feature vector dimension, SIMD width is the SIMD instruction width, GPU cores is the number of GPU cores, FLOPS is the floating-point computing capability of GPU, and Locality is the data locality coefficient; P3-communication optimization: calculate the total task time T, T = max(T CPU ,T GPU )+τ overlap When the total time of real-time computing tasks is too high, the total time of tasks can be reduced by optimizing the communication between CPU and GPU to reduce the overlapping execution time of tasks, τ overlap T is the task overlapping execution time; CPU T is the CPU execution time; GPU The time it takes for the GPU to execute the task.

10. The fingerprint unified identity authentication platform according to claim 1, characterized in that: When executing data saving, the data saving module includes the following steps: C1: Arrange the data to be written into the buffer; C2: Copy the buffer data to a preset memory target address at one time.

Images