A method of biometric data generation, recognition and evaluation

Abstract

The application provides a method for generating, identifying and evaluating biometric data, comprising: a system constructs a virtual biometric database by collecting initial samples and generating virtual data samples using a generator; the system pre-processes and filters the virtual data samples, including denoising and abnormal sample elimination, to form an effective database; the system designs an identification unit to identify the virtual data samples based on an algorithm and output the identification results for testing the equipment to be tested; the system designs an evaluation unit to determine the data difficulty interval based on the identification results and expert knowledge, and score and rank the equipment performance; the system is integrated and deployed, including server deployment and interface design, to realize sample retrieval and repeated testing.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for generating, recognizing, and evaluating biometric data. Background Technology

[0002] Biometric identification technology plays a crucial role in security authentication and identity verification, and is widely used in scenarios such as financial payments, border management, and mobile device unlocking. Its importance lies in its ability to efficiently and accurately verify individual identities and ensure system security. However, in practical applications, biometric identification devices must cope with complex and diverse real-world scenarios, such as iris images under different lighting conditions or fingerprint collection when a finger is injured. This places extremely high demands on device performance. Existing technologies often face limitations in testing and evaluating identification devices, including limited data sources and insufficient coverage of test scenarios, making it difficult to accurately predict the device's performance in real-world environments. The limitation of existing methods lies in the lack of systematic and comprehensiveness in the generation and selection of test data. Many tests rely on limited real-world samples, making it difficult to simulate the complex and diverse changes in biometrics in reality. For example, iris recognition may experience image quality degradation due to changes in lighting or eye diseases, while fingerprint recognition may fail due to skin abrasion or dirt. Current testing methods typically design data only for a single scenario, ignoring the combined effects of diverse environmental factors. This limitation prevents test results from fully reflecting the device's real-world performance in varying scenarios, thus affecting the scientific validity and reliability of the evaluation. The core technical challenge lies in generating highly realistic virtual biometric data to cover diverse application scenarios. The generated data must not only be realistic but also simulate various anomalies that may occur in reality, such as light spot interference in iris images or skin damage in fingerprint images. Current technologies often struggle to balance data diversity and realism during data generation. For example, generated iris samples may lack the details of realistic eye texture, or fingerprint samples may not accurately reflect the effects of factors such as skin aging, humidity, and pressure changes. This inadequacy of generated data directly leads to test samples failing to fully cover the complex situations in real-world applications. Therefore, how to generate highly realistic virtual samples while ensuring coverage of various scenarios, including normal, diseased, and environmentally disturbed conditions, has become a key issue in evaluating the performance of biometric recognition devices. Existing generation technologies struggle to find a balance between diversity and realism, and the generated samples often deviate from real-world scenarios. For example, in iris recognition, virtual samples may not be able to simulate the texture changes caused by pupil constriction under strong light; in fingerprint recognition, virtual samples may ignore the impact of finger humidity and pressure on image quality. These problems prevent test data from fully reflecting the device's performance in complex environments, thus affecting the objectivity of the evaluation results. In practical applications, devices need to meet the recognition needs of users under different lighting conditions and skin conditions. For example, fingerprint recognition devices on bank ATMs may fail due to wet or worn fingers, while iris recognition devices at airport security checkpoints may produce blurred images due to strong light reflection. How to generate highly realistic virtual samples that can cover these complex scenarios, and how to build an effective database through scientific screening to comprehensively test and evaluate the performance of recognition devices, has become a critical problem that urgently needs to be solved in the field of biometric recognition. Summary of the Invention

[0003] This invention provides a method for generating, recognizing, and evaluating biometric data, mainly including: The system constructs a virtual biometric database by collecting initial samples and generating virtual data samples using a generator. The system preprocesses and filters these virtual data samples, including noise reduction and outlier removal, to form a valid database. A recognition unit is designed to identify the virtual data samples based on algorithms and output the recognition results for testing the devices under test. An evaluation unit is designed to determine the data difficulty range based on the recognition results and expert knowledge, and to score and rate the device performance. The system is integrated and deployed, including server deployment and interface design, to enable sample retrieval and repeated testing.

[0004] Furthermore, the construction of the virtual biometric database includes: collecting initial samples as the basis for generation; using a generator to generate virtual data samples for different biometric types, wherein image-type biometrics are generated through feature parameter extraction and adjustment, and speech-type biometrics are generated through multi-feature space mapping and inverse mapping; if the generated virtual data samples meet the judgment criteria, they are added to the database; otherwise, they are discarded and the generation process is repeated; the generator decomposes biometrics into multiple regions, processes the regions through subspace methods or set methods, and combines and reconstructs them to ensure the diversity and simulation degree of the virtual data samples.

[0005] Furthermore, the preprocessing and screening of virtual data samples includes: using specific denoising methods to process noise for image-type virtual data samples; if the processed virtual data samples show no abnormalities after manual or semi-automatic screening, they are retained as valid samples, otherwise they are discarded; the denoising method selects filters or transformation algorithms according to the biometric type to remove light and shadow occlusion or background interference; the screening process is repeated until the database reaches a preset size to ensure the objectivity and usability of valid samples.

[0006] Furthermore, the design recognition unit includes: retrieving virtual data samples from the database as input; performing feature extraction and classification on the virtual data samples based on deep learning algorithms combined with distance metrics or sparse representation models; outputting recognition results if the extracted features meet intra-class stability; transmitting the recognition results to the device under test for simulating real-world testing; and using a sequence model to process correlations for time-series signals to improve recognition accuracy.

[0007] Furthermore, the design evaluation unit includes: calculating performance indicators, including accuracy and false detection rate, based on the identification results; introducing expert knowledge to determine the data difficulty range; generating a score and rating if the performance indicators match the difficulty range; the rating determines the equipment's qualification and supports repeated testing to verify objectivity; and outputting a comprehensive report integrating indicators and range analysis.

[0008] Furthermore, the system integration and deployment includes: deploying the database and units on a server; designing an interface to support sample retrieval and presentation, wherein image samples are visualized and audio samples are played; if the retrieval method is random, then test samples are selected from the database; integrating a data transmission unit to output samples to the device under test; and debugging the system to adapt to a specific environment to ensure the overall process is coherent.

[0009] Furthermore, the generator is used to generate virtual data samples for different biometric types, including: for iris type, decomposing into feature regions, and restoring the image through perturbation and combination; if the restored image meets the judgment criteria, it is confirmed as qualified; for fingerprint type, arranging and combining texture elements, constrained by prior knowledge; for voice type, adjusting key parameters and then inverse mapping to construct samples; repeating the adjustment process to generate a sufficient number of virtual data samples.

[0010] Furthermore, the feature extraction and classification of virtual data samples based on deep learning algorithms combined with distance metrics or sparse representation models includes: extracting global and local features to capture details; performing inter-class distance comparison if the features overcome noise interference; performing time-frequency domain transformation and feature extraction for speech samples; classifying time-series signals using sequence models; and outputting the classification results as the basis for recognition.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for comprehensive testing and evaluation of biometric recognition devices by constructing a virtual biometric database. Addressing the issues of insufficient data diversity and difficulty in covering the complexity of real-world application scenarios in biometric recognition such as iris and fingerprint recognition, this invention innovatively integrates a complete process of virtual data generation, preprocessing and screening, recognition, and evaluation. By collecting diverse initial samples and combining them with a generator based on feature parameter extraction and multi-feature space mapping, this invention generates highly realistic virtual iris and fingerprint samples, covering various conditions including normal, pathological, and environmental interference, ensuring data diversity and authenticity. The preprocessing stage employs adaptive filtering and region segmentation to remove noise and interference, selecting high-quality samples to form an effective database. The recognition unit uses deep learning algorithms combined with distance metrics to extract features, and the evaluation unit incorporates expert knowledge to divide difficulty ranges, generating performance scores and ratings. Through system integration and deployment, this invention supports sample retrieval and repeated testing, ensuring the objectivity and reliability of test results, ultimately achieving a comprehensive evaluation of the recognition device's performance and providing reliable support for applications in high-security and complex scenarios.

[0012] The technological effect is to significantly improve the coverage and simulation of test data, and optimize the scientific nature and efficiency of performance evaluation of identification equipment. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for generating, recognizing, and evaluating biometric data according to the present invention.

[0014] Figure 2 This is the overall framework of a method for generating, recognizing, and evaluating biometric data according to the present invention.

[0015] Figure 3 This invention relates to a method for generating, recognizing, and evaluating biometric data, specifically a virtual face image generation technology process.

[0016] Figure 4 This invention relates to a virtual iris image generation technology process for a method of generating, recognizing, and evaluating biometric data.

[0017] Figure 5 This invention relates to a virtual fingerprint image generation technology process for a method of generating, recognizing, and evaluating biometric data.

[0018] Figure 6 This invention relates to a virtual voice sample generation technology process for a method of generating, recognizing, and evaluating biometric data.

[0019] Figure 7 The main flow of the biometric recognition algorithm for a biometric data generation, recognition, and evaluation method of the present invention is as follows.

[0020] Figure 8 The main components of the method for generating, recognizing, and evaluating biometric data of the present invention are a face, iris, fingerprint, and voice recognition effect evaluator. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0022] like Figure 1-8 This embodiment of a method for generating, recognizing, and evaluating biometric data may specifically include: This invention provides a method for generating, recognizing, and evaluating biometric data. This method achieves comprehensive testing and evaluation of biometric recognition devices through steps such as constructing a virtual biometric database, preprocessing and screening, designing recognition units, designing evaluation units, and system integration and deployment.

[0023] Step S1 involves constructing a virtual biometric database by collecting initial samples and generating virtual data samples using a generator. This step is fundamental to the entire method, providing ample test data sources for subsequent identification and evaluation. The construction of the virtual biometric database requires comprehensive consideration of data diversity, simulation accuracy, and coverage to ensure it can simulate various biometric changes that may be encountered in real-world application scenarios.

[0024] Step S11: Collect initial samples as the basis for generation. Collecting initial samples is the starting point for virtual data generation, requiring the collection of representative biometric data from multiple dimensions. In one embodiment, for the initial sample collection of iris features, the system acquires iris images of people of different ages, ethnicities, and health statuses using a dedicated iris acquisition device. During the acquisition process, the system records basic attribute information for each sample, including parameters such as acquisition time, lighting conditions, acquisition distance, and image resolution. This attribute information provides important reference for subsequent virtual data generation.

[0025] Specifically, the acquisition of iris samples encompassed images of normal irises, irises with minor lesions, and irises affected by external factors. Normal iris samples exhibited clear texture and a complete iris annulus region, providing a standard feature template for the generator. Iris samples with minor lesions contained localized texture variations or pigment deposition; these samples helped the generator learn the patterns of feature changes under abnormal conditions. Iris samples affected by external factors included images under different lighting conditions, partially occluded images, and images taken from different angles; these samples enhanced the generator's adaptability to environmental changes.

[0026] In one embodiment, for the initial sample acquisition of fingerprint features, the system employs multiple fingerprint acquisition technologies to obtain fingerprint images of different quality levels. The acquisition devices include optical fingerprint scanners, capacitive fingerprint scanners, and ultrasonic fingerprint scanners, each acquiring fingerprint images with different feature representations. Fingerprint images acquired by optical scanners have high contrast and can clearly display the ridge and valley structures of the fingerprint. Fingerprint images acquired by capacitive scanners are sensitive to skin moisture and can reflect changes in fingerprint features under different skin conditions. Ultrasonic scanners can penetrate the skin surface to obtain deeper fingerprint structural information.

[0027] Fingerprint sample collection also considered factors such as different finger positions, varying pressure levels, and different skin conditions. Fingerprints from different finger positions have unique texture features; thumb fingerprints typically have a larger core area and rich detail, while little finger fingerprints are relatively smaller and have higher texture density. Different pressure levels affect the clarity and integrity of the fingerprint image; light pressure may cause blurring, while heavy pressure may cause texture deformation. Different skin conditions include dry skin, moist skin, and aging skin, each of which produces fingerprint images with specific characteristics.

[0028] Step S12 involves using a generator to produce virtual data samples for different biometric types. For image-type biometrics, samples are generated through feature parameter extraction and adjustment; for speech-type biometrics, samples are generated through multi-feature space mapping and inverse mapping. The generator is a core component in the entire virtual database construction process, and its design must fully consider the unique properties and variation patterns of different biometrics.

[0029] For the generation process of biometric features from images, the generator first performs deep feature analysis on the initial samples, extracting key geometric, texture, and statistical features. In one embodiment, the feature parameters extracted from the iris image include key parameters such as iris radius, pupil radius, iris center position, texture density distribution, and color distribution. The generator establishes correlation models and variation range models among these parameters by analyzing a large number of initial samples.

[0030] The iris radius typically varies within a specific range. The generator uses statistical analysis to determine the normal radius distribution pattern and makes reasonable parameter adjustments based on this. There is a certain proportional relationship between the pupil radius and the iris radius; the generator learns this proportional relationship and maintains consistency during the generation process. Slight shifts in the iris center position can simulate the effects of different shooting angles and eye movements; the generator produces natural variations by controlling the range of change in the center position.

[0031] Texture density distribution is a crucial component of iris features. The generator analyzes texture patterns in initial samples to establish a model of texture element distribution. These texture elements include basic types such as radial textures, ring textures, and speckled textures. During the generation process, the generator adjusts parameters such as the density, direction, and intensity of these texture elements to create iris textures with natural variations. Color distribution adjustments consider the iris color characteristics of different ethnicities and age groups, enabling the generator to produce rich color variations while maintaining realism.

[0032] To address the generation process of speech biometrics, the generator employs a multi-feature space mapping and inverse mapping method. Speech signals contain rich personal feature information, including fundamental frequency features, formant features, spectral envelope features, prosodic features, etc. The generator first maps the speech signal to multiple feature spaces, each focusing on capturing specific types of speech features.

[0033] In one embodiment, the fundamental frequency feature space specifically handles the pitch variations of speech. The generator analyzes the speaker's fundamental frequency variation patterns to establish a statistical model of the fundamental frequency trajectory. This model includes key parameters such as the average value, range of variation, and rate of variation of the fundamental frequency. The formant feature space focuses on the spectral structure of speech. The generator analyzes the formant positions and intensities corresponding to different phonemes to establish a mapping relationship between phonemes and formant features.

[0034] The spectral envelope feature space processes the overall spectral shape of the speech signal, a feature closely related to the speaker's vocal tract characteristics. By learning the spectral envelope patterns of different speakers, the generator can produce speech spectra with individual characteristics. The prosodic feature space processes suprasegmental features of speech, such as rhythm, stress, and intonation, which reflect the speaker's language habits and emotional state.

[0035] The inverse mapping process involves resynthesizing the adjusted feature parameters into a complete speech signal. The generator employs advanced speech synthesis technology to ensure that the synthesized speech signal retains the original speaker characteristics while possessing natural speech quality. During the inverse mapping process, the generator needs to coordinate parameter changes across different feature spaces to avoid producing inconsistent speech effects.

[0036] Step S13: If the generated virtual data sample meets the judgment criteria, it is added to the database; otherwise, it is discarded and the generation process is repeated. Setting the judgment criteria is a crucial step in ensuring the quality of virtual data, requiring evaluation of the generated samples from multiple dimensions.

[0037] In one embodiment, the criteria for judging iris images include image sharpness, texture naturalness, geometric consistency, and color rationality. Image sharpness is evaluated by calculating the edge strength and texture contrast of the image; images with too low sharpness cannot provide sufficient feature information and are not suitable for recognition testing. Texture naturalness is judged by comparing the texture statistical characteristics of the generated image with those of the real image, including indicators such as texture direction distribution, density distribution, and intensity distribution.

[0038] Geometric consistency checks ensure that the generated iris image has a reasonable geometric structure, including the concentricity of the iris and pupil, the circularity of the iris boundary, and the radial consistency of the texture. Color consistency checks verify whether the color distribution of the generated image conforms to the color characteristics of a real iris, avoiding unnatural color combinations.

[0039] Regarding the criteria for judging fingerprint images, the system focuses on features such as ridge continuity, minutiae distribution, and texture orientation consistency. Ridge continuity ensures that the generated fingerprint has a continuous ridge structure, avoiding breaks or unnatural connections. Minutiae distribution checks whether the distribution of endpoints, bifurcation points, and other detailed features in the generated fingerprint is reasonable; these minutiae are important for fingerprint recognition. Texture orientation consistency verifies whether the directional changes in the fingerprint texture conform to the variation patterns of real fingerprints.

[0040] In step S14, the generator decomposes the biometric features into multiple regions, processes these regions using subspace or set methods, and then reconstructs them to ensure the diversity and simulation accuracy of the virtual data samples. Region decomposition is an important technique for improving generation quality and diversity; by refining the processing of biometric features, the generation process can be better controlled.

[0041] In one embodiment, region decomposition of the iris image divides the entire iris into three main parts: an inner ring region, a middle ring region, and an outer ring region. The inner ring region is close to the pupil boundary and typically contains denser radial textures. The middle ring region is the main part of the iris and contains the richest texture information and personal features. The outer ring region is close to the outer boundary of the iris, with relatively sparse texture but important boundary features.

[0042] The subspace method establishes an independent feature subspace for each region, with each subspace specifically handling the feature variations of the corresponding region. The subspace of the inner ring region mainly handles the density and direction variations of radial textures, the subspace of the middle ring region handles complex texture pattern combinations, and the subspace of the outer ring region handles boundary features and sparse textures. The generator performs independent parameter adjustments within each subspace, and then ensures a natural transition between different regions through a coordination mechanism.

[0043] Ensemble methods treat each region as an independent generation unit, constructing a complete iris image by combining the generation results of different regions. This approach allows for greater flexibility, enabling the generation of new combinations by exchanging regional features from different samples. For example, the inner ring region of one sample can be combined with the middle and outer ring regions of another sample to create entirely new combinations of iris features.

[0044] Step S2 involves the system preprocessing and filtering the virtual data samples, including noise reduction and outlier removal, to form a valid database. Preprocessing and filtering are crucial steps in ensuring database quality; a systematic processing flow improves the usability and testing effectiveness of the virtual data.

[0045] Step S21: Apply a specific denoising method to the virtual data samples of the image type to process noise. Image noise is an important factor affecting the accuracy of biometric recognition, and denoising processing requires the use of appropriate processing methods according to different types of noise.

[0046] In one embodiment, noise in iris images mainly includes acquisition noise, compression noise, and transmission noise. Acquisition noise originates from electronic noise in the image sensor and imperfections in the optical system, typically manifesting as random pixel value fluctuations. To address this type of noise, the system employs an adaptive filtering method, adjusting the filtering intensity based on local image features. In texture-rich areas, the filtering intensity is lower to preserve texture details, while in smooth areas, the filtering intensity is higher to effectively remove noise.

[0047] Compression noise originates from quantization errors in image compression algorithms and is typically more pronounced at image edges and in areas with drastic texture changes. This system employs a frequency-domain-based denoising method, analyzing the image's spectral characteristics to identify and suppress the frequency components of compression noise. This approach can remove noise while preserving image edge sharpness and texture details.

[0048] Transmission noise may include random noise and systematic distortion, which the system addresses using a multi-scale denoising method. This method decomposes the image into components of different scales and employs a suitable denoising strategy at each scale. The coarse-scale component mainly contains the image's main structural information, and noise is removed using edge-preserving filtering. The fine-scale component contains texture details, and denoising is performed using sparse representation.

[0049] For fingerprint image denoising, the system focuses on preserving the ridge structure. Noise in fingerprint images can cause ridge breaks or false connections, affecting the fingerprint's topology. The system employs a direction-adaptive filtering method, adjusting the direction and shape of the filter kernel based on the local ridge direction. A weaker filter is used along the ridge direction, while a stronger filter is used perpendicular to the ridge direction. This approach removes noise while maintaining ridge continuity and clarity.

[0050] In step S22, if the processed virtual data sample shows no abnormalities after manual or semi-automatic screening, it is retained as a valid sample; otherwise, it is discarded. The screening process is a key link in quality control, requiring the establishment of a sound evaluation system and screening criteria.

[0051] During the manual screening process, professional evaluators examine each sample individually according to preset quality standards. In one embodiment, the manual screening standards for iris images include image integrity, texture clarity, geometric correctness, and color naturalness. Image integrity requires the iris region to be fully visible without severe occlusion or loss. Texture clarity requires that the main texture features of the iris, including radial and ring-shaped textures, can be clearly identified. Geometric correctness requires that the shape, size, and positional relationship of the iris and pupil conform to physiological structure. Color naturalness requires that the color distribution of the image be reasonable without obvious color distortion.

[0052] Semi-automatic screening combines the advantages of automated detection and manual review, improving screening efficiency and consistency. The system first uses automated algorithms to perform preliminary screening of samples, identifying obviously unqualified samples. These automated screening algorithms include image quality assessment algorithms, feature consistency checking algorithms, and anomaly detection algorithms. Image quality assessment algorithms evaluate image quality by calculating metrics such as image sharpness, contrast, and signal-to-noise ratio. Feature consistency checking algorithms verify whether the features of the generated samples conform to the statistical regularities of real biological characteristics. Anomaly detection algorithms identify samples with abnormal features, such as excessive distortion or unnatural variations.

[0053] After initial automated screening, the system submits suspicious samples to human review. Human reviewers focus on boundary cases that the automated algorithm struggles to identify, such as minor quality issues or unusual feature combinations. This semi-automatic screening method ensures both accuracy and efficiency.

[0054] Step S23: The denoising method selects a filter or transformation algorithm based on the biometric type to remove light and shadow occlusion or background interference. Different types of biometrics face different interference problems, requiring targeted denoising strategies.

[0055] Occlusion is a common interfering factor in image-based biometrics, mainly manifested as uneven illumination distribution and shadow effects. In one embodiment, the system employs an illumination compensation algorithm to address the occlusion problem in iris images. This algorithm first estimates the illumination distribution model of the image, establishing a spatial distribution function of illumination intensity by analyzing the brightness variation patterns. Then, it compensates the image based on the estimated illumination model, making the illumination across the entire iris region more uniform.

[0056] During illumination compensation, the system pays special attention to maintaining the relative contrast of the iris texture. The texture features of the iris are mainly reflected in local brightness variations; excessive compensation may weaken these important features. Therefore, the algorithm adopts an adaptive compensation strategy, performing stronger compensation in areas of drastic illumination changes and weaker compensation in areas of relatively uniform illumination.

[0057] Background interference mainly originates from objects surrounding the iris, such as eyelids, eyelashes, and glasses. The system employs a region segmentation algorithm to accurately locate the iris region, excluding background interference from the processing scope. This algorithm combines edge detection, circular fitting, and texture analysis techniques to accurately identify the inner and outer boundaries of the iris. For partially occluded iris regions, the system uses texture restoration technology to infer the possible texture of the occluded area based on surrounding texture features.

[0058] To address background interference in fingerprint images, the system primarily processes background noise and artifacts that may occur during the acquisition process. Factors such as surface contamination of the fingerprint acquisition device and incomplete contact between the finger and the acquisition surface can generate background interference in the image. The system employs morphological filtering methods to remove these interferences, distinguishing the true fingerprint structure from background noise by analyzing the morphological characteristics of the fingerprint ridges.

[0059] Step S24: Repeat the screening process until the database reaches the preset size to ensure the objectivity and usability of the valid samples. Determining the database size requires comprehensive consideration of factors such as testing requirements, computing resources, and storage capacity.

[0060] In one embodiment, the preset size of the iris database is determined based on the anticipated testing scenarios. If primarily used for basic functional testing, the database may contain thousands of samples, covering major iris feature variations. If used for performance limit testing, the database may require tens of thousands of samples, including various boundary cases and challenging samples. The database size also needs to consider the balance of different sample categories to ensure sufficient representativeness for various feature types.

[0061] The objectivity of a valid sample is reflected in its unbiasedness and representativeness. The system uses statistical analysis to ensure that the generated samples are consistent with the real samples in terms of feature distribution, avoiding systematic bias. Representativeness requires that the samples can cover various situations that may be encountered in real applications, including different population characteristics, different collection conditions, and different quality levels.

[0062] Usability refers to the ability of the samples to effectively support the testing requirements of the recognition equipment. The system verifies the usability of the samples through pre-testing by inputting the generated samples into the reference recognition system and observing whether the recognition results meet expectations. If the recognition results of a certain type of sample are abnormal, it indicates that the sample may have quality issues and requires further screening or regeneration.

[0063] Step S3: The system design identification unit identifies virtual data samples based on algorithms and outputs the identification results for testing the device under test. The identification unit is the core component of the entire testing system, and its design needs to fully consider the identification characteristics and testing requirements of different biometrics.

[0064] Step S31: Retrieve virtual data samples from the database as input. The sample retrieval strategy directly affects the effectiveness and comprehensiveness of the test, and a suitable retrieval plan needs to be formulated based on the test objectives.

[0065] In one embodiment, the system supports multiple sample retrieval modes, including random retrieval, stratified retrieval, and targeted retrieval. Random retrieval selects samples randomly from the database, suitable for general performance testing, and provides unbiased test results. Stratified retrieval performs stratified sampling based on sample characteristics, ensuring that different types of samples are adequately tested. Targeted retrieval selects appropriate samples for specific testing needs, such as selecting high-difficulty samples for extreme performance testing.

[0066] During sample retrieval, the system records information such as retrieval time, retrieval purpose, and test results for each sample, forming a complete test log. This log information provides important basis for subsequent result analysis and system optimization. The system also supports repeated sample retrieval to verify the consistency and stability of test results.

[0067] Step S32 involves extracting and classifying features from virtual data samples using deep learning algorithms combined with distance metrics or sparse representation models. Feature extraction and classification are the core steps in the recognition process, and the choice and design of the algorithm directly affect the accuracy and robustness of the recognition.

[0068] Deep learning algorithms have demonstrated powerful feature learning capabilities in biometric recognition, automatically learning the mapping relationship from raw data to high-level semantic features. In one embodiment, the deep learning algorithm for iris recognition employs a convolutional neural network architecture, progressively extracting hierarchical features of the iris through multiple layers of convolution and pooling operations.

[0069] The shallow layers of the network primarily extract low-level features, such as basic visual elements like edges and texture orientation. These features correspond to the basic structural information in the iris image, laying the foundation for subsequent high-level feature extraction. The middle layers of the network combine low-level features to form more complex patterns, such as specific texture combinations and local structures. These intermediate features begin to reflect the individual differences in the iris. The deeper layers of the network further abstract these intermediate features, forming highly semantic feature representations with strong discriminative and generalization abilities.

[0070] Distance metrics are used to measure the similarity between different samples and are an important basis for identification decisions. The system employs multiple distance metrics, including Euclidean distance, cosine distance, and Mahalanobis distance. Euclidean distance is suitable for direct comparisons in feature space and is simple to calculate, but may be affected by feature dimension and scale. Cosine distance focuses on the directional similarity of feature vectors and is insensitive to changes in feature amplitude, making it suitable for handling the influence of factors such as illumination variations. Mahalanobis distance considers the correlation between features and can more accurately reflect the true differences between samples.

[0071] Step S33: If the extracted features meet the intra-class stability requirement, output the recognition result. Intra-class stability is an important indicator for evaluating feature quality, reflecting the degree of clustering of different samples of the same individual in the feature space.

[0072] In one embodiment, intra-class stability is assessed by calculating the variance of feature vectors for samples within the same class. The system first calculates the mean of all feature vectors within each class, and then calculates the deviation of each sample feature vector from the mean. If the deviation is within a preset threshold range, the feature exhibits good intra-class stability. For iris features, intra-class stability is primarily reflected in the fact that iris images of the same person acquired at different times should produce similar feature vectors.

[0073] Intra-class stability of features also needs to consider the impact of changes in acquisition conditions on feature stability. Iris images of the same person acquired under different lighting conditions, shooting angles, and time periods may show some differences, but these differences should be smaller than the differences between different individuals. The system establishes a stability assessment model to quantify the degree of influence of these changing factors on feature stability.

[0074] Once the features pass the stability test, the system outputs a recognition result containing confidence information. The confidence level reflects the reliability of the recognition result and is calculated based on the similarity score of feature matching and historical recognition statistics. A high confidence level indicates a high degree of feature matching and reliable results, while a low confidence level may require further verification or the use of other recognition methods.

[0075] Step S34: The identification result is transmitted to the device under test for simulating real-world testing. The transmission of the identification result needs to consider multiple aspects such as data format, transmission protocol, and security to ensure the accuracy and reliability of the testing process.

[0076] The data format of the recognition results needs to be consistent with the interface specifications of the device under test. In one embodiment, the iris recognition results include information such as sample identifier, feature vector, matching score, and confidence level. The sample identifier is used to track the source and attributes of the test sample, the feature vector provides detailed feature information for the device under test to compare, the matching score reflects the degree of similarity with samples in the database, and the confidence level indicates the reliability of the result.

[0077] The selection of a transmission protocol needs to consider factors such as data volume, transmission speed, and reliability. For large-volume image data transmission, the system employs efficient compression algorithms to reduce transmission time. For test scenarios with high real-time requirements, the system uses low-latency transmission protocols to ensure timely response. For applications with high reliability requirements, the system uses transmission protocols with error detection and correction capabilities.

[0078] Security considerations primarily involve data integrity and confidentiality. The system employs digital signature technology to ensure the integrity of transmitted data and prevent tampering during transmission. For test data containing sensitive information, the system uses encryption technology to protect data confidentiality and prevent unauthorized access.

[0079] Step S35: The algorithm uses a sequence model to process the correlation of time-series signals to improve recognition accuracy. Processing time-series signals requires full consideration of their temporal correlation and dynamic characteristics; sequence models can effectively capture these characteristics.

[0080] In one embodiment, the temporal signal processing in voice biometrics employs a recurrent neural network model. This model, through the recurrent connections of hidden states, is able to memorize and utilize historical information, making it suitable for processing time-dependent voice signals. At each time step, the network receives the current voice feature input, combines it with the hidden state from the previous time step, and generates the output for the current time step and an updated hidden state.

[0081] The temporal correlation of speech signals is reflected in the continuity of features between adjacent time frames and the contextual relationships between speech units. Features in adjacent time frames typically exhibit high similarity, and this continuity provides important constraints for recognition. Speech units, such as phonemes and syllables, have complex contextual relationships; preceding speech units influence the articulation features of subsequent speech units.

[0082] Sequence models further enhance their focus on key time periods through attention mechanisms. These mechanisms automatically identify the time periods most important to the recognition task and assign them higher weights. In speech recognition, certain phonemes or syllables may contain more personal feature information; attention mechanisms can automatically identify and prioritize these key components.

[0083] Long Short-Term Memory (LSTM) networks, as an important form of sequence models, solve the gradient vanishing problem of traditional recurrent networks through a gating mechanism. This network comprises three gating units: an input gate, a forget gate, and an output gate, enabling selective memorization and forgetting of information. The input gate controls the degree of new information input, the forget gate controls the degree of retention of historical information, and the output gate controls the degree of output of current state information.

[0084] In step S4, the system design evaluation unit determines the data difficulty range based on the identification results and expert knowledge, and scores and rates the equipment performance. The evaluation unit is a crucial component of the testing system, responsible for objectively and comprehensively evaluating the performance of the equipment under test.

[0085] Step S41: Calculate performance indicators based on the recognition results, including accuracy and false detection rate. The calculation of performance indicators requires a comprehensive evaluation system to ensure the objectivity and comparability of the evaluation results.

[0086] Accuracy is a fundamental metric for measuring the ability of an identification system to correctly identify individuals. It is calculated by dividing the number of correctly identified samples by the total number of test samples. In one embodiment, the accuracy calculation of an iris recognition system needs to differentiate between different types of identification tasks. For authentication tasks, accuracy reflects the system's ability to correctly accept legitimate users and correctly reject illegitimate users. For identity recognition tasks, accuracy reflects the system's ability to correctly select the target identity from the candidate set.

[0087] Accuracy calculations also need to consider the difficulty distribution of the test samples. High accuracy on easy samples may not fully reflect the system's true performance, while accuracy on difficult samples better reflects the system's robustness. The system provides a more detailed performance analysis by calculating accuracy separately for samples of different difficulty levels.

[0088] False positive rate includes two aspects: false acceptance rate and false rejection rate. False acceptance rate reflects the probability that the system incorrectly accepts an unauthorized user, which may lead to security risks. False rejection rate reflects the probability that the system incorrectly rejects a legitimate user, which affects user experience. In practical applications, a balance needs to be found between these two types of errors based on the security and convenience requirements of the application scenario.

[0089] Performance metrics calculations also include response time, throughput, and resource consumption. Response time reflects the time required for the system to process a single recognition request, which is crucial for real-time applications. Throughput reflects the number of recognition requests the system can process per unit of time, demonstrating the system's processing capacity. Resource consumption includes the usage of computing resources, storage resources, network resources, etc., affecting the system's deployment costs and operational efficiency.

[0090] Step S42 involves introducing expert knowledge to determine the data difficulty range. Introducing expert knowledge provides an experience- and theory-based difficulty assessment, making the test more scientific and comprehensive.

[0091] Expert knowledge was acquired through multiple channels, including literature review, expert interviews, and experimental analysis. Literature review involved collecting analyses and conclusions from published research on the difficulty of biometric identification. Expert interviews invited field experts to share the types and characteristics of challenging samples encountered in practical applications. Experimental analysis involved conducting numerous identification experiments to statistically analyze the distribution of identification difficulty for different sample types.

[0092] In one embodiment, the difficulty range for iris recognition is divided considering factors such as image quality, texture complexity, and acquisition conditions. Image quality includes basic indicators such as resolution, sharpness, and contrast; high-quality images typically have lower recognition difficulty. Texture complexity reflects the richness and uniqueness of the iris texture; complex textures provide more recognition information but also increase processing difficulty. Acquisition conditions include environmental factors such as lighting, angle, and distance; ideal acquisition conditions reduce recognition difficulty.

[0093] Expert knowledge also includes experience in handling unusual situations. Some individuals may possess unique biometric characteristics, such as iris abnormalities or worn fingerprints, which are less frequently encountered in standard testing but may arise in real-world applications. Expert experience can help identify these special cases and develop appropriate testing strategies.

[0094] The difficulty level is divided using a multi-level classification method, categorizing test samples into easy, medium, hard, and very hard levels. Each level corresponds to a specific range of sample features and expected recognition performance. This tiered approach facilitates the development of appropriate testing plans for different application needs.

[0095] Step S43: If the performance indicators match the difficulty range, a score and rating are generated. The scoring and rating system needs to establish scientific evaluation criteria to ensure the fairness and credibility of the evaluation results.

[0096] Matching performance metrics with difficulty ranges is achieved by establishing a mapping relationship. In one embodiment, the system establishes a two-dimensional evaluation matrix of performance metrics and difficulty ranges. The rows of the matrix correspond to different difficulty ranges, and the columns correspond to different performance metric levels. Each element in the matrix represents a score for achieving a specific performance level within a specific difficulty range.

[0097] The scoring system employs either a percentage or a grade system to provide intuitive performance evaluation results. A percentage score offers precise numerical assessments, facilitating quantitative comparisons. A grade system categorizes performance into excellent, good, satisfactory, and unsatisfactory levels, making it easy to understand and apply. The scoring calculation comprehensively considers multiple performance indicators, deriving a final score through weighted averaging or other aggregation methods.

[0098] The rating system further transforms the scoring results into application suggestions. Different ratings correspond to different application scenarios and security requirements. High-rated devices are suitable for scenarios with high security requirements, such as financial payments and access control. Medium-rated devices are suitable for scenarios with general security requirements, such as device unlocking and authentication. Low-rated devices may require improvements before being put into practical use.

[0099] The generation of scores and ratings also considers the comprehensiveness and representativeness of the tests. If the test sample coverage is insufficient, or if certain important scenarios are lacking testing, the system will indicate the corresponding limitations in the evaluation report. This approach ensures the accuracy and reliability of the evaluation results.

[0100] Step S44 involves rating and determining the equipment's conformity, and supporting repeated testing to verify objectivity. Determining equipment conformity requires establishing clear standards and procedures to ensure the consistency and repeatability of the results.

[0101] The qualification of equipment is determined based on preset qualification standards, which are formulated according to application requirements and industry specifications. In one embodiment, the qualification standards for iris recognition equipment include specific indicators such as minimum accuracy requirements, maximum false detection rate limits, and upper limits for response time. Qualification standards may vary for different application scenarios, with high-security scenarios requiring more stringent performance indicators.

[0102] The pass / fail assessment employs a multi-round testing approach to avoid the randomness of single test results. The system undergoes multiple independent tests, each using a different sample set to ensure the stability of the results. Consistent results across multiple rounds indicate stable and reliable equipment performance. Significant discrepancies in test results require analysis of the causes and additional testing.

[0103] The design of repeated testing considers both the consistency of test conditions and the diversity of samples. Test conditions include environmental parameters, system configuration, and test procedures, which remain consistent across repeated tests to ensure the comparability of test results. Sample diversity is achieved through random sampling or stratified sampling to ensure that each test covers different types of samples.

[0104] Objectivity is verified through statistical analysis. The system calculates statistical measures such as the mean, variance, and confidence intervals of multiple test results to assess the stability and reliability of the results. If the statistical analysis shows good consistency, the test method is considered objective and reliable. If systematic deviations or abnormal fluctuations are found, the test method and equipment status need to be checked.

[0105] Step S45: Output a comprehensive report integrating indicators and interval analysis. The comprehensive report is the final presentation of the test results and needs to comprehensively, accurately, and clearly reflect the performance status of the equipment.

[0106] The comprehensive report is structured with sections including an executive summary, test methods, test results, performance analysis, and conclusions / recommendations. The executive summary provides the main findings and conclusions of the tests, facilitating a quick understanding of the results. The test methods section details the testing process, sample selection, evaluation criteria, and other information to ensure test repeatability. The test results section lists the specific values ​​and statistical analysis results for each performance indicator.

[0107] The performance analysis section is the core of the report. It uses a combination of charts and text to provide an in-depth analysis of the equipment's performance across different difficulty levels. The analysis includes performance trends, strengths and weaknesses, and influencing factors. Performance trend analysis shows the patterns of equipment performance changes with difficulty, helping to understand the equipment's applicable scope. Strengths and weaknesses analysis identifies the equipment's advantages and disadvantages, providing direction for improvement. Influencing factors analysis explores the key factors affecting equipment performance, providing a basis for optimization.

[0108] Interval analysis provides a more granular performance assessment by breaking down overall performance into sub-performance parameters of varying difficulty. This analytical approach can uncover the performance characteristics of a device under specific conditions, providing more precise guidance for application deployment. For example, a device might perform excellently on easy samples but experience a significant performance drop on difficult samples; this information is invaluable in determining the appropriate application scenarios for the device.

[0109] The conclusions and recommendations section, based on test results and analysis, provides suggestions for equipment improvement and application. The improvement suggestions address identified performance deficiencies and propose specific technical directions for improvement. The application suggestions recommend suitable application scenarios and deployment methods based on the equipment's performance characteristics. These recommendations provide important references for the further development and practical application of the equipment.

[0110] Step S5, system integration and deployment, includes server deployment and interface design to enable sample retrieval and repeated testing. System integration and deployment is a crucial step in integrating various functional modules into a complete testing platform.

[0111] Step S51: Deploy the database and units on the server. Server deployment needs to consider multiple aspects such as performance requirements, reliability requirements, and scalability requirements to ensure that the system can run stably and efficiently.

[0112] The choice of server hardware configuration is based on the system's computing and storage requirements. Computing requirements mainly stem from processing steps such as feature extraction, recognition algorithms, and evaluation analysis, necessitating sufficient processor and memory resources. Storage requirements arise from large amounts of virtual sample data, test result data, and log data, requiring large-capacity storage devices. Network requirements consider data transmission bandwidth and latency, necessitating the configuration of appropriate network interfaces.

[0113] The database is deployed using a distributed architecture, improving system processing power and reliability. Virtual sample data is partitioned and stored according to type and attributes, with each partition capable of independent read and write operations, enhancing concurrent processing capabilities. The database is also configured with backup and recovery mechanisms, regularly backing up important data to ensure data security.

[0114] The identification and evaluation units are deployed using a microservice architecture, with each functional module running as an independent service. This architecture facilitates system maintenance and upgrades; a failure in a single module will not affect the operation of the entire system. Services communicate with each other through standardized interfaces, ensuring system interoperability.

[0115] The load balancing mechanism ensures the system can handle high-concurrency test requests. When multiple test tasks are submitted simultaneously, the load balancer distributes the tasks to different processing nodes, preventing overload of any single node. Dynamic load balancing automatically adjusts the task allocation strategy based on real-time system load conditions.

[0116] Step S52: Design an interface that supports sample retrieval and presentation, including visual display of image samples and playback of audio samples. The user interface design needs to balance functionality and ease of use, providing users with a convenient operating experience.

[0117] The image samples are visualized using multiple display methods to meet different viewing needs. Thumbnail mode displays thumbnails of a large number of samples in a grid format for quick browsing and selection. Detail view mode displays high-resolution images of individual samples, supporting zooming, rotation, and other operations for careful observation of sample details. Comparison view mode displays multiple samples simultaneously for easy comparative analysis.

[0118] The image display functionality also includes feature annotation and analysis tools. Feature annotation can mark key feature points and region boundaries on the image, helping users understand sample characteristics. Analysis tools provide functions such as brightness distribution analysis, contrast analysis, and texture analysis, supporting quantitative assessment of sample quality.

[0119] The voice sample playback function supports multiple audio formats and playback controls. Playback controls include basic functions such as play, pause, fast forward, and rewind, as well as advanced functions such as volume adjustment and playback speed adjustment. The waveform display function graphically shows the time-domain waveform of the voice signal, facilitating the observation of the signal's temporal characteristics. The spectrum display function shows the frequency-domain characteristics of the voice signal, aiding in the analysis of the signal's frequency components.

[0120] The interface also displays sample attribute information, including sample number, generation time, quality level, and difficulty classification. This information helps users understand the basic characteristics of the samples and provides a reference for developing testing plans.

[0121] Step S53: If the retrieval method is random, then test samples are selected from the database. Random retrieval is an important method to ensure the fairness and unbiasedness of the test, and a scientific randomization strategy is required.

[0122] The random selection algorithm employs a pseudo-random number generator to ensure the randomness and repeatability of sample selection. The pseudo-random number generator produces a random sequence based on a seed value; the same seed value produces the same random sequence. This characteristic facilitates the verification and reproduction of test results. The seed value can be selected based on a timestamp, user input, or other random sources.

[0123] Stratified random sampling ensures that different types of samples are appropriately representative. The system first stratifies the samples in the database according to attributes such as type, difficulty, and quality, and then performs random sampling within each stratum. This method avoids the uneven sample distribution problem that may be caused by simple random sampling.

[0124] The sample retrieval process also considers the uniformity of temporal distribution. If the samples in the database were generated at different times, simple random sampling might lead to over-selection of samples from certain time periods. The system employs a time-stratified approach to ensure that samples from different time periods have reasonable representativeness.

[0125] The randomly selected results need to be statistically validated to ensure that the sampling results conform to the expected distribution characteristics. The system calculates various statistical indicators of the sample and compares them with the statistical indicators of the population sample to verify the representativeness of the sampling. If a significant deviation is found, the system will resample or adjust the sampling strategy.

[0126] In step S54, the integrated data transmission unit outputs the sample to the device under test. The data transmission unit is responsible for establishing a reliable data channel between the test system and the device under test, ensuring the accurate transmission of test data.

[0127] The selection of a data transmission protocol needs to consider factors such as transmission efficiency, reliability, and compatibility. For large-capacity image data, the system uses a transmission protocol that supports compression to reduce transmission time and bandwidth consumption. For applications with high real-time requirements, the system uses a low-latency transmission protocol to ensure timely response. For applications with high reliability requirements, the system uses a protocol with error detection and retransmission mechanisms.

[0128] The data format conversion function ensures that the transmitted data is compatible with the interface requirements of the device under test. Different devices may support different data formats, and the transmission unit needs to perform corresponding format conversions according to the device specifications. During the format conversion process, it is essential to maintain data integrity and accuracy to avoid information loss due to format conversion.

[0129] The transmission process monitoring and logging provide real-time feedback and historical tracking of the transmission status. The monitoring function displays current transmission progress, transmission rate, error status, and other information to help users understand the transmission situation. The logging records detailed transmission history, including transmission time, data volume, and transmission results, providing a basis for problem diagnosis and performance optimization.

[0130] Data security mechanisms protect data security during transmission. For test data containing sensitive information, the system employs encryption to prevent interception or tampering during transmission. An authentication mechanism ensures that only authorized devices can receive test data. An integrity verification mechanism verifies the integrity of received data, ensuring that data is not damaged during transmission.

[0131] Step S55: Debug the system to adapt to the specific environment and ensure the overall process is consistent. System debugging is a crucial step in ensuring the normal operation of the test platform, requiring comprehensive verification of the correctness and coordination of each functional module.

[0132] Functional testing verifies whether the basic functions of each module work properly. Functional testing of the database module includes verifying the correctness of data storage, querying, and updating operations. Functional testing of the recognition unit verifies the correctness of algorithms such as feature extraction, matching, and classification. Functional testing of the evaluation unit verifies the accuracy of functions such as performance calculation, scoring, and rating. Functional testing of the interface module verifies the normality of user interaction, data display, and operation response functions.

[0133] Performance testing evaluates the system's performance under different load conditions. Light load testing verifies the system's response time and processing capacity under normal operating loads. Heavy load testing verifies the system's stability and performance under high concurrency and large data volume conditions. Extreme load testing explores the system's performance boundaries, providing a basis for capacity planning.

[0134] Compatibility testing ensures the system can work correctly with different types of devices under test. Testing includes data format compatibility, communication protocol compatibility, and interface specification compatibility. The system needs to support multiple mainstream device types and interface standards, providing broad applicability.

[0135] Stability testing verifies the system's reliability during long-term operation. Long-term operation tests simulate continuous working scenarios in real-world applications, observing for issues such as memory leaks, performance degradation, and functional abnormalities. Fault recovery testing verifies the system's ability to recover from abnormal situations, including handling network interruptions, equipment failures, and data corruption.

[0136] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for generating, identifying, and evaluating biometric data, characterized in that, include: The system constructs a virtual biometric database by collecting initial samples and generating virtual data samples using a generator; The system preprocesses and filters virtual data samples, including noise reduction and outlier removal, to form an effective database. The system design identification unit identifies virtual data samples based on algorithms and outputs the identification results for testing the device under test. The system design evaluation unit determines the data difficulty range based on the identification results and expert knowledge, and scores and rates the equipment performance. System integration and deployment, including server deployment and interface design, to enable sample retrieval and repeated testing.

2. The method for generating, recognizing, and evaluating biometric data as described in claim 1, characterized in that, The construction of the virtual biometric database includes: Initial samples are collected as the basis for generation; A generator is used to generate virtual data samples for different biometric types. For image-type biometrics, the samples are generated by extracting and adjusting feature parameters, while for speech-type biometrics, the samples are generated by multi-feature space mapping and inverse mapping. If the generated virtual data sample meets the judgment criteria, it is added to the database; otherwise, it is discarded and the generation process is repeated. The generator decomposes biological features into multiple regions, processes these regions using subspace or set methods, and combines and reconstructs them to ensure the diversity and realism of the virtual data samples.

3. The method for generating, recognizing, and evaluating biometric data as described in claim 1, characterized in that, The preprocessing and filtering of virtual data samples includes: Specific noise reduction methods are used to process noise in virtual data samples of image type. If the processed virtual data sample shows no abnormalities after manual or semi-automatic screening, it is retained as a valid sample; otherwise, it is discarded. The denoising method selects a filter or transformation algorithm based on the biometric type to remove light and shadow occlusion or background interference. Repeat the screening process until the database reaches the preset size to ensure the objectivity and usability of the valid samples.

4. The method for generating, recognizing, and evaluating biometric data as described in claim 1, characterized in that, The design identification unit includes: Retrieve virtual data samples from the database as input; Feature extraction and classification of virtual data samples are performed based on deep learning algorithms combined with distance metrics or sparse representation models; If the extracted features meet the intra-class stability, the recognition result is output; The identification results are transmitted to the device under test to simulate real-world testing scenarios. The algorithm uses a sequence model to process correlations for time-series signals in order to improve recognition accuracy.

5. The method for generating, recognizing, and evaluating biometric data as described in claim 1, characterized in that, The design evaluation unit includes: Performance metrics, including accuracy and false detection rate, are calculated based on the recognition results. Introduce expert knowledge to determine the range of data difficulty; If the performance metrics match the difficulty range, a score and rating are generated. The rating determines the qualification of the equipment and supports repeated testing to verify its objectivity; Output a comprehensive report integrating indicators and interval analysis.

6. The method for generating, recognizing, and evaluating biometric data as described in claim 1, characterized in that, The system integration and deployment includes: Deploy the database and units on the server; The design interface supports sample retrieval and presentation, including visual display of image samples and playback of audio samples. If the retrieval method is random, then test samples will be selected from the database; The integrated data transmission unit outputs the sample to the device under test; Debug the system to adapt to the specific environment and ensure the overall process is consistent.

7. The method for generating, recognizing, and evaluating biometric data as described in claim 2, characterized in that, The method of using a generator to generate virtual data samples for different biometric types includes: The iris type is decomposed into feature regions, and the image is restored through perturbation and combination. If the restored image meets the judgment criteria, it is deemed qualified. The arrangement and combination of fingerprint type texture elements are constrained by prior knowledge. After adjusting key parameters for speech type, samples are constructed through inverse mapping. Repeat the adjustment process to generate a sufficient number of virtual data samples.

8. The method for generating, recognizing, and evaluating biometric data as described in claim 4, characterized in that, The method of extracting and classifying features from virtual data samples based on deep learning algorithms combined with distance metrics or sparse representation models includes: Extract global and local features to capture details; If the features overcome the noise interference, then class interval comparison is performed; Time-frequency domain transformation and feature extraction are performed on speech samples; Time series signals are classified using sequence models; The classification results are used as the basis for recognition.

Images