Living fingerprint detection method based on multispectral and micro pulse feature fusion

Abstract

The invention relates to the technical field of living body fingerprint detection, and particularly discloses a living body fingerprint detection method based on multispectral and micro pulse feature fusion, which comprises the steps of signal synchronous generation, time domain alignment processing, feature matrix generation, weight dynamic adjustment, cross validation analysis, abnormal region sampling, living body model verification and living body authentication output. According to the method, the verification result is generated by synchronously collecting the multi-spectral reflection intensity and the skin displacement fluctuation sequence of the fingerprint area, dynamically adjusting the spectral absorption and pulsation track feature weight, performing cross validation to output the local abnormal area and obtain the supplementary feature data, and performing physiological coupling rule matching by using the preset living body judgment model. When three consecutive results do not conform to a living body rule, activating an anti-counterfeiting attack alarm protocol; according to the method, multispectral and micro pulse features are fused, and an environment adaptive weight mechanism is combined, so that the accuracy of living body detection is effectively improved, and the environment adaptability of a detection algorithm is enhanced.

Description

Technical Field

[0001] This invention relates to the field of live fingerprint detection technology, and more specifically, to a live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics. Background Technology

[0002] Fingerprint recognition, as a mature and widely used biometric identification technology, has penetrated into many fields such as mobile payment, access control, and information security. To ensure the security of fingerprint recognition systems and prevent fraudulent attacks using fake fingerprints and other forgeries, live fingerprint detection technology has emerged. The core of this technology lies in analyzing whether the collected fingerprint image or signal contains live biological signs to distinguish between real fingers and artificial imitations, making it a crucial link in building a secure authentication system.

[0003] Currently, existing live fingerprint detection technologies mainly rely on detecting single physiological features. For example, some methods determine liveness by analyzing changes in blood flow in the fingerprint area, i.e., photoplethysmography (PPG) signals; others use the elastic deformation or minute pulsations of the skin when a finger is pressed as evidence of liveness. In addition, some technologies attempt to integrate multiple features, but this usually requires the configuration of multiple independent sensors, such as using optical sensors and pressure sensors to collect different signals.

[0004] However, the aforementioned existing technical solutions have significant limitations. Detection methods based on single features are easily countered; for example, a silicone fingerprint mold with a miniature light source could simulate blood flow signals, while a fake fingerprint made of highly elastic material could simulate skin deformation. For multi-sensor solutions, accurate spatiotemporal synchronization of data between different sensors is difficult, and the system structure is complex and costly. Even with multi-feature fusion, a fixed fusion strategy is difficult to adapt to complex and changing usage environments. For example, strong ambient light can severely interfere with the quality of optical signals, and the dryness or moisture level of the user's skin can affect the acquisition of various physiological signals, leading to unstable detection performance. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: This invention provides a live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse features, including the following steps: S1, signal synchronous generation: acquire the synchronously collected multispectral reflectance intensity sequence of the fingerprint region and the skin sub-pixel displacement fluctuation sequence to generate a spatiotemporally synchronized original signal.

[0007] S2. Time-domain alignment processing: Perform time-domain alignment processing on the original spatiotemporal synchronization signals to form a set of spatiotemporally correlated data pairs.

[0008] S3. Feature Matrix Generation: Extract the spectral absorption feature vector of the multispectral reflectance intensity sequence in the spatiotemporal correlated data pair set, and simultaneously extract the pulsation trajectory feature vector of the skin subpixel displacement fluctuation sequence at the corresponding time point. The spectral absorption feature vector and the pulsation trajectory feature vector are fused and encoded according to the timestamp to generate a dynamic coupling feature matrix.

[0009] S4. Dynamic weight adjustment: The weight ratio of spectral absorption features and pulsation trajectory features in the dynamic coupling feature matrix is ​​dynamically adjusted according to environmental interference parameters to generate adaptive weight decision rules.

[0010] S5. Cross-validation analysis: Based on the adaptive weight decision rule, cross-validation analysis is performed on the dynamic coupling feature matrix to output the identifier of local feature anomaly regions.

[0011] S6. Abnormal Area Sampling: Triggers an enhanced sampling command for local abnormal area markers to obtain supplementary feature data with extended acquisition cycles.

[0012] S7. Live model verification: Input the supplementary feature data into the preset liveness determination model to match the physiological coupling law and generate liveness verification results.

[0013] S8. Liveness Authentication Output: Based on the liveness verification result, output a liveness fingerprint authentication pass signal.

[0014] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention acquires physiological signals of two different modalities, namely multispectral reflectance intensity and skin subpixel displacement fluctuation, simultaneously and performs in-depth fusion analysis on them. Since the signals originate from the same physical region and are perfectly synchronized in time, the registration and synchronization errors of the multi-sensor scheme are fundamentally eliminated, ensuring the high fidelity and intrinsic correlation of the original data, thereby providing a reliable data basis for accurate discrimination and significantly improving the accuracy of liveness detection.

[0015] (2) By introducing an environmental adaptive weight decision-making mechanism, this invention can monitor interference factors such as ambient light intensity and skin dryness in real time, and dynamically adjust the weight ratio of spectral absorption features and pulsation trajectory features in the final decision. This intelligent adaptive capability enables the system to selectively rely on the current signal source with better quality, effectively compensating for signal distortion in complex and ever-changing application scenarios, thereby enhancing the robustness and environmental adaptability of the detection algorithm.

[0016] (3) This invention establishes a hierarchical intelligent verification process. When preliminary cross-validation reveals contradictions in local features, it does not directly reject the application but instead triggers enhanced sampling and deep verification of the abnormal region. By analyzing the stability and physiological coupling patterns of extended period signals, this process can distinguish between genuine physiological abnormalities and malicious forgery attacks with higher confidence. This refined processing strategy effectively reduces the false rejection rate caused by individual differences or poor contact while ensuring high security, thus improving the user experience. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0019] Figure 2 This is a diagram illustrating the conflict area of ​​the present invention.

[0020] Figure 3 This is a graph showing the relationship between the phase offset and the fluctuation amplitude of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, the present invention provides a live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse features, including: S1, signal synchronous generation: acquiring the synchronously collected multispectral reflectance intensity sequence of the fingerprint region and the skin sub-pixel displacement fluctuation sequence to generate a spatiotemporally synchronized original signal.

[0023] In a specific embodiment of the present invention, the specific process of generating the spatiotemporal synchronization original signal is as follows: controlling a multi-band light source to alternately irradiate the fingerprint area at a preset frequency.

[0024] It should be noted that the control system drives a multi-band light source integrating multiple light-emitting units with different center wavelengths, such as blue light with a wavelength of 450nm, green light with a wavelength of 530nm, red light with a wavelength of 660nm, and near-infrared light with a wavelength of 940nm. The control system cyclically triggers these light-emitting units at a precise and fixed preset frequency, causing them to alternately illuminate the fingerprint area placed on the sensor surface.

[0025] In a specific embodiment of the present invention, the preset frequency can be 100Hz. The choice of this frequency is based on the fact that it needs to be precisely matched with the acquisition capability of the high frame rate macro optical sensor. The sensor synchronously captures skin reflection images under each band of illumination with millisecond-level time accuracy. For example, the first frame image is acquired immediately when blue light is on, and the second frame image is acquired when green light is on. By cyclically triggering four light sources of 450nm blue light, 530nm green light, 660nm red light and 940nm near-infrared light at a fixed frequency of 100Hz, it is ensured that the timing of each band of illumination corresponds strictly with the sensor exposure, thereby generating a raw multispectral video stream without synchronization error, providing a high-fidelity data foundation for the subsequent extraction of skin subpixel displacement fluctuation sequences.

[0026] A sequence of skin surface reflection images under each wavelength of light is captured synchronously using a macro optical sensor.

[0027] It should be noted that, while the light source alternates, a high-frame-rate macro optical sensor is simultaneously capturing a sequence of skin surface reflection images of the fingerprint area under each single wavelength of illumination. This synchronization is a key technical feature, meaning that the exposure time of each frame of the sensor strictly corresponds to the illumination time of a specific wavelength of light source. For example, in the... Milliseconds later, blue light illuminates, and the sensor captures the first frame of the image; in the second... Milliseconds later, a green light illuminates, the sensor captures the second frame, and so on. This continuous acquisition creates a raw video stream containing multispectral information—a sequence of images showing reflections from the skin surface.

[0028] The multispectral reflectance intensity sequence is separated from the skin surface reflectance image sequence.

[0029] It should be noted that the skin surface reflectance image sequence is processed in parallel to separate two target signals. The first is a multispectral reflectance intensity sequence. The processing method involves defining multiple regions of interest (ROIs) in the image sequence, typically located in areas with high fingerprint image quality. For each frame, the average grayscale value of the pixels within its corresponding ROI is automatically acquired. Since each frame corresponds to a known light source band, these average grayscale values ​​are organized according to time sequence and band information to form the multispectral reflectance intensity sequence. .in Represents the intensity of reflection. Represents spectral bands, Represents time.

[0030] The displacement fluctuation of the skin surface between adjacent frames is calculated based on the image subpixel displacement algorithm, and a skin subpixel displacement fluctuation sequence is generated.

[0031] It should be noted that the specific process of generating the skin subpixel displacement fluctuation sequence is as follows: 1) Selecting frame pairs: Selecting two adjacent frames from a continuous image sequence; 2) Calculating correlation: Calculating the correlation between the two frames. and Performing two-dimensional Fourier transforms on each yields the frequency domain representation: ,in, The cross-power spectrum is obtained by multiplying the frequency domain representations of the two frames conjugately and normalizing the results. , ;in, yes The complex conjugate, with the denominator being the modulus normalization term, for the cross power spectrum Perform a two-dimensional inverse Fourier transform: ; It is a matrix where the value of each point represents the similarity between two frames at that offset; 3) Subpixel localization: Locating peak points: Finding the maximum point in the correlation matrix, and its coordinates This refers to the integer pixel-level displacement between two frames of images. Within a small neighborhood around the peak point, a quadratic function is used to fit the data to obtain more accurate coordinates of the extreme points, assuming that the area around the peak point conforms to a quadratic surface model. Parameters fitted using the least squares method extreme point coordinates This is subpixel-level displacement; 4) Generate sequence: Repeat the above calculation for all consecutive frame pairs in the entire image sequence to obtain a time-varying two-dimensional displacement vector sequence. The magnitude of this sequence or its projection in a specific direction constitutes the skin subpixel-level displacement fluctuation sequence. ,in t represents the displacement fluctuation, and t represents time.

[0032] This invention enables the parallel extraction of two different modalities of live physiological signals from the same sequence of reflective images acquired by a single macro optical sensor. This single-source synchronous acquisition mechanism fundamentally ensures perfect temporal alignment of the two signals, avoiding synchronization errors and calibration challenges that may be introduced by multi-sensor solutions. It provides a high-quality, high-fidelity raw data foundation for subsequent feature fusion analysis, significantly improving the reliability and robustness of live fingerprint detection algorithms in distinguishing between real and forged fingerprints, as forged materials cannot simultaneously simulate the precise physiological coupling between these two signals.

[0033] S2. Time-domain alignment processing: Perform time-domain alignment processing on the original spatiotemporal synchronization signals to form a set of spatiotemporally correlated data pairs.

[0034] In a specific embodiment of the present invention, the specific process of forming a spatiotemporally correlated data pair set is as follows: identifying the starting synchronization timestamps of the multispectral reflectance intensity sequence and the skin subpixel displacement fluctuation sequence.

[0035] It should be noted that the processing begins with the initial synchronization timestamp between the multispectral reflectance intensity sequence and the skin subpixel-level displacement fluctuation sequence. Although the two signals are synchronized during acquisition, the analysis typically starts from a point in time after the fingerprint and sensor have reached stable contact, to avoid noise interference in the initial stage. This point in time is the initial synchronization timestamp, denoted as […]. .

[0036] The two sequences are truncated using a preset time window to generate time-aligned data segments.

[0037] In one specific embodiment of the present invention, the duration of the preset time window is... Typically, the step size is set based on the general pattern of the human cardiac cycle, specifically 1.5 seconds, to ensure that each window can completely capture the physiological information of at least one heartbeat event. The step size needs to ensure that each slide can cover at least half of the cardiac cycle, thereby stably capturing the physiological coupling characteristics of skin displacement and spectral reflectance. Therefore, the step size can be set to 0.75 seconds, so that the windows overlap by 50% after each slide, which can both ensure the continuity of features and control the amount of computation.

[0038] The spectral reflectance intensity value at the same time point within each time window is linked to the displacement fluctuation as a data pair.

[0039] It should be noted that for data extracted within any time window, at each discrete time sampling point... Above, there exists a vector composed of light intensities in different spectral bands. and a corresponding skin displacement amount ,in, Indicates different spectral bands, Indicates the total number of spectral bands. Indicates the discrete time sampling point number. This method converts the spectral reflectance intensity vector at the same time point within each time window into... With skin displacement Bind as a related data pair. For example, at a point in time. The generated data pairs can be represented as The associated data pairs generated from all sampling points within a time window are aggregated, and then the associated data pairs generated from all sliding windows are merged to form a spatiotemporal associated data pair set.

[0040] This invention successfully transforms two continuous, dynamically changing physiological signal streams into a set of discretized, structured, and internally time-series-preserving spatiotemporally correlated data pairs. The core technical advantage of this processing method lies in its ability to not only ensure the alignment of multimodal data over macroscopic time periods but also achieve precise binding at microscopic time points. This lays a solid foundation for subsequent feature extraction, ensuring that features extracted from spectral signals can be directly compared and fused with physical displacement features caused by arterial pulsation at the same instant. This allows for in-depth exploration of the inherent physiological coupling patterns between the two signals, significantly improving the analytical accuracy and reliability of the liveness detection model.

[0041] S3. Feature Matrix Generation: Extract the spectral absorption feature vector of the multispectral reflectance intensity sequence in the spatiotemporal correlated data pair set, and simultaneously extract the pulsation trajectory feature vector of the skin subpixel displacement fluctuation sequence at the corresponding time point. The spectral absorption feature vector and the pulsation trajectory feature vector are fused and encoded according to the timestamp to generate a dynamic coupling feature matrix.

[0042] In a specific embodiment of the present invention, the specific process of generating the dynamic coupling feature matrix is ​​as follows: calculate the gradient value of the spectral reflection intensity of each associated data pair as the band changes, and construct the spectral absorption feature vector.

[0043] It should be noted that the gradient value of the spectral reflectance intensity of each correlated data pair as a function of the spectral band is calculated to form the spectral absorption feature vector. For any time within a time window... Related data Extract its multispectral components Spectral absorption eigenvectors The calculation is accomplished by evaluating the rate of change of spectral intensity between adjacent bands, i.e., the spectral gradient. For example, the vector's... Each element can be calculated using the following formula: ,in, Is the vector at the th order? The value of each element, and They are time points Below, the center wavelength is and The reflected intensity value corresponding to the light source.

[0044] Extract the frequency domain energy distribution of displacement fluctuations within the corresponding time window to form the pulsation trajectory feature vector.

[0045] It should be noted that the frequency domain energy distribution of displacement fluctuations within the corresponding time window is extracted to construct the pulsation trajectory feature vector. This step focuses on the entire time window. Subpixel displacement fluctuation sequence of skin within The specific process for constructing the pulsation trajectory feature vector is as follows: 1) Fast Fourier Transform: using a programming language to analyze the sub-pixel-level displacement fluctuation sequence of the skin. Perform a Fast Fourier Transform to obtain the frequency domain sequence. 2) Determine the frequency resolution and frequency range: Frequency resolution ,in It is the sampling frequency. It is the length of the time-domain sequence, representing the interval between two adjacent frequency points in the frequency domain. According to the sampling theorem, the frequency range in the frequency domain is from 0 to... 3) Locating the physiological heart rate frequency range: Based on actual needs and research subjects, determine the specific physiological heart rate frequency range. Based on the frequency resolution and frequency range, find the frequency domain sequence. The frequency point index range corresponding to the physiological heart rate frequency range, assuming frequency points Then find the satisfying 4) Calculate energy: frequency domain sequence It is a complex number, and its magnitude It can be done ;in and They are The real and imaginary parts. For each frequency point within the physiological heart rate frequency range. Calculate its amplitude For each frequency point Its energy Within the physiological heart rate frequency range, the energy at each frequency point is calculated; 5) The energy or amplitude of each frequency component within the calculated physiological heart rate frequency range is arranged in ascending order of frequency, thus forming the pulsation trajectory feature vector of this time window. .

[0046] The spectral absorption feature vectors and pulsation trajectory feature vectors of the same time window are tensor-concatenated to form a three-dimensional feature matrix.

[0047] It should be noted that the spectral absorption feature vector and the pulsation trajectory feature vector within the same time window are tensor-concatenated to form a three-dimensional feature matrix. In this step, the two features calculated within the same time window are fused. Because the spectral absorption feature vector... It changes over time, while the pulsation trajectory feature vector This is a summary of the entire window, therefore, merging requires aligning the latter with the former in the time dimension. Specifically, for each point in time within this time window... The global pulsation trajectory feature vector Spectral absorption eigenvectors appended to this moment Then, the new combined vectors of all time points within the window are stacked in chronological order to form a two-dimensional matrix. When we stack such two-dimensional matrices generated from multiple consecutive, possibly overlapping, time windows along a new dimension, we form the three-dimensional feature matrix described in this invention, also known as the dynamically coupled feature matrix. Its three dimensions represent the time window index, the time step within the window, and the fused features, respectively.

[0048] S4. Dynamic weight adjustment: The weight ratio of spectral absorption features and pulsation trajectory features in the dynamic coupling feature matrix is ​​dynamically adjusted according to environmental interference parameters to generate adaptive weight decision rules.

[0049] In a specific embodiment of the present invention, the specific process of generating adaptive weight decision rules is as follows: detecting ambient light intensity parameters and skin dryness parameters as environmental interference parameters.

[0050] It should be noted that the ambient light intensity parameter It can be directly measured using a dedicated ambient light sensor integrated into the fingerprint sensor module. Skin dryness parameter. Data was collected using a professional skin impedance meter.

[0051] When the ambient light intensity exceeds a preset threshold, the weighting coefficient of the spectral absorption feature vector is reduced.

[0052] It should be noted that strong ambient light can interfere with the signals emitted by active light sources, polluting the purity of the spectral absorption feature vector and reducing its signal-to-noise ratio. Therefore, an ambient light threshold is set. .when At that time, the weighting coefficients of the spectral features will be reduced. , These are the weighting coefficients for the adjusted spectral features. These are the initial weighting coefficients for the spectral features. Represents the natural constant. This represents the attenuation factor, and its specific value was obtained through extensive experimental testing, measuring the signal-to-noise ratio changes of the spectral absorption eigenvector under different ambient light intensities. .

[0053] In one specific embodiment of the present invention, the ambient light threshold can be set to 700 lux. If the setting is too low, the weight coefficient of the spectral absorption feature vector may be incorrectly reduced under normal lighting conditions, affecting normal signal analysis. If the setting is too high, it may not be able to effectively deal with strong ambient light interference in a timely manner. Therefore, setting it to 700 lux can effectively deal with possible strong ambient light interference while ensuring normal signal processing.

[0054] When skin dryness exceeds a preset threshold, the weight coefficient of the pulsation trajectory feature vector is increased.

[0055] It should be noted that excessively dry skin reduces the efficiency of light penetration into the dermis, resulting in a weakening of spectral signals reflecting changes in blood flow. However, it has a relatively small impact on skin mechanical displacement directly driven by arterial pulsation, i.e., the characteristics of pulsation trajectory. Therefore, setting a skin dryness threshold is necessary. .when At that time, the weighting coefficient of the pulsation trajectory characteristics will be increased. , This represents the weighting coefficients of the adjusted pulsation trajectory characteristics. These are the initial weighting coefficients for the pulsation trajectory characteristics. This represents the adjustment coefficient, which determines the rate at which the function increases with skin dryness. Its specific value was obtained through extensive experimental testing. .

[0056] In a specific embodiment of the present invention, based on a large amount of experimental and clinical data statistics, when normal skin is measured using a skin impedance meter, if the skin impedance value measured by the skin impedance meter is greater than the threshold of 70kΩ, the skin is considered to be in a relatively dry state. Therefore, the skin dryness threshold value is set at 70kΩ.

[0057] Decision boundary conditions based on reconstructing the dynamically coupled feature matrix using weight coefficients.

[0058] It should be noted that the decision boundary conditions are reconstructed based on the weighting coefficients of the dynamically coupled feature matrix. This step does not directly modify the numerical values ​​of the dynamically coupled feature matrix, but rather applies these weights in subsequent classification or discriminant analysis. When the liveness detection model analyzes the feature matrix, it multiplies the extracted spectral absorption features by the adjusted weighting coefficients. Multiply the pulsation trajectory characteristics by This essentially changes the contribution of different features to the final decision, thereby dynamically adjusting the decision hyperplane, i.e., the decision boundary conditions, in the high-dimensional feature space that distinguishes between live and fake fingerprints.

[0059] S5. Cross-validation analysis: Based on the adaptive weight decision rule, cross-validation analysis is performed on the dynamic coupling feature matrix to output the identifier of local feature anomaly regions.

[0060] Please see Figure 2 As shown, in a specific embodiment of the present invention, the specific process of identifying the abnormal region of the output local features is as follows: under the reconstructed decision boundary conditions, identify conflict regions where the spectral absorption features conform to the characteristics of living organisms but the pulsation trajectory features are below the intensity threshold.

[0061] It should be noted that the cross-validation analysis method described in this invention is a key step in verifying the consistency of different modal features within the dynamically coupled feature matrix, guided by the adaptive weight decision rules generated in the previous step. This analysis is not an independent judgment of a single feature, but rather a verification of the intrinsic correlation between two features, and the process unfolds in the spatial dimension of the fingerprint image. First, under the reconstructed decision boundary conditions, conflict regions are identified where the spectral absorption features conform to the characteristics of a living organism, but the pulsation trajectory features are below the intensity threshold. Specifically, the effective fingerprint region is divided into multiple sub-regions for analysis. For each sub-region, its two features are evaluated in parallel. On the one hand, a preset living organism spectral model is used to determine whether the spectral absorption features of the region conform to the typical pattern of the living blood absorption spectrum. On the other hand, the energy of the corresponding pulsation trajectory features in the region is detected to exceed a preset pulsation intensity threshold. When the analysis results of a certain sub-region show that its spectral absorption features are determined to conform to the typical pattern of the living blood absorption spectrum, but the energy of its pulsation trajectory features fails to reach the threshold, this contradictory feature phenomenon is identified, and the sub-region is defined as a conflict region.

[0062] It should also be noted that the specific method for determining whether the spectral absorption characteristics of the region conform to the typical pattern of the absorption spectrum of living blood using the preset live spectral model is as follows: extract a threshold range of spectral absorption characteristics determined based on a large amount of spectral data from living blood samples from the preset live spectral model, compare the spectral absorption characteristic value extracted from the fingerprint sub-region spectrum with the threshold range, and if the spectral absorption characteristic value is within the threshold range, it is determined that the spectral absorption characteristic value of the region conforms to the typical pattern of the absorption spectrum of living blood; if the spectral absorption characteristic value is not within the threshold range, it is determined that the spectral absorption characteristic value of the region does not conform to the typical pattern of the absorption spectrum of living blood.

[0063] In a specific embodiment of the present invention, the preset typical value of the pulsation intensity threshold can be set to 0.2. The value is mainly based on the energy analysis of the pulsation trajectory characteristics of a large number of healthy living finger samples. Statistical analysis found that when normal living blood pulsates, the energy of the corresponding area is generally within a certain range. The typical value is taken as the value that can more accurately distinguish between living and non-living people and has a certain degree of fault tolerance, so as to ensure that pulsation that conforms to the characteristics of living people can be effectively identified, while avoiding misjudgment caused by individual differences or environmental interference.

[0064] The conflict area is marked as a local feature anomaly region identifier.

[0065] It should be noted that the conflict areas are marked as local feature anomaly regions. This step involves aggregating all sub-regions identified as conflict areas to generate a marker. This marker can be a binary mask image, where the pixel value corresponding to the conflict area is 1, and other areas are 0, to accurately indicate the spatial location and range of the feature inconsistency phenomenon on the fingerprint image.

[0066] Calculate the ratio of the total area of ​​the conflict region to the total effective area of ​​the fingerprint, i.e. the proportion of the conflict region area. When the proportion of the conflict region area exceeds the preset tolerance, send an enhanced sampling command to the acquisition module.

[0067] It should be noted that the effective total area of ​​the fingerprint is usually obtained by directly acquiring the original fingerprint image through an optical fingerprint sensor, followed by image preprocessing to accurately delineate and identify the effective total area of ​​the fingerprint. The preset tolerance typical value can be set to 15%. This value is based on a comprehensive consideration of the reasonable fluctuations in conflict areas caused by individual differences and equipment errors during normal fingerprint acquisition, as well as the threshold ratio set to ensure the accuracy and reliability of fingerprint liveness detection and avoid misjudgment due to minor conflict areas. This threshold ratio can tolerate a certain reasonable error while effectively identifying abnormal situations.

[0068] S6. Abnormal Area Sampling: Triggers an enhanced sampling command for local abnormal area markers to obtain supplementary feature data with extended acquisition cycles.

[0069] In a specific embodiment of the present invention, the specific process of obtaining supplementary feature data with extended acquisition period is as follows: extend the data acquisition time of the corresponding position of the local feature anomaly region identifier to a set multiple of the base time.

[0070] It should be noted that the data acquisition duration at the location corresponding to the identified local feature anomaly area is extended to a set multiple of the baseline duration. Here, the baseline duration is the preset time window duration used in the initial analysis. .

[0071] In one specific embodiment of the present invention, the multiplier can be set to 2 times, which is intended to capture at least two complete physiological cardiac cycles. This extension of duration is crucial for confirming the periodicity of the signal, because fluctuations in a single cycle may be caused by random noise or artifacts, while the stable repetition of multiple consecutive cycles is strong evidence of in vivo physiological signals.

[0072] Extract the periodic fluctuation pattern of pulsation trajectory characteristics within an extended acquisition period.

[0073] It should be noted that extracting the periodic fluctuation pattern of pulsation trajectory features within an extended acquisition period refers to obtaining more complete physiological cycle data by extending the data acquisition time of local abnormal feature areas, analyzing the skin subpixel-level displacement fluctuation sequence, and extracting pulsation frequency and amplitude, i.e., the number of pulsations and displacement fluctuations per unit time, which exhibit periodic changes over time. These parameters maintain a relatively stable repetitive pattern in different physiological cycles.

[0074] Please see Figure 3 As shown, S7, Live Model Verification: Input the supplementary feature data into the preset live determination model to perform physiological coupling law matching and generate live verification results.

[0075] In a specific embodiment of the present invention, the specific process of generating the liveness verification result is as follows: establishing a reference range for the fluctuation amplitude of the spectral absorption characteristics within the pulse cycle.

[0076] It is important to note that a benchmark is needed, namely, establishing a baseline range for the fluctuation amplitude of spectral absorption characteristics within the pulse cycle. This baseline range is not calculated temporarily, but rather is a preset parameter obtained through prior calibration of a large number of real live fingerprint samples. It defines a physiologically reasonable interval. This represents the normal fluctuation range of spectral absorption intensity caused by changes in blood volume during a single cardiac cycle.

[0077] The phase shift between the spectral absorption fluctuations and the peaks of the pulsation trajectory in the supplementary feature data is detected.

[0078] It should be noted that we are now entering the actual detection phase of the currently acquired supplementary feature data. The first detection step is to detect the phase shift between the spectral absorption fluctuations and the peaks of the pulsation trajectory in the supplementary feature data. From the supplementary feature data, by analyzing the periodicity of the skin's sub-pixel-level displacement fluctuation sequence, we have accurately identified the starting point and the time point of the peak appearance of each pulse cycle. Simultaneously, within the exact same time period, the changes in the multispectral reflectance intensity sequence were analyzed to identify the time when the absorption characteristic extreme point occurs due to the blood filling reaching its maximum. Due to the characteristics of hemodynamics, there is a small but stable time delay between the peak of mechanical pressure and the peak of blood volume. This time difference is calculated to quantify the phase shift. .

[0079] When the fluctuation amplitude is within the reference range and the phase offset is less than the preset error threshold, it is determined to conform to the physiological coupling law of living organisms. When the fluctuation amplitude is not within the reference range or the phase offset is greater than or equal to the preset error threshold, it is determined to not conform to the physiological coupling law of living organisms.

[0080] It should be noted that the final determination is based on a preset coupling rule. The determination criteria include two aspects. First, the average fluctuation amplitude of the spectral absorption characteristics in the supplementary feature data over multiple complete pulse cycles is obtained. And verify whether it falls within the preset benchmark range, i.e. On the other hand, the calculated phase offset... The system compares the fluctuation amplitude with a preset error threshold. Only when the fluctuation amplitude is within the baseline range and the phase offset is less than the preset error threshold will it be finally determined to conform to the physiological coupling law of living organisms.

[0081] In one specific embodiment of the present invention, the typical value of the preset error threshold can be set to 0.1 seconds. The value is based on the study of the physiological characteristics of a large number of real live fingerprint samples. Considering that there is a certain time delay from the blood being pumped out of the heart to causing mechanical displacement of the finger skin, but this delay is relatively stable and within a small range, experimental statistical analysis has found that when the phase offset is less than 0.1 seconds, it can distinguish between live and non-live fingerprints more accurately. This avoids misjudgment caused by individual differences and effectively eliminates abnormal situations that may occur in forged fingerprints because they cannot simulate this precise physiological coupling relationship, thereby ensuring the accuracy and reliability of liveness detection.

[0082] S8. Liveness Authentication Output: Based on the liveness verification result, output a liveness fingerprint authentication pass signal.

[0083] In a specific embodiment of the present invention, the specific process of outputting the live fingerprint authentication pass signal is as follows: when the live verification results are inconsistent with the live physiological coupling law for three consecutive times, the anti-counterfeiting attack alarm protocol is activated.

[0084] Freeze the current fingerprint authentication process and record attack signature patterns.

[0085] The attack signature pattern is uploaded to the security certification center database.

[0086] It should be noted that the additional security measures implemented after outputting the liveness verification result are an intelligent response protocol designed to counter persistent attack attempts. This process is not triggered immediately after a single verification failure, but rather based on a cumulative counting mechanism for consecutive failures. The system maintains a failure counter for each authentication session. After completing each full liveness detection process and generating a liveness verification result, if the verification result does not conform to the physiological coupling law of liveness, it means that this attempt has been judged as non-liveness, and the failure counter is incremented by one. Conversely, if any verification passes successfully, the counter is immediately reset to zero.

[0087] It should also be noted that the intelligent response protocol is triggered when three consecutive verification results fail to conform to the physiological coupling law of living organisms. That is, when the failure counter accumulates to 3, the system determines that it is currently under continuous and potentially malicious attacks, and automatically activates the anti-spoofing attack alarm protocol. This protocol is a predefined, automatically executed response procedure. Once activated, the protocol immediately freezes the current fingerprint authentication process, stops accepting any new fingerprint input, and displays authentication lock or failure information to the user interface, effectively preventing attackers from repeatedly trying to find system vulnerabilities. Simultaneously, the system performs data solidification, recording the attack signature pattern. This pattern is a data packet containing all the key information leading to the consecutive failures, which may include the multispectral reflectance intensity sequence, skin sub-pixel displacement fluctuation sequence, extracted spectral absorption features and pulsation trajectory features from the three failed attempts, the final generated dynamic coupling feature matrix, and anomaly markers from all intermediate analysis stages. Finally, the system uploads the encapsulated attack signature pattern to the security authentication center database via an encrypted channel.

[0088] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics, characterized in that, Includes the following steps: S1. Signal Synchronization Generation: Acquire the multispectral reflectance intensity sequence of the fingerprint region and the sub-pixel displacement fluctuation sequence of the skin synchronously collected, and generate the spatiotemporal synchronous raw signal; S2. Time-domain alignment processing: Perform time-domain alignment processing on the original spatiotemporal synchronization signal to form a set of spatiotemporally correlated data pairs; S3. Feature Matrix Generation: Extract the spectral absorption feature vector of the multispectral reflectance intensity sequence in the spatiotemporal correlated data set, and simultaneously extract the pulsation trajectory feature vector of the skin sub-pixel displacement fluctuation sequence at the corresponding time point. Fusion and encoding of the spectral absorption feature vector and the pulsation trajectory feature vector according to the timestamp to generate a dynamic coupled feature matrix. S4. Dynamic weight adjustment: The weight ratio of spectral absorption features and pulsation trajectory features in the dynamic coupling feature matrix is ​​dynamically adjusted according to environmental interference parameters to generate adaptive weight decision rules. S5. Cross-validation analysis: Based on the adaptive weight decision rule, cross-validation analysis is performed on the dynamic coupling feature matrix to output the local feature anomaly region identifier; S6. Abnormal Area Sampling: Triggers an enhanced sampling command for local abnormal area markers to obtain supplementary feature data with extended acquisition period; S7. Live model verification: Input the supplementary feature data into the preset live determination model to match the physiological coupling law and generate live verification results; S8. Liveness Authentication Output: Based on the liveness verification result, output a liveness fingerprint authentication pass signal.

2. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 1, characterized in that: The specific process for generating the original spatiotemporal synchronization signal is as follows: Control the multi-band light source to alternately illuminate the fingerprint area at a preset frequency; A sequence of skin surface reflection images under each wavelength of light is captured synchronously using a macro optical sensor; The multispectral reflectance intensity sequence is extracted from the skin surface reflectance image sequence; The displacement fluctuation of the skin surface between adjacent frames is calculated based on the image subpixel displacement algorithm, and a skin subpixel displacement fluctuation sequence is generated.

3. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 1, characterized in that: The specific process for forming the spatiotemporally correlated data pair set is as follows: Identify the initial synchronization timestamps of the multispectral reflectance intensity sequence and the subpixel displacement fluctuation sequence of the skin; The two sequences are truncated using a preset time window to generate time-aligned data segments; The spectral reflectance intensity value at the same time point within each time window is linked to the displacement fluctuation as a data pair.

4. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 3, characterized in that: The specific process for generating the dynamically coupled feature matrix is ​​as follows: Calculate the gradient value of spectral reflectance intensity as a function of band in each associated data pair to form a spectral absorption feature vector; Extract the frequency domain energy distribution of displacement fluctuation within the corresponding time window to construct the pulsation trajectory feature vector; The spectral absorption feature vectors and pulsation trajectory feature vectors of the same time window are tensor-concatenated to form a three-dimensional feature matrix.

5. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 1, characterized in that: The specific process for generating adaptive weight decision rules is as follows: Ambient light intensity and skin dryness parameters were used as environmental interference parameters. When the ambient light intensity exceeds a preset threshold, the weighting coefficient of the spectral absorption feature vector is reduced. When skin dryness exceeds a preset threshold, the weight coefficient of the pulsation trajectory feature vector is increased; Decision boundary conditions based on reconstructing the dynamically coupled feature matrix using weight coefficients.

6. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 5, characterized in that: The specific process for identifying local feature anomaly regions in the output is as follows: Under the reconstructed decision boundary conditions, identify conflict regions where the spectral absorption characteristics conform to in vivo characteristics but the pulsation trajectory characteristics are below the intensity threshold; The conflict area is marked as a local feature anomaly region identifier; Calculate the ratio of the total area of ​​the conflict region to the total effective area of ​​the fingerprint, i.e. the proportion of the conflict region area. When the proportion of the conflict region area exceeds the preset tolerance, send an enhanced sampling command to the acquisition module.

7. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 6, characterized in that: The specific process for obtaining supplementary feature data to extend the acquisition period is as follows: Extend the data acquisition duration at the location corresponding to the local feature anomaly region identifier to a set multiple of the baseline duration; Extract the periodic fluctuation pattern of pulsation trajectory characteristics within an extended acquisition period.

8. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 7, characterized in that: The specific process for generating the liveness verification result is as follows: Establish a baseline range for the fluctuation amplitude of spectral absorption characteristics within the pulse cycle; Detect the phase shift between spectral absorption fluctuations and pulsation trajectory peaks in supplementary feature data; When the fluctuation amplitude is within the reference range and the phase offset is less than the preset error threshold, it is determined to conform to the physiological coupling law of living organisms. When the fluctuation amplitude is not within the reference range or the phase offset is greater than or equal to the preset error threshold, it is determined to not conform to the physiological coupling law of living organisms.

9. The live fingerprint detection method based on the fusion of multispectral and micro-arterial pulse characteristics according to claim 8, characterized in that: The specific process of outputting the live fingerprint authentication signal is as follows: When three consecutive liveness verification results do not conform to the physiological coupling law of liveness, the anti-spoofing attack alarm protocol is activated. Freeze the current fingerprint authentication process and record attack signature patterns; The attack signature pattern is uploaded to the security certification center database.

Images