Fingerprint image acquisition method based on bio-electricity signal enhancement

Abstract

The invention discloses a fingerprint image acquisition method based on bio-electricity signal enhancement. The method comprises the following steps: S1, constructing an original signal data set; s2, carrying out dynamic modeling, and generating a spatial distribution model and adaptability enhancement parameters of the signals; s3, finger sliding, rotation and pressure changes are monitored, and spatial and temporal distribution data are generated in combination with the spatial distribution model; s4, compensating texture deviation caused by motion according to the space-time distribution data, and generating a signal distribution model; s5, signal contrast is enhanced through nerve-fingerprint feature interaction enhancement and a nerve response synchronous amplification mechanism, and an enhanced signal is generated; s6, constructing a multi-modal collaborative network to optimize texture distribution characteristics, and generating an optimized characteristic signal; and S7, decoding the optimized feature signal by using a spatial-temporal feature decoding technology to generate a final fingerprint image. According to the invention, the multi-source bioelectricity signal fusion and dynamic optimization technology is utilized to realize high-precision fingerprint image acquisition in a complex scene.

Description

Technical Field

[0001] The present invention relates to the technical fields of biometric identification and fingerprint acquisition, and particularly to a fingerprint image acquisition method based on the enhancement of bioelectric signals. Background Art

[0002] Fingerprint recognition technology, as an important part of the biometric identification field, has been widely applied in multiple fields such as identity authentication, security protection, and unlocking of intelligent devices. Traditional fingerprint acquisition technologies mainly rely on optical, electrostatic, or capacitive sensors to generate digital images by obtaining fingerprint surface feature information. However, in complex scenarios, such as wet hands, dry hands, aging skin, low contact pressure, etc., traditional technologies often face the problem of decreased acquisition quality. This quality degradation may manifest as blurred fingerprint images, loss of details, or increased noise, thereby affecting the accuracy and robustness of fingerprint recognition systems.

[0003] Optical fingerprint acquisition technology uses reflected or transmitted light to obtain fingerprint textures. However, under wet hand conditions, the refraction and scattering effects of moisture will significantly reduce the image quality. At the same time, optical technology is sensitive to surface contamination and is easily interfered by oil stains, dust, etc. Electrostatic fingerprint acquisition technology generates texture features by capturing electrostatic signals between the finger and the acquisition surface. However, when the contact pressure is insufficient or the skin surface is too dry, the intensity of the electrostatic signal will be significantly reduced, resulting in incomplete texture features. Capacitive fingerprint acquisition technology constructs fingerprint images by detecting capacitance changes between the skin and the electrodes. However, under special conditions such as aging skin, due to the decrease in skin elasticity and conductivity, the collected capacitance signals may deviate, affecting the clarity of the image.

[0004] In addition, traditional fingerprint acquisition technologies usually only rely on a single signal source and lack the comprehensive utilization of multimodal information. This single signal dependence makes it difficult for the system to adaptively adjust in the face of complex acquisition scenarios. For example, under wet hand conditions, the electrostatic signal may completely fail, and relying solely on the capacitive signal may not be able to capture sufficient detail information. In addition, the signal processing and image generation methods of existing technologies are mostly static processing and cannot dynamically adapt to the real-time changes of signals, resulting in insufficient performance of the system in dealing with scenarios such as dynamic finger movement, pressure fluctuation, or rotation.

[0005] Another important technical limitation lies in the deficiencies of traditional fingerprint acquisition systems in signal enhancement and texture optimization. Existing technologies usually enhance the signal acquisition ability through hardware, but this method is costly and has limited adaptability to complex scenarios. For signal processing, traditional methods mostly adopt fixed rules or simple filtering algorithms, which cannot fully explore the internal correlations of biological signals and lack the ability to deeply model dynamically changing signals. For example, static signal processing methods are difficult to capture the spatio-temporal characteristics of signal distribution during finger movement, resulting in dynamic distortion or blurred boundaries in the generated fingerprint images.

[0006] In terms of signal enhancement and feature optimization, existing technologies usually rely on low-dimensional feature modeling and cannot fully capture the high-dimensional distribution characteristics of multi-source signals and complex spatio-temporal dependence relationships. In addition, for the dynamic changes of signals, existing technologies lack an efficient compensation mechanism. For example, in scenarios with low contact pressure or slight sliding, fingerprint texture features are prone to deformation or offset, and traditional static compensation algorithms are difficult to adjust the signal distribution in real time, resulting in a decline in the quality of the acquired images.

[0007] Therefore, how to provide a fingerprint image acquisition method based on bioelectric signal enhancement is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] An object of the present invention is to propose a fingerprint image acquisition method based on bioelectric signal enhancement. The present invention utilizes multi-source bioelectric signal fusion and dynamic optimization technologies, and through the joint modeling and enhancement of neural signals, electrostatic signals, and capacitance signals, realizes high-precision fingerprint image acquisition in complex scenarios. Through technologies such as dynamic compensation, feature interaction enhancement, multi-modal collaborative optimization, and spatio-temporal decoding, the problem of the decline in fingerprint acquisition quality under conditions such as wet hands, dry hands, aging skin, and low contact pressure is effectively solved. The generated fingerprint images have high resolution, clear texture details, and global consistency, and have the advantages of strong adaptability, good robustness, and high acquisition stability, providing new support for fingerprint recognition technology.

[0009] A fingerprint image acquisition method based on bioelectric signal enhancement according to an embodiment of the present invention includes the following steps:

[0010] S1. Collect multi-source bioelectric signals in the fingerprint contact area, where the multi-source bioelectric signals include neural signals, electrostatic signals, and capacitance signals, and construct an original signal data set;

[0011] S2. Perform dynamic modeling on the original signal data set based on a high-order variational autoencoder to generate a spatial distribution model of the signal and adaptive enhancement parameters;

[0012] S3. Use the contact force sensor and displacement sensing module to monitor the changes in finger sliding, rotation, and contact pressure in real time, and generate spatio-temporal distribution data of finger movement by combining with the spatial distribution model of the signal;

[0013] S4. According to the spatio-temporal distribution data of finger movement, compensate the texture deviation caused by finger movement in real time, and generate a dynamically compensated signal distribution model;

[0014] S5. Through the neural-fingerprint feature interaction enhancement mechanism, perform dynamic correspondence modeling between the dynamically compensated signal distribution model and the fingerprint texture features, and use the neural response synchronous amplification mechanism to enhance the signal contrast to generate an enhanced signal;

[0015] S6. Construct a multi-modal collaborative network jointly optimized by the electrostatic signal and the enhanced signal, optimize the texture distribution characteristics through the generative adversarial network, and generate an optimized feature signal;

[0016] S7. Use spatio-temporal feature decoding technology to decode the optimized feature signal to generate the final fingerprint image.

[0017] Optionally, the S2 specifically includes:

[0018] S21. Decompose the multi-source bioelectric signals in the original signal dataset, and represent the time series of neural signals, electrostatic signals, and capacitance signals as:

[0019]

[0020] where, I k (t) represents the signal intensity of the k-th type of signal at time t, N k represents the number of frequency components of the k-th type of signal, a kn represents the amplitude of the n-th component of the k-th type of signal, f kn represents the frequency of the n-th component of the k-th type of signal, represents the initial phase of the n-th component of the k-th type of signal;

[0021] S22. Perform spectral analysis on I k (t) of each type of signal, and construct a spectral feature matrix Fk:

[0022]

[0023] where, F k (i, j) represents the complex amplitude of the component with frequency f i in the k-th type of signal at time slice j, T represents the sampling time window, f i represents the frequency of the i-th component, and j represents the time slice index;

[0024] S23. For the spectral feature matrix Fk Perform dimensionality reduction and map it into a latent feature matrix:

[0025]

[0026] Among them, Z k represents the latent feature matrix of the k-th type of signal, and V k represents the projection matrix obtained by principal component analysis;

[0027] S24. Construct a dynamic distribution model based on the latent feature matrix:

[0028]

[0029] Among them, p(Z k ) represents the probability distribution of the latent feature matrix, M k represents the number of latent variables, Z ki represents the value of the i-th latent variable, μ ki represents the mean of the i-th latent variable, and σ ki represents the standard deviation of the i-th latent variable;

[0030] S25. Use the dynamic distribution model to map the latent variables into the spatial distribution characteristics of the signal and generate the spatial distribution model of the signal:

[0031]

[0032] Among them, S k (x, y) represents the spatial distribution of the k-th type of signal at the spatial coordinates (x, y), and Φ i (x, y) represents the i-th orthogonal basis function at the spatial coordinates (x, y);

[0033] S26. Extract the adaptive enhancement parameters according to the spatial distribution model of the signal:

[0034] θ k ={μ ki , σ ki , M k};

[0035] Among them, θ k represents the set of enhancement parameters of the k-th type of signal.

[0036] Optionally, the specific steps of S3 include:

[0037] S31. Use the contact force sensor to collect the pressure change on the finger contact surface in real time:

[0038]

[0039] Among them, P(t) represents the contact pressure at time t, F(t) represents the contact force at time t, and A represents the contact area between the finger and the contact surface;

[0040] S32. Real-time monitor the two-dimensional displacement of the finger on the contact surface through the displacement sensing module:

[0041]

[0042] Among them, Δd(t) represents the displacement of the finger on the plane at time t, x(t) and y(t) represent the two-dimensional position coordinates of the finger at time t, x 0 and y 0 represent the initial two-dimensional position coordinates of the finger;

[0043] S33. Monitor the change in the rotation angle of the finger on the contact surface through the rotation sensor:

[0044] Δθ(t) = θ(t) - θ 0 ;

[0045] Among them, Δθ(t) represents the change in the rotation angle of the finger at time t, θ(t) represents the angle value at time t, and θ 0 represents the initial angle;

[0046] S34. Use the spatial distribution model of the signal, combine the changes in finger pressure, displacement, and rotation angle, and generate a preliminary finger motion feature matrix:

[0047] M(t) = [P(t) Δd(t) Δθ(t)] · S(x, y);

[0048] Among them, M(t) represents the finger motion feature matrix at time t, and S(x, y) represents the spatial distribution of the signal at the spatial coordinates (x, y);

[0049] S35. Perform spatio-temporal feature fusion on the finger motion feature matrix to generate spatio-temporal distribution data of finger motion:

[0050]

[0051] Among them, S(t, x, y) represents the spatio-temporal distribution data of finger motion at time t and spatial coordinates (x, y), G(x, y) represents the spatial weight function at spatial coordinates (x, y), t 0 represents the start time of the motion, and t n represents the end time of the motion.

[0052] Optionally, the specific content of S4 includes:

[0053] S41. Obtain the spatio-temporal distribution data S(t, x, y) of finger movement, sample the spatio-temporal distribution data according to time t and spatial coordinates (x, y), and represent the movement offset of the finger at each sampling point as Δx(t, x, y) and Δy(t, x, y);

[0054] S42. Divide the original fingerprint texture mapping into grid regions of a fixed size, and represent the texture data of each grid region with a texture feature matrix;

[0055] S43. Calculate the texture deviation value using the spatio-temporal distribution data of finger movement:

[0056]

[0057] Among them, ΔT(t, x, y) represents the texture deviation value, V(x, y) represents the current texture feature matrix, and V 0 (x, y) represents the initial reference texture feature matrix, represents the gradient of the current texture feature matrix in the x direction, represents the gradient of the current texture feature matrix in the y direction;

[0058] S44. Perform dynamic weight adjustment on the texture deviation value ΔT(t, x, y) to generate a compensation weight matrix for locally optimizing the deviations at different positions. The adjustment process is controlled by the texture gradient and the deviation magnitude:

[0059]

[0060] Among them, W(x, y) represents the dynamic compensation weight at the spatial coordinates (x, y), K represents the texture gradient adjustment coefficient, exp represents the exponential function, represents the gradient magnitude of the current texture feature matrix V(x, y), λ represents the deviation weight adjustment coefficient, and |ΔT(t, x, y)| represents the absolute value of the texture deviation;

[0061] S45. Based on the compensation weight matrix W(x, y) and the texture deviation value ΔT(t, x, y), calculate the dynamically compensated signal distribution model:

[0062] S f (x, y) = S 0 (x, y) + W(x, y)·ΔT(t, x, y);

[0063] Among them, S f (x, y) represents the dynamically compensated signal distribution model, and S 0 (x, y) represents the initial signal distribution model.

[0064] Optionally, the specific content of S5 includes:

[0065] S51. Based on the neural-fingerprint feature interaction mechanism, dynamically model the signal distribution model after dynamic compensation and the fingerprint texture features to generate an interaction enhancement matrix. The neural-fingerprint feature interaction mechanism includes:

[0066] Perform point-by-point correlation analysis on the signal distribution model after dynamic compensation and the fingerprint texture features in space, and introduce spatial weights using the Gaussian kernel function to enhance the influence between adjacent regions;

[0067] Map the local changes of the signal distribution model after dynamic compensation and the fingerprint texture features to the interaction space, and couple multi-scale features together through integral operations;

[0068] Calculate the weighted sum of the signal distribution model after dynamic compensation and the fingerprint texture features within the local neighborhood:

[0069]

[0070] where R(x, y) represents the interaction enhancement matrix at spatial coordinates (x, y), S(u, v) represents the spatial distribution of the signal at spatial coordinates (u, v), T(x - u, y - v) represents the distribution of the fingerprint texture feature values at the offset (x - u, y - v), exp represents the exponential function, and σ represents the scale parameter of the Gaussian kernel function;

[0071] S52. Introduce a neural response synchronous amplification mechanism to enhance the signal contrast, and dynamically couple the intensity distribution of the neural signal with the interaction enhancement matrix to generate an enhanced signal:

[0072]

[0073] where E(x, y) represents the enhanced signal at spatial coordinates (x, y), η represents the neural response amplification coefficient, N(x, y) represents the intensity distribution of the neural signal at spatial coordinates (x, y), and max(N(x, y)) represents the maximum value of the neural signal intensity distribution.

[0074] Optionally, the S6 specifically includes:

[0075] S61. Obtain the electrostatic signal and the enhanced signal, and perform timing alignment and normalization processing on the electrostatic signal and the enhanced signal;

[0076] S62. Construct a multi-modal collaborative network, which consists of two parts: an electrostatic signal feature extraction branch and an enhanced signal feature extraction branch. The electrostatic signal feature extraction branch extracts the local intensity distribution features of the electrostatic signal through convolution operations, and the enhanced signal feature extraction branch extracts the global features of the enhanced signal and generates a joint feature map through feature fusion operations;

[0077] S63. Input the joint feature map into a generative adversarial network, which includes a generator and a discriminator. The generator generates an optimized texture distribution characteristic based on the joint feature map. By learning the internal relationship between the electrostatic signal and the enhanced signal, an optimized feature signal is generated. The discriminator is used to evaluate the similarity between the texture distribution characteristic output by the generator and the actual texture feature, and guide the generator to optimize.

[0078] S64. Finally, output the optimized feature signal.

[0079] Optionally, S7 specifically includes:

[0080] S71. Use spatio-temporal feature decoding technology to decode the optimized feature signal. The decoding process includes time decoding and space decoding. The time decoding is used to extract the dynamic feature of the optimized feature signal in the time series and generate a time characteristic distribution. The space decoding is used to extract the distribution characteristic of the optimized feature signal at different spatial positions and generate a spatial characteristic mapping.

[0081] S72. Jointly process the time characteristic distribution and the spatial characteristic mapping to form a spatio-temporal feature mapping. The joint processing includes feature alignment and weight balance.

[0082] S73. Reconstruct the decoded spatio-temporal feature mapping to generate the final fingerprint image. The reconstruction process includes noise suppression, texture enhancement, and boundary correction.

[0083] The beneficial effects of the present invention are as follows:

[0084] First of all, by fusing multi-source bioelectrical signals, including nerve signals, electrostatic signals, and capacitance signals, the present invention solves the problem of single-signal dependence in the prior art. In complex acquisition scenarios, even if a certain signal source is restricted due to wet hands, dry hands, or skin state changes, other signal sources can still provide supplementary information to ensure the integrity and stability of fingerprint image acquisition. In addition, by dynamically modeling multi-source signals through a high-order variational autoencoder, the present invention can generate an accurate signal space distribution model and adaptive enhancement parameters, so as to achieve efficient fusion and feature extraction of multi-modal signals, enabling the signals to accurately reflect the spatial characteristics of fingerprint textures.

[0085] Secondly, the present invention designs a dynamic compensation mechanism, which can correct the texture deviation caused by finger sliding, rotation, and pressure changes in real time, and generate a signal distribution model after dynamic compensation. This mechanism is particularly suitable for scenarios such as low contact pressure or slight sliding, effectively avoiding the problems of image blurring and texture distortion caused by finger movement in traditional methods. By analyzing and compensating the spatio-temporal distribution data of finger movement, the present invention ensures the stability and consistency of fingerprint images under dynamic acquisition conditions.

[0086] In addition, the neural-fingerprint feature interaction enhancement mechanism of the present invention dynamically enhances the signal contrast of key regions through the neural response synchronization amplification technology. This mechanism can not only improve the feature clarity in complex texture regions but also enhance the contrast effect of fingerprint textures under low signal intensity conditions, thereby providing richer and more accurate feature data for high-precision fingerprint recognition. Especially in complex scenarios such as wet hands and aging skin, this mechanism significantly improves the signal quality and image clarity.

[0087] Furthermore, the present invention constructs a multi-modal collaborative network, which further improves the integrity and consistency of the fingerprint texture distribution through the joint optimization of electrostatic signals and enhanced signals. The application of the generative adversarial network in texture optimization enables the present invention to learn the deep association between electrostatic signals and enhanced signals and generate optimized feature signals with high resolution and strong contrast. Through this optimization, the present invention solves the problem of decreased recognition accuracy caused by incomplete textures or noise interference in traditional methods.

[0088] Finally, the present invention decodes the optimized feature signals into a unified fingerprint image through spatio-temporal feature decoding technology. The decoding process combines the Gaussian-multiple sample reconstruction model to further optimize the texture details and global distribution characteristics of the fingerprint image. The generated fingerprint image has the characteristics of high resolution, clear details, and global consistency, can accurately reflect the spatial texture distribution of the fingerprint, and meets the high-quality acquisition requirements in complex application scenarios. Description of the Drawings

[0089] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0090] Figure 1 is the overall flowchart of a fingerprint image acquisition method based on bioelectric signal enhancement proposed by the present invention;

[0091] Figure 2 is the flowchart of the spatio-temporal distribution data generation and dynamic compensation mechanism of finger movement for a fingerprint image acquisition method based on bioelectric signal enhancement proposed by the present invention. Detailed Embodiments

[0092] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0093] Refer to Figure 1 and Figure 2 , a fingerprint image acquisition method based on bioelectric signal enhancement includes the following steps:

[0094] S1. Collect multi-source bioelectrical signals in the fingerprint contact area. The multi-source bioelectrical signals include neural signals, electrostatic signals, and capacitance signals, and construct an original signal dataset;

[0095] S2. Based on a high-order variational autoencoder, perform dynamic modeling on the original signal dataset to generate a spatial distribution model of the signals and adaptive enhancement parameters;

[0096] S3. Use a contact force sensor and a displacement sensing module to monitor the sliding, rotation, and contact pressure changes of the finger in real time, and combine the spatial distribution model of the signals to generate spatio-temporal distribution data of the finger movement;

[0097] S4. According to the spatio-temporal distribution data of the finger movement, compensate the texture deviation caused by the finger movement in real time to generate a dynamically compensated signal distribution model;

[0098] S5. Through a neural-fingerprint feature interaction enhancement mechanism, perform dynamic correspondence modeling between the dynamically compensated signal distribution model and the fingerprint texture features, and use a neural response synchronization amplification mechanism to enhance the signal contrast to generate enhanced signals;

[0099] S6. Construct a multi-modal collaborative network jointly optimized by electrostatic signals and enhanced signals, and optimize the texture distribution characteristics through a generative adversarial network to generate optimized feature signals;

[0100] S7. Adopt spatio-temporal feature decoding technology to decode the optimized feature signals to generate the final fingerprint image.

[0101] In this embodiment, the specific content of S2 includes:

[0102] S21. Decompose the multi-source bioelectrical signals in the original signal dataset, and represent the time series of neural signals, electrostatic signals, and capacitance signals as:

[0103]

[0104] where I k (t) represents the signal intensity of the k-th type of signal at time t, N k represents the number of frequency components of the k-th type of signal, a kn represents the amplitude of the n-th component of the k-th type of signal, f kn represents the frequency of the n-th component of the k-th type of signal, represents the initial phase of the n-th component of the k-th type of signal;

[0105] S22. Perform spectral analysis on I k (t) of each type of signal to construct a spectral feature matrix Fk:

[0106]

[0107] Among them, F k (i, j) represents the complex amplitude of the component with frequency f in the k-th type of signal at time slice j. T represents the sampling time window, f i represents the frequency of the i-th component, and j represents the time slice index; i

[0108] S23. Perform dimensionality reduction on the spectral feature matrix F k and map it to a latent feature matrix:

[0109]

[0110] Among them, Z k represents the latent feature matrix of the k-th type of signal, and V k represents the projection matrix obtained by principal component analysis;

[0111] S24. Construct a dynamic distribution model based on the latent feature matrix:

[0112]

[0113] Among them, p(Z k ) represents the probability distribution of the latent feature matrix, M k represents the number of latent variables, Z ki represents the value of the i-th latent variable, μ ki represents the mean of the i-th latent variable, and σ ki represents the standard deviation of the i-th latent variable;

[0114] S25. Use the dynamic distribution model to map the latent variables to the spatial distribution characteristics of the signal and generate a spatial distribution model of the signal:

[0115]

[0116] Among them, S k (x, y) represents the spatial distribution of the k-th type of signal at spatial coordinates (x, y), and Φ i (x, y) represents the i-th orthogonal basis function at spatial coordinates (x, y);

[0117] S26. Extract the adaptive enhancement parameters according to the spatial distribution model of the signal:

[0118] θ k ={μ ki , σ ki , M k};

[0119] Among them, θ k represents the enhancement parameter set of the k-th type of signal.​

[0120] In this embodiment, step S3 specifically includes:

[0121] S31. Using a contact force sensor to collect the pressure change on the finger contact surface in real time:

[0122]

[0123] Among them, P(t) represents the contact pressure at time t, F(t) represents the contact force at time t, and A represents the contact area between the finger and the contact surface;

[0124] S32. Using a displacement sensing module to monitor the two-dimensional displacement of the finger on the contact surface in real time:

[0125]

[0126] Among them, Δd(t) represents the displacement of the finger on the plane at time t, x(t) and y(t) represent the two-dimensional position coordinates of the finger at time t, and x 0 and y 0 represent the initial two-dimensional position coordinates of the finger;

[0127] S33. Using a rotation sensor to monitor the change in the rotation angle of the finger on the contact surface:

[0128] Δθ(t) = θ(t) - θ 0 ;

[0129] Among them, Δθ(t) represents the change in the rotation angle of the finger at time t, θ(t) represents the angle value at time t, and θ 0 represents the initial angle;

[0130] S34. Using the spatial distribution model of the signal, and combining the changes in finger pressure, displacement, and rotation angle, to generate a preliminary finger motion feature matrix:

[0131] M(t) = [P(t) Δd(t) Δθ(t)] · S(x, y);

[0132] Among them, M(t) represents the finger motion feature matrix at time t, and S(x, y) represents the spatial distribution of the signal at the spatial coordinates (x, y);

[0133] S35. Performing spatio-temporal feature fusion on the finger motion feature matrix to generate spatio-temporal distribution data of finger motion:

[0134]

[0135] Among them, S(t, x, y) represents the spatio-temporal distribution data of finger movement at time t and spatial coordinates (x, y), G(x, y) represents the spatial weight function at spatial coordinates (x, y), and t 0 represents the starting time of the movement, and t n represents the ending time of the movement.

[0136] In this embodiment, the S4 specifically includes:

[0137] S41. Obtain the spatio-temporal distribution data S(t, x, y) of finger movement, sample the spatio-temporal distribution data according to time t and spatial coordinates (x, y), and represent the movement offset of the finger at each sampling point as Δx(t, x, y) and Δy(t, x, y);

[0138] S42. Map and divide the original fingerprint texture into grid regions of a fixed size, and represent the texture data of each grid region with a texture feature matrix;

[0139] S43. Calculate the texture deviation value using the spatio-temporal distribution data of finger movement:

[0140]

[0141] Among them, ΔT(t, x, y) represents the texture deviation value, V(x, y) represents the current texture feature matrix, and V0(x, y) represents the initial reference texture feature matrix. represents the gradient of the current texture feature matrix in the x direction, represents the gradient of the current texture feature matrix in the y direction;

[0142] S44. Perform dynamic weight adjustment on the texture deviation value ΔT(t, x, y) to generate a compensation weight matrix for locally optimizing the deviations at different positions, and the adjustment process is controlled by the texture gradient and the deviation magnitude:

[0143]

[0144] Among them, W(x, y) represents the dynamic compensation weight at spatial coordinates (x, y), K represents the texture gradient adjustment coefficient, exp represents the exponential function, represents the gradient magnitude of the current texture feature matrix V(x, y), λ represents the deviation weight adjustment coefficient, and |ΔT(t, x, y)| represents the absolute value of the texture deviation;

[0145] S45. Based on the compensation weight matrix W(x, y) and the texture deviation value ΔT(t, x, y), calculate the dynamically compensated signal distribution model:

[0146] S f (x, y) = S 0(x, y) + W(x, y)·ΔT(t, x, y);

[0147] Wherein, S f (x, y) represents the signal distribution model after dynamic compensation, and S 0 (x, y) represents the initial signal distribution model.

[0148] In this embodiment, the S5 specifically includes:

[0149] S51. Based on the neural-fingerprint feature interaction mechanism, perform dynamic modeling on the signal distribution model after dynamic compensation and the fingerprint texture features to generate an interaction enhancement matrix. The neural-fingerprint feature interaction mechanism includes:

[0150] Perform point-by-point correlation analysis in space on the signal distribution model after dynamic compensation and the fingerprint texture features, and introduce spatial weights using the Gaussian kernel function to enhance the influence between adjacent regions;

[0151] Map the local changes of the signal distribution model after dynamic compensation and the fingerprint texture features to the interaction space, and couple multi-scale features together through integral operations;

[0152] Calculate the weighted sum of the signal distribution model after dynamic compensation and the fingerprint texture features within the local neighborhood:

[0153]

[0154] Wherein, R(x, y) represents the interaction enhancement matrix at the spatial coordinates (x, y), S(u, v) represents the spatial distribution of the signal at the spatial coordinates (u, v), T(x - u, y - v) represents the distribution of the fingerprint texture feature values at the offset (x - u, y - v), exp represents the exponential function, and σ represents the scale parameter of the Gaussian kernel function;

[0155] S52. Introduce a neural response synchronous amplification mechanism to enhance the signal contrast, and dynamically couple the intensity distribution of the neural signal with the interaction enhancement matrix to generate an enhanced signal:

[0156]

[0157] Wherein, E(x, y) represents the enhanced signal at the spatial coordinates (x, y), η represents the neural response amplification coefficient, N(x, y) represents the intensity distribution of the neural signal at the spatial coordinates (x, y), and max(N(x, y)) represents the maximum value of the neural signal intensity distribution.

[0158] In this embodiment, the S6 specifically includes:

[0159] S61. Obtain the electrostatic signal and the enhanced signal, and perform timing alignment and normalization processing on the electrostatic signal and the enhanced signal;

[0160] S62. Construct a multi-modal collaborative network, which consists of two parts: an electrostatic signal feature extraction branch and an enhanced signal feature extraction branch. The electrostatic signal feature extraction branch extracts the local intensity distribution features of the electrostatic signal through convolution operations, and the enhanced signal feature extraction branch extracts the global features of the enhanced signal and generates a joint feature map through feature fusion operations;

[0161] S63. Input the joint feature map into a generative adversarial network, which includes a generator and a discriminator. The generator generates an optimized texture distribution characteristic according to the joint feature map, and generates an optimized feature signal by learning the internal relationship between the electrostatic signal and the enhanced signal; the discriminator is used to evaluate the similarity between the texture distribution characteristic output by the generator and the actual texture feature, and guide the generator to optimize;

[0162] S64. Finally, output the optimized feature signal.

[0163] In this embodiment, the specific steps of S7 are as follows:

[0164] S71. Use spatio-temporal feature decoding technology to decode the optimized feature signal. The decoding process includes time decoding and space decoding; the time decoding is used to extract the dynamic features of the optimized feature signal in the time series and generate a time characteristic distribution; the space decoding is used to extract the distribution characteristics of the optimized feature signal at different spatial positions and generate a spatial characteristic mapping;

[0165] S72. Jointly process the time characteristic distribution and the spatial characteristic mapping to form a spatio-temporal feature mapping. The joint processing includes feature alignment and weight balance;

[0166] S73. Reconstruct the decoded spatio-temporal feature mapping to generate the final fingerprint image. The reconstruction process includes noise suppression, texture enhancement, and boundary correction.

[0167] Example 1:

[0168] In order to verify the feasibility of the present invention in practice, the present invention is applied to an identity verification system. The test scenario is set in a security check channel of a high-traffic airport, simulating different finger states and environmental conditions, and evaluating the acquisition accuracy, stability, and adaptability of the system.

[0169] During the test, 200 volunteers were selected as the test subjects. The age distribution of the volunteers was from 20 to 60 years old, and the finger states included wet hands, dry hands, aging skin, and normal state. The test device used a customized acquisition terminal that supported the method of the present invention. This terminal was equipped with a multi-source bioelectric signal acquisition module, a high-order variational autoencoder, a dynamic compensation mechanism, a neural-fingerprint feature interaction enhancement module, and a multi-modal collaborative optimization network.

[0170] During the specific test process, the volunteers were required to collect fingerprints in different finger states, including directly collecting after wet hands, collecting after drying with a tissue, collecting after contacting alcohol, and collecting in the natural state. During the collection process, the clarity of the fingerprint image, the texture integrity, and the system response time were recorded in real time. At the same time, the usability and accuracy of the image were evaluated through subsequent identity comparison.

[0171] To fully verify the superiority of the present invention, the test was also compared with existing optical fingerprint acquisition technology and single electrostatic signal acquisition technology to evaluate the acquisition quality and robustness under the same conditions.

[0172] Under the condition of wet hands, the image clarity of the method of the present invention reached 97%, which was much higher than 72% of the optical technology and 65% of the electrostatic technology. This was because the present invention effectively offset the interference of wet hands on the electrostatic signal through multi-source signal fusion and a dynamic compensation mechanism, and optimized the texture feature contrast through neural-fingerprint feature interaction enhancement technology. Under the condition of dry hands, the texture integrity index of the present invention was 96%, while those of the traditional optical technology and electrostatic technology were 78% and 68% respectively. This result indicated that the adaptive enhancement parameters of the present invention could significantly improve the signal quality in the state of dry skin. Under the condition of low contact pressure, the identity comparison accuracy rate of the present invention still remained at 98%, while those of the traditional methods were 70% and 64% respectively. In addition, the average response time of the present invention was 0.8 seconds, which was reduced by about 35% compared with the traditional method.

[0173] Through data verification, the fingerprint image acquisition quality of the present invention in complex scenarios was significantly better than that of the existing technology, which could effectively solve the problem of the decline in acquisition quality under special conditions such as wet hands, dry hands, and aging skin, and had strong real-time performance and stability.

[0174] Table 1 Comparative analysis table of fingerprint acquisition effects in complex scenarios

[0175]

[0176] As can be seen from Table 1 above, the fingerprint image acquisition performance of the present invention in complex scenarios is significantly better than that of traditional optical technology and electrostatic technology. Under wet hand conditions, the image clarity of the present invention reaches 97%, significantly higher than 72% of optical technology and 65% of electrostatic technology. This is due to the fact that the present invention effectively eliminates the interference to electrostatic signals under wet hand conditions and enhances the clarity and contrast of fingerprint textures through the multi-source bioelectric signal fusion technology, combined with the dynamic compensation and neural-fingerprint feature interaction enhancement mechanism.

[0177] In the dry hand scenario, the texture integrity of the present invention is 96%, which is also significantly better than 78% of optical technology and 68% of electrostatic technology. This is because traditional technologies usually cannot capture sufficient signal details in the dry skin state, while the present invention ensures the signal quality and texture stability under dry skin conditions through the dynamic adjustment of adaptive enhancement parameters and the optimization of the multi-modal collaborative network.

[0178] For the test of aging skin, the image clarity and texture integrity of the present invention are 95% and 94% respectively, while those of optical technology are only 75% and 73%, and those of electrostatic technology are even lower, only 63% and 61%. This result shows that when dealing with the situation of decreased skin conductivity or blurred texture, the present invention can effectively capture texture details through the deep modeling of multi-source signals and the optimization of high-order distributions, ensuring that the collected fingerprint images are clear and complete.

[0179] In the low contact pressure scenario, the identity comparison accuracy rate of the present invention is as high as 98%, while those of optical technology and electrostatic technology are 70% and 64% respectively. Traditional methods are highly sensitive to contact pressure and are prone to insufficient signal strength, resulting in the inability to generate high-quality images. However, the present invention uses the dynamic compensation mechanism and texture deviation correction function to ensure the integrity and contrast of signals even under low contact pressure.

[0180] For the test of the normal state, the present invention is superior to traditional methods in all indicators. The image clarity and texture integrity reach 98% and 97% respectively, and the identity comparison accuracy rate is as high as 99%. At the same time, the response time is only 0.7 seconds, significantly faster than 1.1 seconds of optical technology and 1.3 seconds of electrostatic technology. This fully demonstrates the real-time performance and high efficiency of the present invention.

[0181] Generally speaking, the present invention shows excellent adaptability and robustness in wet hands, dry hands, aging skin, low contact pressure, and normal states. It effectively solves the problem of the decline in acquisition quality of traditional technologies in complex scenarios, and at the same time has a higher response speed and recognition efficiency, providing an innovative direction and practical value for the development of fingerprint image acquisition technology.

[0182] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes should be covered within the protection scope of the present invention.

Claims

1. A fingerprint image acquisition method based on bioelectric signal enhancement, characterized in that: The steps include: S1. Collect multi-source bioelectric signals of the fingerprint contact area, wherein the multi-source bioelectric signals include neural signals, electrostatic signals and capacitive signals, and construct an original signal data set; S2, dynamically modeling the original signal data set based on high-order variational autoencoders to generate the spatial distribution model and adaptive enhancement parameters of the signal; S3, using the contact force sensor and displacement sensing module to monitor the sliding, rotation and contact pressure changes of the finger in real time, and combining the spatial distribution model of the signal to generate the spatiotemporal distribution data of the finger movement; S4, compensating the texture deviation caused by the finger movement in real time according to the spatiotemporal distribution data of the finger movement, and generating a signal distribution model after dynamic compensation; S5. Through the neural-fingerprint feature interactive enhancement mechanism, the signal distribution model after dynamic compensation is dynamically modeled with the fingerprint texture feature, and the neural response synchronous amplification mechanism is used to enhance the signal contrast and generate an enhanced signal; S6. Construct a multimodal collaborative network for joint optimization of electrostatic signals and enhanced signals, optimize texture distribution characteristics through generative adversarial networks, and generate optimized feature signals; S7. Use spatiotemporal feature decoding technology to decode the optimized feature signal to generate a final fingerprint image.

2. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1 is characterized in that: The S2 specifically includes: S21. Decompose the multi-source bioelectric signals in the original signal data set, and express the time series of neural signals, electrostatic signals and capacitive signals as: Among them, I k (t) represents the signal strength of the k-th signal at time t, N k represents the number of frequency components of the k-th signal, a kn represents the amplitude of the nth component of the kth signal, f kn represents the frequency of the nth component of the kth signal, Represents the initial phase of the nth component of the kth type signal; S22, I for each type of signal k (t) Perform spectrum analysis and construct the spectrum feature matrix F k : Among them, F k (i,j) indicates that the frequency of the k-th signal is f i The complex amplitude of the component of at time slice j, T represents the sampling time window, f i represents the frequency of the i-th component, and j represents the time slice index; S23, spectrum feature matrix F k Perform dimensionality reduction and map it into a potential feature matrix: Among them, Z k represents the potential feature matrix of the k-th type of signal, V k represents the projection matrix obtained by principal component analysis; S24. Constructing a dynamic distribution model based on the potential feature matrix: Among them, p(Z k ) represents the probability distribution of the potential feature matrix, M k represents the number of latent variables, Z ki represents the value of the i-th latent variable, μ ki represents the mean of the i-th latent variable, σ ki represents the standard deviation of the i-th latent variable; S25. Using the dynamic distribution model, the latent variables are mapped to the spatial distribution characteristics of the signal to generate the spatial distribution model of the signal: Among them, S k (x,y) represents the spatial distribution of the k-th signal at the spatial coordinate (x,y), Φ i (x,y) represents the i-th orthogonal basis function at the spatial coordinate (x,y); S26. Extracting adaptive enhancement parameters according to the spatial distribution model of the signal: i k ={μ ki ,s ki ,M k }; Among them, θ k Represents the set of enhancement parameters for the k-th type of signal.

3. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1 is characterized in that: The S3 specifically includes: S31. Using the contact force sensor to collect the pressure change of the finger contact surface in real time: Wherein, P(t) represents the contact pressure at time t, F(t) represents the contact force at time t, and A represents the contact area between the finger and the contact surface; S32, real-time monitoring of the two-dimensional displacement of the finger on the contact surface through the displacement sensing module: Wherein, Δd(t) represents the displacement of the finger on the plane at time t, x(t) and y(t) represent the two-dimensional position coordinates of the finger at time t, and x0 and y0 represent the initial two-dimensional position coordinates of the finger; S33, monitoring the rotation angle change of the finger on the contact surface through the rotation sensor: Δθ(t)=θ(t)-θ0; Where Δθ(t) represents the rotation angle change of the finger at time t, θ(t) represents the angle value at time t, and θ0 represents the initial angle; S34. Generate a preliminary finger motion feature matrix by using the spatial distribution model of the signal and combining the changes in finger pressure, displacement and rotation angle: M(t)=[P(t) Δd(t) Δθ(t)]·S(x,y); Where M(t) represents the finger motion feature matrix at time t, and S(x,y) represents the spatial distribution of the signal at the spatial coordinate (x,y); S35, performing spatiotemporal feature fusion on the finger motion feature matrix to generate spatiotemporal distribution data of the finger motion: Among them, S(t,x,y) represents the spatiotemporal distribution data of the finger movement at time t and the spatial coordinates (x,y), G(x,y) represents the spatial weight function on the spatial coordinates (x,y), t0 represents the start time of the movement, t n Indicates the end time of the exercise.

4. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1 is characterized in that: The S4 specifically includes: S41, obtaining the spatiotemporal distribution data S(t,x,y) of the finger movement, sampling the spatiotemporal distribution data according to time t and spatial coordinates (x,y), and expressing the movement offset of the finger at each sampling point as Δx(t,x,y) and Δy(t,x,y); S42, dividing the original fingerprint texture mapping into grid areas of fixed size, and the texture data of each grid area is represented by a texture feature matrix; S43, using the spatiotemporal distribution data of the finger movement to calculate the texture deviation value: Among them, ΔT(t,x,y) represents the texture deviation value, V(x,y) represents the current texture feature matrix, and V0(x,y) represents the initial reference texture feature matrix. Represents the gradient of the current texture feature matrix in the x direction, Represents the gradient of the current texture feature matrix in the y direction; S44, dynamically adjust the weight of the texture deviation value ΔT(t,x,y) to generate a compensation weight matrix for local optimization of the deviations at different positions. The adjustment process is controlled by the texture gradient and the deviation size: Among them, W(x,y) represents the dynamic compensation weight on the spatial coordinate (x,y), κ represents the texture gradient adjustment coefficient, and exp represents the exponential function. represents the gradient size of the current texture feature matrix V(x,y), λ represents the deviation weight adjustment coefficient, and |ΔT(t,x,y)| represents the absolute value of the texture deviation; S45. Based on the compensation weight matrix W(x, y) and the texture deviation value ΔT(t, x, y), the signal distribution model after dynamic compensation is calculated: S f (x,y)=S0(x,y)+W(x,y)·ΔT(t,x,y); Among them, S f (x, y) represents the signal distribution model after dynamic compensation, and S0(x, y) represents the initial signal distribution model.

5. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1, characterized in that: The S5 specifically includes: S51. Based on the neural-fingerprint feature interaction mechanism, dynamically model the signal distribution model and fingerprint texture features after dynamic compensation to generate an interactive enhancement matrix. The neural-fingerprint feature interaction mechanism includes: The signal distribution model after dynamic compensation and fingerprint texture features are subjected to point-by-point correlation analysis in space, and the Gaussian kernel function is used to introduce spatial weights to enhance the influence between adjacent areas. The signal distribution model after dynamic compensation and the local changes of fingerprint texture features are mapped to the interaction space, and the multi-scale features are coupled together through the integration operation; Calculate the weighted sum of the signal distribution model and fingerprint texture features after dynamic compensation in the local neighborhood: Where R(x,y) represents the interaction enhancement matrix at the spatial coordinates (x,y), S(u,v) represents the spatial distribution of the signal at the spatial coordinates (u,v), T(xu,yv) represents the distribution of the fingerprint texture feature value at the offset (xu,yv), exp represents the exponential function, and σ represents the scale parameter of the Gaussian kernel function; S52. Introduce a neural response synchronous amplification mechanism to enhance signal contrast, dynamically couple the intensity distribution of neural signals with the interactive enhancement matrix, and generate an enhanced signal: Among them, E(x,y) represents the enhanced signal at the spatial coordinates (x,y), η represents the neural response amplification factor, N(x,y) represents the neural signal intensity distribution at the spatial coordinates (x,y), and max(N(x,y)) represents the maximum value of the neural signal intensity distribution.

6. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1, characterized in that: The S6 specifically includes: S61, acquiring an electrostatic signal and an enhanced signal, and performing time alignment and normalization processing on the electrostatic signal and the enhanced signal; S62, constructing a multimodal collaborative network, the multimodal collaborative network consisting of two parts: an electrostatic signal feature extraction branch and an enhanced signal feature extraction branch, the electrostatic signal feature extraction branch extracts local intensity distribution features of the electrostatic signal through a convolution operation, the enhanced signal feature extraction branch extracts global features of the enhanced signal, and generates a joint feature map through a feature fusion operation; S63, inputting the joint feature map into a generative adversarial network, the generative adversarial network includes a generator and a discriminator, the generator generates an optimized texture distribution characteristic according to the joint feature map, and generates an optimized feature signal by learning the intrinsic relationship between the electrostatic signal and the enhanced signal; the discriminator is used to evaluate the similarity between the texture distribution characteristic output by the generator and the actual texture characteristic, and guide the generator to optimize; S64. Finally, the optimized characteristic signal is output.

7. The fingerprint image acquisition method based on bioelectric signal enhancement according to claim 1, characterized in that: The S7 specifically includes: S71, using the spatiotemporal feature decoding technology to decode the optimized feature signal, the decoding process includes time decoding and space decoding; the time decoding is used to extract the dynamic characteristics of the optimized feature signal in the time series and generate the time characteristic distribution; the space decoding is used to extract the distribution characteristics of the optimized feature signal in different spatial positions and generate the space characteristic map; S72, jointly processing the temporal characteristic distribution and the spatial characteristic mapping to form a temporal and spatial characteristic mapping, wherein the joint processing includes feature alignment and weight balancing; S73, reconstructing the decoded spatiotemporal feature map to generate a final fingerprint image, wherein the reconstruction process includes noise suppression, texture enhancement and boundary correction.

Images