A biological multimodal feature recognition method based on dynamic fuzzy reasoning

Abstract

The present invention proposes a biometric multimodal feature recognition method and device based on dynamic fuzzy reasoning. The method comprises: preprocessing a target object's facial, fingerprint, and iris images; performing feature extraction on the preprocessed facial, fingerprint, and iris images; and determining the three-dimensional Euclidean distances between the target object and an object to be verified in a database based on the feature extraction results. Fuzzy reasoning is performed based on the three-dimensional Euclidean distances of the face, fingerprint, and iris to obtain a recognition score. The recognition score is compared with a preset decision threshold. If the recognition score is greater than or equal to the decision threshold, the target object is considered a match; otherwise, it is considered a mismatch. The decision threshold is trained based on a preset target recognition accuracy using a reward / penalty mechanism. The present invention can effectively cope with various environmental changes, enhance recognition accuracy, and is suitable for applications requiring extremely high security.

Description

Technical Field

[0001] The present invention relates to the field of biometric identification technology, and in particular to a method, device and computer-readable storage medium for biometric multimodal feature recognition based on dynamic fuzzy reasoning. Background Art

[0002] In the information age, biometric technology, due to its unique security and convenience, is widely used in fields such as access control, financial services, and public security. The security of identity authentication systems is crucial in critical infrastructure such as the State Grid's ultra-high voltage converter stations. As a core component of the power transmission system, the security management of converter stations directly impacts the stability and reliability of power supply. Traditional identity authentication technologies, such as passwords and physical ID cards, are increasingly unable to meet high security requirements, especially in the face of internal security threats and external attacks.

[0003] After searching, the Chinese patent publication number is CN107507286A, a dual-modal biometric sign-in system based on face and handwritten signature, which performs image acquisition, preprocessing, fusion feature extraction and recognition judgment of face and handwritten signature through the identity information collection and detection module, information preprocessing module, convolutional neural network module, feature normalization and fusion module and recognition module connected in sequence.

[0004] The above-mentioned identity recognition method uses dual-modal recognition, but faces multiple challenges in high-security environments such as converter stations. It is easily affected by environmental factors such as dust, pollution or moisture, resulting in unstable recognition accuracy. In addition, facial and handwritten signature recognition systems are more vulnerable to counterfeit attacks, so there is room for improvement. Summary of the Invention

[0005] The purpose of the present invention is to address the shortcomings of the prior art and to propose a biological multimodal feature recognition method, device and computer-readable storage medium based on dynamic fuzzy reasoning, which has the advantage of enhancing the accuracy of the recognition system and effectively reducing the misrecognition rate.

[0006] According to a first aspect of the present invention, a method for biometric multimodal feature recognition based on dynamic fuzzy reasoning is provided. The method comprises the following steps:

[0007] Preprocessing the acquired face image, fingerprint image and iris image of the target object respectively;

[0008] Perform feature extraction on the pre-processed face image, fingerprint image and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint and iris between the target object and the object to be verified in the database according to the feature extraction results;

[0009] Fuzzy reasoning is performed based on the three-dimensional Euclidean distance between face, fingerprint and iris to obtain the recognition score;

[0010] The recognition score is compared with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched; wherein the decision threshold is trained based on the preset target recognition accuracy using a reward / penalty mechanism.

[0011] Furthermore, the acquired face image, fingerprint image and iris image of the target object are preprocessed respectively, including:

[0012] A multi-task convolutional neural network is used to detect facial regions in the acquired face image; the face is aligned and cropped based on the detected face boundaries and key points to obtain a preprocessed face image;

[0013] Adjusting the grayscale distribution of the acquired fingerprint image and removing noise by applying a median filter using an N*N window; using Gabor filters of multiple directions and multiple frequencies to enhance the ridges and valleys in the denoised fingerprint image based on the local direction and frequency characteristics of the fingerprint ridges, thereby obtaining a preprocessed fingerprint image; wherein the multiple directions are multiple directions covering the fingerprint, and the multiple frequencies are adjusted based on the average ridge spacing of the fingerprint;

[0014] The acquired iris image is converted into a grayscale image, and the image noise is removed by Gaussian filtering. The iris boundary of the denoised iris image is located using Hough transform, and a circular iris region is cropped out. Polar coordinate transformation is applied to the cropped circular iris region to convert the circular iris region into a rectangular form in a Cartesian coordinate system to obtain a normalized iris image. Histogram equalization is performed on the normalized iris image to obtain a preprocessed iris image.

[0015] Furthermore, feature extraction is performed on the pre-processed face image, fingerprint image, and iris image, and the three-dimensional Euclidean distances of the face, fingerprint, and iris between the target object and the object to be verified in the database are determined based on the feature extraction results, including:

[0016] The pre-processed face image is input into the pre-trained FaceNet model to obtain a multi-dimensional face feature vector representing the visual features of the face and perform L2 normalization.

[0017] The preprocessed fingerprint image is input into the NLNet model to obtain the fingerprint feature vector representing the fingerprint feature of each dimension and perform L2 normalization;

[0018] Using the forward propagation process of the pre-trained DeepIrisNet model, the iris feature vector is extracted from the pre-processed iris image and L2 normalized;

[0019] Calculate the Euclidean distance between the normalized face, fingerprint and iris feature vectors and the face, fingerprint and iris feature vectors of the object to be verified respectively.

[0020] Furthermore, fuzzy reasoning is performed based on the three-dimensional Euclidean distance of the face, fingerprint, and iris to obtain the recognition score, including:

[0021] Construct a fuzzy set {L, M, H}, where L, M, and H are three fuzzy subsets of the fuzzy set, corresponding to the three matching degrees of "low matching degree", "average matching degree", and "high matching degree" respectively;

[0022] A membership function is constructed by combining a dynamic exponential decreasing function and a Gaussian function, and the membership of the three-dimensional Euclidean distance of the face, fingerprint and iris to each fuzzy subset is determined according to the membership function;

[0023] Construct a set of fuzzy rules to map the matching degree of face, fingerprint and iris to the corresponding overall matching degree;

[0024] determining a triggering weight of each fuzzy rule in the set of fuzzy rules based on the membership of each fuzzy subset based on the three-dimensional Euclidean distance of the face, fingerprint and iris;

[0025] The overall matching degree corresponding to each fuzzy rule is quantified and weighted averaged in combination with the trigger weight to obtain the recognition score.

[0026] Furthermore, the dynamic exponential decreasing function and the Gaussian function are combined to construct a membership function, including expressing the membership function as:

[0027]

[0028] Where x is the input variable, which is the Euclidean distance d1, d2 or d3 of the face, fingerprint or iris; is the weight coefficient; is the center value of the Gaussian function; is the standard deviation, which is used to control the width of the membership function; is a positive constant of proportionality.

[0029] Furthermore, based on the membership of each fuzzy subset of the three-dimensional Euclidean distance of the face, fingerprint, and iris, a triggering weight of each fuzzy rule in the set of fuzzy rules is determined, including determining the triggering weight of each fuzzy rule as:

[0030]

[0031] Among them, Rule(i,j,k) represents the rules with matching degrees of face, fingerprint and iris i, j and k respectively, where i, j, k∈{L,M,H}; ω Rule(i,j,k) is the trigger weight corresponding to the fuzzy rule Rule(i,j,k); μ d1 、μ d2 、μ d3 Respectively represent the membership of the three-dimensional Euclidean distance of face, fingerprint and iris to each fuzzy subset.

[0032] Furthermore, the recognition score is expressed as:

[0033]

[0034] Where S is the recognition score; ν Rule(i,j,k) It is a numerical value that quantitatively represents the overall matching degree corresponding to the fuzzy rule Rule(i,j,k).

[0035] Furthermore, the decision threshold is trained based on the following formula:

[0036] T new =T old +α×(Precision-Target Precision)

[0037] Among them, α is the learning rate, which is used to set the sensitivity of threshold adjustment; Precision is the current recognition accuracy of the system, and TargetPrecision is the target accuracy; T old is the decision threshold used in the previous decision; T new The decision threshold to be used for the next decision;

[0038] Among them, when the current recognition accuracy of the system is higher than the target accuracy, T new Less than T old ; When the current recognition accuracy of the system is lower than the target accuracy, T new Greater than T old .

[0039] According to a second aspect of the present invention, a device for biometric multimodal feature recognition based on dynamic fuzzy reasoning is provided. The device adopts the biometric multimodal feature recognition method based on dynamic fuzzy reasoning as described in the first aspect of the present invention, comprising:

[0040] A preprocessing module, used to preprocess the acquired face image, fingerprint image and iris image of the target object respectively;

[0041] A feature extraction module is used to extract features from the pre-processed face image, fingerprint image and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint and iris between the target object and the object to be verified in the database according to the feature extraction results;

[0042] The fuzzy reasoning module is used to perform fuzzy reasoning based on the three-dimensional Euclidean distance of face, fingerprint and iris to obtain the recognition score;

[0043] The matching verification module is used to compare the recognition score with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched; wherein the decision threshold is trained based on the preset target recognition accuracy using a reward / penalty mechanism.

[0044] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method according to the first aspect of the present invention.

[0045] The beneficial effects of the present invention are:

[0046] 1. Improve recognition accuracy: By integrating multiple biometric features (face, fingerprint, and iris), the accuracy of the recognition system is enhanced and the false recognition rate is effectively reduced;

[0047] 2. Enhance system robustness: Apply fuzzy reasoning and dynamic threshold adjustment technology to enable the system to adapt to environmental changes and fluctuations in input quality, thereby maintaining high efficiency and stability under various operating conditions;

[0048] 3. By combining a dynamic exponentially decreasing function with a Gaussian function to construct a membership function, stable processing of the input signal and suppression of outliers can be achieved. This is because the exponentially decreasing membership function can quickly change the output value when the input value changes slightly, making it very suitable for applications that require sensitivity to initial changes, such as intrusion detection in security systems and high-precision matching within a small range in biometric systems. The Gaussian membership function is suitable for applications where the input value has a high membership near a certain central value and decays smoothly when it moves away from the central value.

[0049] 4. Improved security: The integrated multimodal biometric method improves the system's ability to resist forgery and deception attacks, which is especially important in application scenarios with high security requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic diagram of the process structure of a biological multimodal feature recognition method based on dynamic fuzzy reasoning proposed by the present invention;

[0051] Figure 2 This is a schematic diagram of facial feature extraction based on a biometric multimodal feature recognition method based on dynamic fuzzy reasoning proposed by the present invention;

[0052] Figure 3 This is a schematic diagram of fingerprint feature extraction of a biometric multimodal feature recognition method based on dynamic fuzzy reasoning proposed by the present invention;

[0053] Figure 4 This is a schematic diagram of iris feature extraction in a biometric multimodal feature recognition method based on dynamic fuzzy reasoning proposed by the present invention. DETAILED DESCRIPTION

[0054] The following describes in detail embodiments of the present invention, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0055] Example 1

[0056] Reference Figure 1-4 This embodiment relates to a biological multimodal feature recognition method based on dynamic fuzzy reasoning, comprising the following steps:

[0057] Step 1: Preprocess the acquired face image, fingerprint image and iris image of the target object respectively.

[0058] Face image preprocessing

[0059] In this embodiment, the acquired facial image is a high-definition RGB facial image. Facial image preprocessing includes: using a multi-task convolutional neural network to perform facial region detection on the acquired facial image, generating candidate face windows and bounding boxes in the P-Net stage, performing fine screening in the R-Net stage to eliminate most false positives, and providing the positions of five key facial feature points (left eye, right eye, nose, left corner of the mouth, and right corner of the mouth) in the O-Net stage;

[0060] The face is aligned and cropped according to the detected facial boundaries and key points, ensuring that the eyes are on the horizontal line and the nose is in the center, and adjusted to a fixed size to prepare for subsequent feature extraction to obtain a preprocessed face image.

[0061] Fingerprint image preprocessing

[0062] Obtain fingerprint image, perform histogram equalization, and improve the contrast of the entire image by adjusting the grayscale distribution of the fingerprint image;

[0063] Noise filtering: Because salt and pepper noise introduced during the scanning process can significantly affect the continuity and clarity of ridges, we apply a median filter using a 5x5 window to remove this noise from the image. Median filtering is particularly effective in removing this type of random noise while maintaining sharp boundaries.

[0064] Based on the local orientation and frequency characteristics of the fingerprint ridges, Gabor filters are applied to enhance the ridges and valleys in the image using multiple orientations and frequencies. The resulting preprocessed fingerprint image is obtained. Each filter orientation is set to 0°, 45°, 90°, and 135° to cover the main orientations of the fingerprint; the frequency parameter is adjusted based on the average ridge spacing of the fingerprint to ensure maximum enhancement.

[0065] The Gabor filter can be defined in two-dimensional space as a combination of a Gaussian kernel function and a modulated cosine wave (real part) and a sine wave (imaginary part). Its expression is:

[0066]

[0067] x′=xcosθ+ysinθ (2)

[0068] y′=-xsinθ+ycosθ (3)

[0069] Where λ is the wavelength, which is related to the average spacing of the ridges, θ is the direction of the Gabor function, which represents the filter's deflection angle, ψ is the phase offset, which is usually set to 0, σ is the standard deviation of the Gaussian envelope, which determines the filter's spatial size, and γ is the spatial aspect ratio, which determines the filter's ellipticity.

[0070] Iris image preprocessing

[0071] The original color iris image is converted into a grayscale image, and the image noise is removed by Gaussian filtering to smooth the image;

[0072] Use Hough transform to locate the iris boundary. Hough circle transform is a technique used to detect circular shapes in images. It can effectively identify the inner and outer boundaries of the iris and accurately crop the iris area.

[0073] Polar coordinate transformation is applied to the cropped iris region to convert the circular iris image into a rectangular form in the Cartesian coordinate system to achieve iris normalization.

[0074] The normalized iris image is subjected to histogram equalization to improve the image contrast and highlight the iris texture, thereby obtaining the preprocessed iris image.

[0075] Step 2: Perform feature extraction on the preprocessed face image, fingerprint image, and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint, and iris between the target object and the object to be verified in the database based on the feature extraction results.

[0076] Facial feature extraction

[0077] Use the pre-trained FaceNet model, which learns and encodes facial features from images through a multi-layer convolutional network, including convolutional layers, activation layers, pooling layers, and normalization layers. Finally, a fully connected layer outputs 128-dimensional embedding vectors, each of which closely represents the visual features of a face.

[0078] L2 normalization is applied to each output 128-dimensional feature vector to ensure that the Euclidean length of the vector is 1.

[0079] Fingerprint feature extraction

[0080] The pre-processed fingerprint image is fed into the NLNet model, which automatically identifies and encodes key fingerprint features, such as ridges, minutiae, and other relevant patterns, through a multi-layer network structure. The output feature vector provides a unique digital representation for each fingerprint, capturing its essential characteristics.

[0081] The feature vectors output from NLNet are L2 normalized to ensure that the length of each vector is consistent.

[0082] Iris feature extraction

[0083] Using the pre-trained DeepIrisNet model, which contains a multi-layer convolutional network specifically designed to extract the unique texture features of the iris, forward propagation is performed to extract the feature vector containing the key information of the iris;

[0084] The feature vectors output from DeepIrisNet are L2 normalized to ensure that the length of each vector is 1.

[0085] For each face, fingerprint, and iris feature vector stored in the database, calculate the Euclidean distance between it and the face, fingerprint, and iris feature vector to be verified. The formula is as follows:

[0086]

[0087] Where is the face, fingerprint, or iris feature vector of a person in the database, is the face, fingerprint, or iris feature vector of the person to be verified, and is the dimension of the vector. The three Euclidean distances d1, d2, and d3 represent the similarity indicators of the face, fingerprint, and iris, respectively.

[0088] Step 3: Perform fuzzy inference based on the three-dimensional Euclidean distance of face, fingerprint and iris to obtain the recognition score.

[0089] This step specifically includes the following sub-steps:

[0090] Step 31: Construct a fuzzy set {L, M, H}, where L, M, and H are three fuzzy subsets of the fuzzy set, corresponding to the three matching degrees of "low matching degree", "medium matching degree", and "high matching degree", respectively.

[0091] The purpose of this step is to fuzzify the feature parameters later, that is, the three-dimensional Euclidean distances d1, d2, and d3 of the face, fingerprint, and iris are used as three input variables, and each variable is mapped to the three fuzzy sets.

[0092] Step 32: Combine the dynamic exponential decreasing function and the Gaussian function to construct a membership function, and determine the membership of the three-dimensional Euclidean distance of the face, fingerprint, and iris to each fuzzy subset based on the membership function.

[0093] The dynamic membership function is designed to determine the membership of faces, fingerprints, and irises to various fuzzy subsets. The exponentially decreasing membership function can quickly change the output value when the input value changes slightly, making it very suitable for applications that need to be sensitive to initial changes, such as intrusion detection in security systems and high-precision matching response within a small range in biometric systems. The Gaussian membership function is suitable for input values with a high membership near a certain central value and smoothly decays when leaving the central value. By combining the calculation results of the dynamic exponentially decreasing function and the Gaussian function, stable processing of the input signal and outlier suppression can be achieved. The formula is as follows:

[0094]

[0095] Where x is the input variable, namely the three-dimensional Euclidean distances d1, d2, and d3 of the face, fingerprint, and iris; is the weight coefficient, which controls the relative contribution of the two membership functions; is the center value of the Gaussian function; is the standard deviation, which controls the width of the membership function; σ is a positive proportionality constant that controls the rate at which the exponential function decays and determines its sensitivity to input changes. It requires careful tuning based on the characteristics of each feature and the specific requirements of the system. Fingerprints and irises are relatively stable and rich in detail, so a higher k value can be used to quickly distinguish subtle differences, helping to improve system sensitivity. A good setting for k is 0.3. The average matching value c based on the fingerprint and iris features is set to 15, and σ is set to 3 to provide sufficient width to cover the variation range of most features. Face recognition is significantly affected by lighting variations. An ambient light sensor is used to capture light intensity in real time. The system automatically adjusts the decay rate k in the membership function based on preset light intensity thresholds, which are categorized as low light, medium light, and high light. When light intensity is below the low light threshold, k is set to 0.05 to improve system tolerance. When light intensity is between the low and medium light thresholds, k is set to 0.1. When light intensity is between the medium and high light thresholds, k is set to 0.15. When light intensity is above the high light threshold, k is set to 0.3 to enhance the system's sensitivity to detail. The average matching degree c based on facial features is set to 10 to 25, and σ is set to 2 to 5, which is adjusted according to the illumination intensity.

[0096] Step 33: Construct a set of fuzzy rules for mapping the matching degrees of face, fingerprint and iris to the corresponding overall matching degrees.

[0097] To perform fuzzy logic reasoning, after fuzzifying the precise parameters, a set of rules should be constructed and stored in the fuzzy rule base. The fuzzy rules are in the form of "IfX1=A,X2=B,X3=CThenY=D", which can be expressed as:

[0098] “Rule(i,j,k):If X1 is A i and X2 is B j and X3 is C k ,Then Y is D i,j,k ,i,j,k=1,2,3”

[0099] Where Rule(i,j,k) represents the rule. The fuzzy rule table is as follows:

[0100]

[0101]

[0102] Step 34: Determine the triggering weight of each fuzzy rule in the set of fuzzy rules based on the membership of the three-dimensional Euclidean distance of the face, fingerprint and iris to each fuzzy subset.

[0103] The trigger weight of each rule is as follows:

[0104]

[0105] Among them, Rule(i,j,k) represents the rules with matching degrees of face, fingerprint and iris i, j and k respectively, where i, j, k∈{L,M,H}; ω Rule(i,j,k) is the trigger weight corresponding to the fuzzy rule Rule(i,j,k); Respectively represent the membership of the three-dimensional Euclidean distance of face, fingerprint and iris to each fuzzy subset.

[0106] Step 35: quantify the overall matching degree of the personnel corresponding to each fuzzy rule, and perform weighted averaging based on the trigger weight to obtain the recognition score.

[0107] Dynamic defuzzification operation: In fuzzy reasoning, y i It is a quantitative expression of the degree of matching of the personnel according to the i-th fuzzy rule. Here, L, M, and H represent three different matching degrees, namely low matching, medium matching, and high matching. These three outputs are assigned numerical values for mathematical processing:

[0108] L(low matching)=0 (7)

[0109] M (medium match) = 0.5 (8)

[0110] H (high match) = 1 (9)

[0111] The recognition score is used to comprehensively consider the output of all fuzzy rules and convert it into a real number recognition score to which the decision threshold can be applied. The formula is as follows:

[0112]

[0113] Where S is the recognition score; ν Rule(i,j,k) It is a numerical value that quantitatively represents the overall matching degree corresponding to the fuzzy rule Rule(i,j,k), and is used for mathematical calculations of defuzzification.

[0114] Step 4: Compare the recognition score with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched. The decision threshold is obtained by training based on a preset target recognition accuracy using a reward / penalty mechanism.

[0115] In order to convert from a comprehensive score to a binary classification result, a decision threshold needs to be set.

[0116]

[0117] If the fuzzy output result is greater than the threshold, it is a "match"; if it is less than the threshold, it is a "mismatch". If the system environment or input conditions are dynamic, the threshold can be dynamically adjusted based on historical performance through a reward / penalty mechanism to optimize the subsequent identity recognition process. The formula for dynamic threshold adjustment is as follows:

[0118] T new =T old +α×(Precision-Target Precision) (12)

[0119] Where α is the learning rate, which sets the sensitivity of the threshold adjustment; Precision is the current system's recognition accuracy, and Target Precision is the target accuracy. When the system's recognition accuracy exceeds the target accuracy, the threshold is reduced to make the system more sensitive to matches. When the system's recognition accuracy falls below the target accuracy, the threshold is increased to reduce the possibility of false matches and improve system security. The new threshold is used for the next decision. This dynamic threshold adjustment mechanism enables the system to self-adjust based on actual operating conditions, rapidly optimizing the threshold configuration and improving recognition accuracy and system robustness.

[0120] Example 2

[0121] Reference Figure 1-4 A biometric multimodal feature recognition method based on dynamic fuzzy reasoning is disclosed. The difference between the method and the first embodiment lies in that the Daugman calculus is used to locate the inner and outer circle boundaries of the iris, thereby segmenting the effective iris area; the positioning accuracy of the iris area is further improved, and the noise resistance performance is better.

[0122] A biological multimodal feature recognition device based on dynamic fuzzy reasoning. The device adopts the biological multimodal feature recognition method based on dynamic fuzzy reasoning described in the first embodiment of the present invention, comprising:

[0123] A preprocessing module, used to preprocess the acquired face image, fingerprint image and iris image of the target object respectively;

[0124] A feature extraction module is used to extract features from the pre-processed face image, fingerprint image and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint and iris between the target object and the object to be verified in the database according to the feature extraction results;

[0125] The fuzzy reasoning module is used to perform fuzzy reasoning based on the three-dimensional Euclidean distance of face, fingerprint and iris to obtain the recognition score;

[0126] The matching verification module is used to compare the recognition score with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched; wherein the decision threshold is trained based on the preset target recognition accuracy using a reward / penalty mechanism.

[0127] A computer-readable storage medium stores a computer program, which implements the above method when executed by a processor.

[0128] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0129] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0130] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0131] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., utilizing an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions. The electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A biological multimodal feature recognition method based on dynamic fuzzy reasoning, characterized in that: The following steps are involved: Preprocessing the acquired face image, fingerprint image and iris image of the target object respectively; Perform feature extraction on the pre-processed face image, fingerprint image and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint and iris between the target object and the object to be verified in the database according to the feature extraction results; Fuzzy reasoning based on 3D Euclidean distance of face, fingerprint and iris; The method includes: constructing a fuzzy set {L, M, H}, where L, M and H are three fuzzy subsets of the fuzzy set, corresponding to three matching degrees of "low matching degree", "medium matching degree" and "high matching degree" respectively; The membership function is constructed by combining the dynamic exponential decreasing function and the Gaussian function; The membership function is expressed as: Where x is the input variable, which is the three-dimensional Euclidean distance of face, fingerprint or iris 、 or ; is the weight coefficient; is the center value of the Gaussian function; is the standard deviation, which is used to control the width of the membership function; is a positive constant of proportionality; Construct a set of fuzzy rules to map the matching degree of face, fingerprint and iris to the corresponding overall matching degree; determining a triggering weight of each fuzzy rule in the set of fuzzy rules based on the membership of each fuzzy subset based on the three-dimensional Euclidean distance of the face, fingerprint and iris; The overall matching degree corresponding to each fuzzy rule is quantified and weighted averaged in combination with the trigger weight to obtain a recognition score; The recognition score is compared with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched; wherein the decision threshold is trained based on the preset target recognition accuracy using a reward / penalty mechanism.

2. The biological multimodal feature recognition method based on dynamic fuzzy reasoning according to claim 1 is characterized in that: The acquired face image, fingerprint image and iris image of the target object are preprocessed separately, including: A multi-task convolutional neural network is used to detect facial regions in the acquired face image; the face is aligned and cropped based on the detected face boundaries and key points to obtain a preprocessed face image; Adjusting the grayscale distribution of the acquired fingerprint image and removing noise by applying a median filter using an N*N window; using Gabor filters of multiple directions and multiple frequencies to enhance the ridges and valleys in the denoised fingerprint image based on the local direction and frequency characteristics of the fingerprint ridges, thereby obtaining a preprocessed fingerprint image; wherein the multiple directions are multiple directions covering the fingerprint, and the multiple frequencies are adjusted based on the average ridge spacing of the fingerprint; The acquired iris image is converted into a grayscale image, and the image noise is removed by Gaussian filtering. The iris boundary of the denoised iris image is located using Hough transform, and a circular iris region is cropped out. Polar coordinate transformation is applied to the cropped circular iris region to convert the circular iris region into a rectangular form in a Cartesian coordinate system to obtain a normalized iris image. Histogram equalization is performed on the normalized iris image to obtain a preprocessed iris image.

3. The biological multimodal feature recognition method based on dynamic fuzzy reasoning according to claim 1 is characterized in that: Perform feature extraction on the pre-processed face image, fingerprint image, and iris image, and determine the three-dimensional Euclidean distance between the target object and the object to be verified in the database based on the feature extraction results, including: The pre-processed face image is input into the pre-trained FaceNet model to obtain a multi-dimensional face feature vector representing the visual features of the face and perform L2 normalization. The preprocessed fingerprint image is input into the NLNet model to obtain the fingerprint feature vector representing the fingerprint feature of each dimension and perform L2 normalization; Using the forward propagation process of the pre-trained DeepIrisNet model, the iris feature vector is extracted from the pre-processed iris image and L2 normalized; Calculate the Euclidean distance between the normalized face, fingerprint and iris feature vectors and the face, fingerprint and iris feature vectors of the object to be verified respectively.

4. The biological multimodal feature recognition method based on dynamic fuzzy reasoning according to claim 1 is characterized in that: Determining a triggering weight of each fuzzy rule in the set of fuzzy rules based on the membership of each fuzzy subset of the three-dimensional Euclidean distance of the face, fingerprint, and iris, including determining the triggering weight of each fuzzy rule as: in, The rules that represent the matching degrees of face, fingerprint and iris are i, j, k respectively, where i, j, k {L,M,H}; Fuzzy rules Corresponding trigger weight; 、 、 Respectively represent the membership of the three-dimensional Euclidean distance of face, fingerprint and iris to each fuzzy subset.

5. The biological multimodal feature recognition method based on dynamic fuzzy reasoning according to claim 4 is characterized in that: The recognition score is expressed as: Where S is the recognition score; It is a fuzzy rule The corresponding numerical value that quantifies the overall matching degree.

6. The biological multimodal feature recognition method based on dynamic fuzzy reasoning according to claim 1 is characterized in that: The decision threshold is trained based on the following formula: in, is the learning rate, which is used to set the sensitivity of the threshold adjustment; Precision is the current recognition accuracy of the system, and Target Precision is the target accuracy; T old is the decision threshold used in the previous decision; T new The decision threshold to be used for the next decision; Among them, when the current recognition accuracy of the system is higher than the target accuracy, T new Less than T old ; When the current recognition accuracy of the system is lower than the target accuracy, T new Greater than T old .

7. A biological multimodal feature recognition device based on dynamic fuzzy reasoning, characterized in that: The method for biometric multimodal feature recognition based on dynamic fuzzy reasoning according to any one of claims 1 to 6 comprises: A preprocessing module, used to preprocess the acquired face image, fingerprint image and iris image of the target object respectively; A feature extraction module is used to extract features from the pre-processed face image, fingerprint image and iris image respectively, and determine the three-dimensional Euclidean distance of the face, fingerprint and iris between the target object and the object to be verified in the database according to the feature extraction results; The fuzzy reasoning module is used to perform fuzzy reasoning based on the three-dimensional Euclidean distance of face, fingerprint and iris to obtain the recognition score; The matching verification module is used to compare the recognition score with a preset decision threshold. When the recognition score is greater than or equal to the decision threshold, the target object is considered to match the object to be verified, otherwise it is not matched; wherein the decision threshold is trained based on the preset target recognition accuracy using a reward / penalty mechanism.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images