A biometric matching system

Abstract

The present application provides a biometric feature comparison system, including a biometric feature collection module, a feature extraction module, a time pattern analysis module, a feature comparison module and a decision module. The present application records the daily unlocking pattern of the user by setting up a time pattern analysis module, and can unlock the door after successfully comparing the partial fingerprint image collected during the unlocking time period of the pattern with the information of the same part of the template in the database, and can also infer and compensate the missing part of the fingerprint information through the deep learning technology provided in the feature extraction module, and unlock the door after successfully comparing the compensated fingerprint information with the complete template information in the database, thereby avoiding the problem of multiple fingerprint unlocking failures caused by the user's wet fingers or dust on the fingers making the fingerprint part unclear, thereby improving the accuracy and reliability of fingerprint unlocking, and also improving the efficiency of fingerprint unlocking and the comfort of user experience.

Description

Technical Field

[0001] The present invention relates to the field of feature comparison systems, and in particular to a biological feature comparison system. Background Art

[0002] The rapid development of fingerprint recognition technology in recent years has made fingerprint locks widely used in the field of identity authentication. Due to the uniqueness of fingerprints, fingerprint locks have a wide range of use and application value. Applying fingerprint recognition technology to access control locks can more effectively achieve the goals of intelligent door lock products and anti-theft. The core function of fingerprint locks is affected by the accuracy of fingerprint recognition.

[0003] At present, some hotel and apartment residents have begun to use door locks with fingerprint unlocking function. However, during use, the fingerprint part may be unclear due to moisture or dust on the user's fingers, resulting in multiple fingerprint unlocking failures, and even causing the door lock to alarm for a long time, which brings inconvenience to the guests. At the same time, the alarm sound also affects the comfort of other guests.

[0004] For example, the Chinese invention patent (application number: CN201811008189.3) discloses a “high-precision fingerprint recognition method and system based on image processing”. Its description discloses: “In the process of realizing the present invention, the inventor found that there are at least the following problems in the prior art: in the existing security field, especially for fingerprint locks, the accuracy of one-to-one recognition of fingerprint images of different qualities collected in different environments is not ideal, and the recognition rate and search time for large-scale fingerprint libraries are far from meeting the needs of practical applications. There may also be problems such as broken fingerprints, unclear fingerprints, etc. that cannot be correctly recognized in complex environments”; the above patents can prove the defects of the prior art;

[0005] Therefore, we make improvements to this and propose a biometric matching system. Summary of the invention

[0006] The purpose of the present invention is to solve the current problem that the fingerprint part is unclear due to the user's wet fingers or dust on the fingers, which leads to multiple fingerprint unlocking failures.

[0007] In order to achieve the above-mentioned object of the invention, the present invention provides the following biometric feature comparison system to improve the above-mentioned problem.

[0008] The specific application is as follows:

[0009] A biometric feature comparison system includes the following modules:

[0010] Biometrics collection module: It has multiple biometrics collection submodules, including face collection module, iris collection module, and fingerprint collection module, which are used to collect multiple biometric data of users and transmit them to the feature extraction module for processing;

[0011] Feature extraction module: used to receive raw data from different biometric feature collection modules, and perform preprocessing and feature extraction. Feature extraction uses deep learning technology to preferentially extract key features from the collected biometric features. In fingerprint comparison, the feature extraction module identifies the detailed features of the ridges and valleys in the fingerprint and saves them. When part of the fingerprint pattern is missing, the module infers and compensates based on the existing key features to restore the complete fingerprint information for subsequent feature comparison and decision-making.

[0012] Time pattern analysis module: used to analyze the user's unlocking time pattern. By collecting the user's past unlocking records, the user's unlocking habits can be identified, including the regular time after get off work and the time when returning from other trips.

[0013] Feature comparison module: used to compare the extracted features with the information in the database. In fingerprint comparison, the collected fingerprint features are matched with the fingerprint templates stored in the database;

[0014] Decision-making module: Based on the collected biometric information, time pattern analysis results and feature comparison results, the decision-making module comprehensively judges and controls the unlocking, and by setting the priority, the decision-making module selects the unlocking method according to the set rules. The priority rules are as follows:

[0015] If the module detects that the unlocking time is within the regular unlocking time period, the unlocking method based on time regularity will be used first. If the module detects that the unlocking time is within the irregular unlocking time period, the unlocking method based on deep learning will be used.

[0016] The feature extraction module works as follows:

[0017] 1. Receive the original fingerprint image data from the fingerprint collection module, which includes the digitized fingerprint image and detailed fingerprint pattern information;

[0018] 2. Preprocess the original fingerprint image;

[0019] 3. Extract features from the preprocessed fingerprint image through deep learning technology;

[0020] 4. The extracted key features include the endpoints and bifurcation points of the ridges and valleys, and are saved to form the feature template and feature vector of the fingerprint;

[0021] 5. The feature extraction module uses a deep learning model and existing key features to infer and compensate for fingerprint images with missing lines.

[0022] 6. After the above steps, the feature extraction module outputs the extracted feature information to the subsequent comparison module and decision module;

[0023] The working steps of the time rule analysis module are as follows:

[0024] Ⅰ. Collect the user's historical unlocking records, including the time and method of each unlocking;

[0025] Ⅱ. Clean the collected raw data, remove duplicate, erroneous and incomplete records, and sort the data in chronological order;

[0026] III. Use statistical methods to identify the regularity of users' unlocking time, analyze the distribution of users' unlocking time, and find out the time periods with strong regularity, including the time after get off work every day, the travel time on weekends, and the changing trend of users' unlocking time with the change of seasons and working days;

[0027] IV. Based on the identified time pattern, construct the user's unlocking habit pattern, which includes unlocking behavior at fixed time points, periodic unlocking patterns, and possible abnormal unlocking behavior;

[0028] V. The unlocking time pattern obtained by analysis is recorded and stored, and the record is shared with the decision-making module for determining the unlocking time;

[0029] VI. Continuously collect new unlocking records, update users’ unlocking habit patterns, and regularly re-analyze users’ unlocking time patterns to adapt to changes in user habits and updates in security needs;

[0030] The working steps of the biometric feature comparison system for fingerprint unlocking are divided into two situations, as follows:

[0031] The first type: There is no abnormality in the fingerprint and the lock is unlocked normally;

[0032] The first step is to collect fingerprint patterns through the fingerprint collection module and transmit them to the feature extraction module;

[0033] In the second step, the feature extraction module uses deep learning technology to process the collected fingerprint images and accurately extract the key features of the ridges and valleys in the fingerprints;

[0034] In the third step, the comparison module compares the extracted fingerprint features with the fingerprint template stored in the database;

[0035] Step 4: The decision module determines the fingerprint matching situation based on the comparison results. If the match is successful, the decision module issues an unlock command. If the match is unsuccessful, the decision module does not issue an unlock command. If the match is unsuccessful for multiple times, the decision module issues an alarm command. The number of multiple unmatched times here can be freely set by the system.

[0036] The second situation: the fingerprint contains water, dust and fingerprint fracture phenomenon;

[0037] S10, transmitting the collected clear part of the image to the feature extraction module through the fingerprint collection module;

[0038] S20, the feature extraction module determines that the information is a partially clear image, and sends the determination result to the decision module, and the decision module determines the time relationship between the unlocking time at this time and the regular unlocking time period through the time rule analysis module;

[0039] S30, the decision module determines through the time rule analysis module that the unlocking time is within the regular unlocking time period, then selects to use the unlocking method based on the time rule, and sends the decision instruction to the feature extraction module;

[0040] The decision module determines through the time rule analysis module that the unlocking time is not within the regular unlocking time period, and then chooses to use the unlocking method based on deep learning, and sends the decision instruction to the feature extraction module;

[0041] S40, the unlocking method based on time rule includes the following steps:

[0042] S401, based on the first and second steps above, the collected partially clear image is transmitted to the feature comparison module through the feature extraction module;

[0043] S402, the feature comparison module compares the received partially clear image with the complete information image stored in the database, and sends the comparison result to the decision module;

[0044] S403, if the comparison result of the clear image of the part matches the same part of the complete information image in the database, the decision module issues an unlocking instruction; if the comparison result of the clear image of the part does not match the same part of the complete information image in the database, the decision module does not issue an unlocking instruction; if multiple matches are unsuccessful, the decision module issues an alarm instruction, and the number of multiple mismatches here can be freely set by the system;

[0045] S50, the unlocking method based on deep learning includes the following steps:

[0046] S501, based on the first and second steps above, the collected partially clear image is inferred and compensated by the feature extraction module to restore the complete fingerprint information image and transmit it to the feature comparison module;

[0047] S502, the feature comparison module compares the restored complete fingerprint information image with the complete information image stored in the database, and sends the comparison result to the decision module;

[0048] S503. If the comparison result between the restored complete fingerprint information image and the complete information image in the database matches, an unlocking command is issued through the decision module. If the comparison result between the restored complete fingerprint information image and the complete information image in the database does not match, no unlocking command is issued through the decision module. If multiple matches are unsuccessful, an alarm command is issued through the decision module. The number of multiple mismatches here can be freely set by the system.

[0049] As a preferred technical solution of this application, the image preprocessing steps in the two steps are as follows:

[0050] Step 1: Use the weighted average method to calculate the final grayscale value based on the contribution of the three color channels of red, green and blue to the grayscale value, and convert the color fingerprint image into a grayscale image;

[0051] Step 2: Adjust the fingerprint images collected from different devices and environments to a uniform size and orientation;

[0052] Step 3: Use filtering algorithm to improve the clarity and contrast of fingerprint image;

[0053] Step 4: Use the adaptive threshold method to convert the enhanced grayscale image into a binary image containing only black and white colors;

[0054] Step 5: Use iterative stripping method to thin the lines in the binary fingerprint image into lines with a single pixel width;

[0055] Step 6: Further eliminate existing noise points and false features.

[0056] As the preferred technical solution of this application, the deep learning feature extraction working steps in the three steps are as follows:

[0057] S1. Establish a deep neural network model, which consists of multiple hidden layers, each hidden layer contains multiple neurons, and the neurons are interconnected through weights and biases. The weights and biases are parameters that need to be learned during the training process of the model;

[0058] S2, the model randomly initializes weights and biases, and sends data to the neural network for forward propagation calculation. The input data will pass through each layer of the model, and generate output through weighted summation and activation function operations;

[0059] S3. The model compares the output calculated by forward propagation with the true label and calculates the loss function. The loss function is used to measure the difference between the model's predicted results and the true results. The model calculates the gradient of the loss function with respect to each weight and bias, and updates the weights and biases based on these gradients.

[0060] S4. Repeat the above steps S2 and S3 to form an iterative process. In each iteration, the model will perform forward propagation based on the current weights and biases, calculate the loss function, and then perform backpropagation and update the weights and biases. This process will continue until the preset number of iterations is reached and the loss function value drops below the preset threshold of 0.1.

[0061] As a preferred technical solution of the present application, the calculation formula for the weighted sum in step S2 is: ,in: It is the first The weight of the input, It is Inputs, is the bias term, is the result of weighted summation.

[0062] As a preferred technical solution of the present application, the calculation formula of the activation function in step S2 is: ,in: is the result of weighted summation, is the activation function, is the output of the neuron. It will serve as the input of the next layer of neurons and continue to participate in the calculation of the network.

[0063] As a preferred technical solution of this application, the loss function calculation formula in step S3 is as follows:

[0064] ,in: is the value of the loss function, is the number of samples, It is The true label of a sample is usually a binary variable. The model is The probability that a sample is predicted to be the positive class.

[0065] As the preferred technical solution of the present application, the specific implementation of the weighted average method in step one is calculated by the formula gray value = 0.299×R+0.587×G+0.114×B.

[0066] As a preferred technical solution of the present application, the working steps of the adaptive threshold method in step 4 to binarize the grayscale image are as follows:

[0067] S221, reading a grayscale image to be processed, which includes a pre-processed grayscale image;

[0068] S222, calculating local statistics of the image, including local mean and local standard deviation, by applying a rectangular region of fixed size to the image, and calculating statistical characteristics of the local neighborhood of each pixel in the rectangular region;

[0069] S223, based on the above local statistics, determine a threshold for each pixel and each local area, the threshold may be a fixed offset based on the grayscale mean, or a dynamic value based on the standard deviation;

[0070] S224, comparing the gray value of each pixel in the image with the corresponding local threshold, setting the pixels whose gray value is greater than or equal to the threshold to white, and setting the pixels whose gray value is less than the threshold to black;

[0071] S225, through the above steps, the original grayscale image is converted into a binary image, which contains only black and white colors and is used to represent fingerprint lines and other features;

[0072] S226, removing noise and filling holes from the binary image to further improve the binarization result;

[0073] S227. Output and save the generated binary image.

[0074] As a preferred technical solution of the present application, the calculation formula of the weighted average value in step S213 is as follows:

[0075] ,in: is the position in the Gaussian kernel The weight value of is the position in the image The pixel value of is half the radius of the Gaussian kernel, is the sum of all weight values ​​in the Gaussian kernel and is used to normalize the weighted sum.

[0076] As a preferred technical solution of the present application, the Gaussian function calculation formula in step S213 is as follows:

[0077] ,in: is the Gaussian function at the point The value of, i.e., the weight,

[0078] and is the coordinate relative to the center point of the filter, is the standard deviation, which determines the width of the Gaussian function, that is, the width of the weight distribution.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] In the scheme of this application:

[0081] 1. In order to solve the technical problem that a biometric feature matching system in the prior art fails to unlock the door multiple times due to unclear fingerprints caused by damp or dusty fingers, the present application records the daily unlocking rules of the user by providing a time rule analysis module, and can unlock the door after successfully comparing the partial fingerprint image collected during the regular unlocking time period with the information of the same part of the template in the database. The missing part of the fingerprint information can also be inferred and compensated by the deep learning technology provided in the feature extraction module, and the fingerprint information after compensation is successfully compared with the complete template information in the database to achieve unlocking, thereby avoiding the problem of multiple fingerprint unlocking failures caused by unclear fingerprints caused by damp or dusty fingers, improving the accuracy and reliability of fingerprint unlocking, and also improving the efficiency of fingerprint unlocking and the comfort of user experience;

[0082] 2. When fingerprint collection is incomplete, the reliability of the system is improved again by judging whether the user's fingerprint unlocking is within the regular unlocking time period and then selecting different fingerprint unlocking methods. In addition, when unlocking within the regular time period, the process of inferring and compensating the missing fingerprint image can be omitted, effectively reducing the power consumption of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A schematic diagram of a biometric feature comparison system provided for this application;

[0084] Figure 2 A workflow diagram of fingerprint unlocking in a biometric feature comparison system provided in this application. DETAILED DESCRIPTION

[0085] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0086] In order to solve this technical problem, the present invention provides a biometric feature comparison system, which is used to unlock the door better and faster when the fingerprint is wet, broken and dusty.

[0087] Specifically, please refer to Figure 1 and Figure 2 , a biometric matching system, comprising the following modules:

[0088] Biometrics collection module: It has multiple biometrics collection submodules, including face collection module, iris collection module, and fingerprint collection module, which are used to collect multiple biometric data of users and transmit them to the feature extraction module for processing;

[0089] Feature extraction module: used to receive raw data from different biometric feature collection modules, and perform preprocessing and feature extraction. Feature extraction uses deep learning technology to preferentially extract key features from the collected biometric features. In fingerprint comparison, the feature extraction module identifies the detailed features of the ridges and valleys in the fingerprint and saves them. When part of the fingerprint pattern is missing, the module infers and compensates based on the existing key features to restore the complete fingerprint information for subsequent feature comparison and decision-making.

[0090] Time pattern analysis module: used to analyze the user's unlocking time pattern. By collecting the user's past unlocking records, the user's unlocking habits can be identified, including the regular time after get off work and the time when returning from other trips. This information is very helpful for optimizing the unlocking process and improving security.

[0091] Feature matching module: used to compare the extracted features with the information in the database. In fingerprint matching, the collected fingerprint features are matched with the fingerprint templates stored in the database. The matching module will consider various factors during the matching process, such as the integrity and clarity of the fingerprint, to ensure the accuracy of the matching;

[0092] Decision-making module: Based on the collected biometric information, time pattern analysis results and feature comparison results, the module makes a comprehensive judgment to control unlocking, and sets priorities. The decision-making module selects the unlocking method according to the set rules. The priority rules are as follows:

[0093] If the module detects that the unlocking time is within the regular unlocking time period, the unlocking method based on time regularity will be used first. If the module detects that the unlocking time is within the irregular unlocking time period, the unlocking method based on deep learning will be used.

[0094] The working steps of the feature extraction module are as follows:

[0095] 1. Receive the original fingerprint image data from the fingerprint collection module, which includes the digitized fingerprint image and detailed fingerprint pattern information;

[0096] 2. Preprocess the original fingerprint image;

[0097] 3. Extract features from the preprocessed fingerprint image through deep learning technology;

[0098] 4. The extracted key features include the endpoints and bifurcation points of the ridges and valleys, and are saved to form the feature template and feature vector of the fingerprint, which are used in the subsequent comparison and recognition process;

[0099] 5. The feature extraction module uses a deep learning model and existing key features to infer and compensate for fingerprint images with missing textures to restore complete fingerprint information;

[0100] 6. After the above steps, the feature extraction module outputs the extracted feature information to the subsequent comparison module and decision module. These feature information will be used to compare with other fingerprint features in the database to verify identity or make other related decisions;

[0101] The working steps of the time pattern analysis module are as follows:

[0102] Ⅰ. Collect the user's historical unlocking records, including the time and method of each unlocking, such as password, fingerprint, mobile phone APP, etc.;

[0103] Ⅱ. Clean the collected raw data, remove duplicate, erroneous and incomplete records, and sort the data in chronological order to facilitate subsequent analysis;

[0104] III. Use statistical methods, such as frequency analysis and time series analysis, to identify the regularity of users' unlocking time, analyze the distribution of users' unlocking time, and find out the time periods with strong regularity, including the time after get off work every day, the travel time on weekends, and the changing trend of users' unlocking time with the change of seasons and working days;

[0105] IV. Based on the identified time pattern, construct the user's unlocking habit pattern, which includes unlocking behavior at fixed time points, periodic unlocking patterns, and possible abnormal unlocking behavior;

[0106] V. The unlocking time pattern obtained by analysis is recorded and stored, and the record is shared with the decision-making module for determining the unlocking time;

[0107] VI. Continuously collect new unlocking records, update users’ unlocking habit patterns, and regularly re-analyze users’ unlocking time patterns to adapt to changes in user habits and updates in security needs;

[0108] The working steps of a biometric feature comparison system for fingerprint unlocking are divided into two situations, as follows:

[0109] The first type: There is no abnormality in the fingerprint and the lock is unlocked normally;

[0110] The first step is to collect fingerprint patterns through the fingerprint collection module and transmit them to the feature extraction module;

[0111] In the second step, the feature extraction module uses deep learning technology to process the collected fingerprint images and accurately extract the key features of the ridges and valleys in the fingerprints;

[0112] In the third step, the comparison module compares the extracted fingerprint features with the fingerprint template stored in the database;

[0113] Step 4: The decision module determines the fingerprint matching situation based on the comparison results. If the match is successful, the decision module issues an unlock command. If the match is unsuccessful, the decision module does not issue an unlock command. If the match is unsuccessful for multiple times, the decision module issues an alarm command. The number of multiple unmatched times here can be freely set by the system.

[0114] The second situation: the fingerprint contains water, dust and fingerprint fracture phenomenon;

[0115] S10, transmitting the collected clear part of the image to the feature extraction module through the fingerprint collection module;

[0116] S20, the feature extraction module determines that the information is a partially clear image, and sends the determination result to the decision module, and the decision module determines the time relationship between the unlocking time at this time and the regular unlocking time period through the time rule analysis module;

[0117] S30, the decision module determines through the time rule analysis module that the unlocking time is within the regular unlocking time period, then selects to use the unlocking method based on the time rule, and sends the decision instruction to the feature extraction module;

[0118] The decision module determines through the time rule analysis module that the unlocking time is not within the regular unlocking time period, and then chooses to use the unlocking method based on deep learning, and sends the decision instruction to the feature extraction module;

[0119] S40, the unlocking method based on time rule includes the following steps:

[0120] S401, based on the first and second steps above, the collected partially clear image is transmitted to the feature comparison module through the feature extraction module;

[0121] S402, the feature comparison module compares the received partially clear image with the complete information image stored in the database, and sends the comparison result to the decision module;

[0122] S403, if the comparison result of the clear image of the part matches the same part of the complete information image in the database, the decision module issues an unlocking instruction; if the comparison result of the clear image of the part does not match the same part of the complete information image in the database, the decision module does not issue an unlocking instruction; if multiple matches are unsuccessful, the decision module issues an alarm instruction, and the number of multiple mismatches here can be freely set by the system;

[0123] S50, the unlocking method based on deep learning includes the following steps:

[0124] S501, based on the first and second steps above, the collected partially clear image is inferred and compensated by the feature extraction module to restore the complete fingerprint information image and transmit it to the feature comparison module;

[0125] S502, the feature comparison module compares the restored complete fingerprint information image with the complete information image stored in the database, and sends the comparison result to the decision module;

[0126] S503. If the comparison result between the restored complete fingerprint information image and the complete information image in the database matches, an unlocking command is issued through the decision module. If the comparison result between the restored complete fingerprint information image and the complete information image in the database does not match, no unlocking command is issued through the decision module. If multiple matches are unsuccessful, an alarm command is issued through the decision module. The number of multiple mismatches here can be freely set by the system.

[0127] The present invention provides a biometric feature comparison system, which can realize unlocking more accurately and quickly when the fingerprint is wet, broken or covered with dust.

[0128] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0129] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0130] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0131] Embodiment 1, a biometric feature comparison system, the image preprocessing steps in the second step are as follows:

[0132] Step 1: Use the weighted average method to calculate the final grayscale value based on the contribution of the three color channels of red, green and blue to the grayscale value, and convert the color fingerprint image into a grayscale image;

[0133] Step 2: Adjust the fingerprint images collected from different devices and environments to a uniform size and orientation for subsequent feature extraction and matching;

[0134] Step 3: Use filtering algorithm to improve the clarity and contrast of fingerprint image;

[0135] Step 4: Use the adaptive threshold method to convert the enhanced grayscale image into a binary image containing only black and white colors;

[0136] Step 5: Use iterative stripping method to refine the lines in the binary fingerprint image into lines with a single pixel width, so as to reduce redundant information and highlight the main features of the fingerprint lines, so as to facilitate subsequent feature extraction and matching;

[0137] Step 6: Further eliminate existing noise points and pseudo features;

[0138] Furthermore, the deep learning feature extraction process in the three steps is as follows:

[0139] S1. Establish a deep neural network model, which consists of multiple hidden layers. Each hidden layer contains multiple neurons. Neurons are connected to each other through weights and biases. Weights and biases are parameters that need to be learned during the training process of the model.

[0140] S2. At the beginning of training, the model randomly initializes weights and biases, and sends data to the neural network for forward propagation calculation. The input data will pass through each layer of the model, and generate output through weighted summation and activation function operations. That is, the input of each neuron is the weighted sum of the outputs of all neurons in the previous layer. Then this weighted sum is passed through the activation function to generate the output of the neuron. This process will continue until the last layer of the model, that is, the output layer;

[0141] S3. The model compares the output calculated by forward propagation with the true label and calculates the loss function. The loss function is used to measure the difference between the model's predicted results and the true results. The model calculates the gradient of the loss function with respect to each weight and bias, and updates the weights and biases based on these gradients.

[0142] S4. Repeat the above steps S2 and S3 to form an iterative process. In each iteration, the model will perform forward propagation based on the current weights and biases, calculate the loss function, and then perform backpropagation and update the weights and biases. This process will continue until the preset number of iterations is reached and the loss function value drops below the preset threshold of 0.1.

[0143] The calculation formula for the weighted sum in step S2 is: ,in: It is the first The weight of the input, It is Inputs, is the bias term, is the result of weighted summation;

[0144] The calculation formula of the activation function in step S2 is: ,in: is the result of weighted summation, is the activation function, is the output of the neuron. It will serve as the input of the next layer of neurons and continue to participate in the calculation of the network;

[0145] The loss function calculation formula in step S3 is as follows:

[0146] ,in: is the value of the loss function, is the number of samples, It is The true label of a sample is usually a binary variable. The model is The probability of a sample being predicted as a positive class;

[0147] The specific interconnection between neurons through weights and biases in step S1 is as follows:

[0148] Weight is a numerical parameter that connects two neurons. In a neural network, each connection has a weight associated with it. The weight determines the degree of influence of the output of the previous layer of neurons on the input of the current neuron. Specifically, the input of the current neuron is the sum of the output of the previous layer of neurons multiplied by the corresponding weight.

[0149] Suppose we have a simple neural network, where the first hidden layer has 3 neurons (let's call them A, B, C), and the second hidden layer has 1 neuron (let's call it D). Then, the connections from A, B, C to D will have their own weights. For example, the weight from A to D can be expressed as (w_{AD}), the weight from B to D is (w_{BD}), and the weight from C to D is (w_{CD}). When the input of the D neuron is calculated, it will take the output of the A, B, and C neurons (usually the value processed by the activation function), and then multiply them with the corresponding weights respectively, and finally add these products;

[0150] Bias is another important parameter, which is added to the weighted input of the neuron. The bias can be regarded as a learnable threshold used to adjust the activation degree of the neuron. In the above example, a bias term (b_D) is added to the input calculation of neuron D. The introduction of bias increases the flexibility of the model, allowing the neuron to still have a non-zero output without any input (that is, all inputs are 0);

[0151] After calculating the weighted input of the neuron, the input is processed through the activation function to obtain the final output of the neuron;

[0152] Furthermore, the steps of inferring and compensating the fingerprint image with missing lines of key features through the deep learning model in the five steps are as follows:

[0153] Inference process:

[0154] S100, a deep learning model, captures local and global contextual information in the image and uses the continuity, direction and spacing of the surrounding complete lines to infer the characteristics of the missing parts in the fingerprint image;

[0155] S200, the model searches for feature fragments similar to the missing part from the existing fingerprint data set to estimate the features of the missing part;

[0156] Compensation process:

[0157] S101, based on the inferred missing features, the model can generate new feature information to fill the missing part, by adding a generation unit in the output layer of the model, the generation unit includes a decoder, and the decoder encodes and compensates the inferred result of the missing part of the image, so that the model can output a complete fingerprint image;

[0158] S102, the model merges the inferred features with the surrounding existing features in a jigsaw manner to smooth the transition and reduce the abruptness;

[0159] S103, performing denoising, smoothing, and enhancing operations on the compensated fingerprint image, further improving the performance and reliability of fingerprint recognition by optimizing the quality and accuracy of the fingerprint image, and providing a more accurate and stable fingerprint recognition service for practical applications;

[0160] Furthermore, the specific implementation of the weighted average method in step 1 is calculated by the formula gray value = 0.299×R+0.587×G+0.114×B, where:

[0161] R: represents the value of the red channel, ranging from 0 to 255. In a color image, the red component of each pixel is represented by this value;

[0162] G: represents the value of the green channel, also ranging from 0 to 255, which indicates the intensity of green in the pixel;

[0163] B: represents the value of the blue channel, also ranging from 0 to 255, which represents the intensity of blue in the pixel;

[0164] 0.299: is the weight of the red channel. The human eye is more sensitive to red than green and less sensitive than blue, so this weight value is set to a value in the middle;

[0165] 0.587: is the weight of the green channel. The human eye is more sensitive to green than to red and blue, so this weight value is the largest.

[0166] 0.114: is the weight of the blue channel. The human eye is less sensitive to blue than to red and green, so this weight value is the smallest.

[0167] The weight value refers to the contribution of the red, green, and blue color channels to the final grayscale value during the image grayscale conversion process.

[0168] Embodiment 2 further optimizes a biometric feature comparison system provided in Embodiment 1. Specifically, the steps of binarizing the grayscale image using the adaptive threshold method in step 4 are as follows:

[0169] S221, reading a grayscale image to be processed, which includes a pre-processed grayscale image;

[0170] S222, calculating local statistics of the image, including local mean and local standard deviation, by applying a rectangular region of fixed size to the image, and calculating statistical characteristics of the local neighborhood of each pixel in the rectangular region;

[0171] S223, based on the above local statistics, determine a threshold for each pixel and each local area, the threshold may be a fixed offset based on the grayscale mean, or a dynamic value based on the standard deviation;

[0172] S224, comparing the gray value of each pixel in the image with the corresponding local threshold, setting the pixels whose gray value is greater than or equal to the threshold to white, and setting the pixels whose gray value is less than the threshold to black;

[0173] S225, through the above steps, the original grayscale image is converted into a binary image, which contains only black and white colors and is used to represent fingerprint lines and other features;

[0174] S226, removing noise and filling holes from the binary image to further improve the binarization result;

[0175] S227, output and save the generated binary image;

[0176] Furthermore, the standard deviation of the Gaussian filter in step S211 is determined as follows:

[0177] The standard deviation mainly depends on the type and degree of noise in the fingerprint image. If the noise is mainly high-frequency noise, such as randomly distributed bright spots and dark spots, a small standard deviation should be selected to smooth the noise while retaining the details of the fingerprint. If the noise is low-frequency, such as blurring of the entire area, a large standard deviation should be selected to smooth the area. This usually needs to be determined through experiments. Try different standard deviation values, observe the effect after filtering, and find the best value that can remove noise and retain details.

[0178] The size of the Gaussian filter in step S211 is determined as follows:

[0179] The size of the filter determines the area covered by the filter. The size of the filter is determined by the resolution of the fingerprint image and the distribution of the noise. If the noise is locally distributed, a small filter should be selected. If the noise is distributed throughout the image, a larger filter should be selected to cover the area. This requires experiments to determine the optimal filter size.

[0180] The calculation formula of the weighted average in step S213 is as follows:

[0181] ,in: is the position in the Gaussian kernel The weight value of is the position in the image The pixel value of is half the radius of the Gaussian kernel, is the sum of all weight values ​​in the Gaussian kernel, used to normalize the weighted sum;

[0182] The Gaussian function calculation formula in step S213 is as follows:

[0183] ,in: is the Gaussian function at the point The value of, that is, the weight, and is the coordinate relative to the center point of the filter, is the standard deviation, which determines the width of the Gaussian function, that is, the width of the weight distribution;

[0184] The weighted average calculation formula visits all pixels covered by the Gaussian kernel, multiplies each pixel value by the corresponding weight, then sums and normalizes to obtain the final weighted average value, which will be used as the new value of the corresponding pixel in the filtered image;

[0185] Furthermore, the steps of using the iterative stripping method to thin the lines in the binary fingerprint image into lines of single pixel width are as follows:

[0186] (1) Load the binary fingerprint image;

[0187] (2) Create a labeled image of the same size as the original image, with all pixels initialized to an unprocessed state;

[0188] (3) Select a neighborhood window size to check the status of the surrounding pixels of the current pixel, visit each pixel in the image, check the distribution of black and white pixels in its neighborhood, define black pixels with white pixels in its neighborhood, and white pixels with black pixels in its neighborhood as edge pixels and mark them;

[0189] (4) Mark the detected edge pixels as pending;

[0190] (5) Setting priority rules to determine the edge pixels that should be stripped first;

[0191] (6) According to the priority rule, select some edge pixels for stripping. For each selected edge pixel, change its color to the same color as its non-edge neighboring pixels according to the color of its neighboring pixels, update the marked image, and mark the stripped pixels as processed;

[0192] (7) Repeat the process of edge detection, marking edge pixels, and stripping operations until a stopping condition is met. The stopping condition may be that a predetermined number of iterations is reached and there is no longer an obvious line edge in the image;

[0193] (8) By setting the size threshold of the connected area, the isolated noise points generated by the stripping process are removed, and the image is smoothed using morphological operations;

[0194] (9) Carefully observe the refined image to ensure that the lines remain continuous and without breaks;

[0195] (10) If you get a satisfactory refinement result, save the image as a new file;

[0196] Furthermore, the priority method in step (5) is as follows:

[0197] Priority settings based on grain direction:

[0198] S301, analyzing the directionality of the grain lines. In the fingerprint image, the grain lines usually have clear directions, including horizontal, vertical and oblique directions;

[0199] S302, for each edge pixel, calculating the angle between its direction and the main direction of the grain line;

[0200] S303, determining the peeling priority of edge pixels according to the size of the angle, the smaller the angle, the higher the priority;

[0201] Priority settings based on grain width:

[0202] S401, analyzing the width of the grain lines;

[0203] S402, for each edge pixel, calculating the width of the black pixels around it;

[0204] S403: Determine the stripping priority of edge pixels according to the width of the grain line. The larger the grain line width is, the higher the priority is.

[0205] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection, an electrical connection, or communication with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0206] Obviously, the embodiments described above are only some embodiments of the present invention, rather than all embodiments. The preferred embodiments of the present invention are given in the accompanying drawings, but they do not limit the patent scope of the present invention. The present invention can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive. Although the present invention has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific embodiments, or to perform equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of the present invention, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present invention.

Claims

1. A biometric feature comparison system, characterized in that: Includes the following modules: Biometrics collection module: It has multiple biometrics collection submodules, including face collection module, iris collection module, and fingerprint collection module, which are used to collect multiple biometric data of users and transmit them to the feature extraction module for processing; Feature extraction module: used to receive raw data from different biometric feature collection modules, and perform preprocessing and feature extraction. Feature extraction uses deep learning technology to preferentially extract key features from the collected biometric features. In fingerprint comparison, the feature extraction module identifies the detailed features of the ridges and valleys in the fingerprint and saves them. When part of the fingerprint pattern is missing, the module infers and compensates based on the existing key features to restore the complete fingerprint information and provide a basis for subsequent feature comparison and decision-making; Time pattern analysis module: used to analyze the user's unlocking time pattern and identify the user's unlocking habits by collecting the user's past unlocking records; Feature matching module: used to compare the extracted features with the information in the database. In fingerprint matching, the collected fingerprint features are matched with the fingerprint templates stored in the database, and the matching results are fed back to the decision module; Decision-making module: Based on the collected biometric information, time pattern analysis results and feature comparison results, the decision-making module comprehensively judges and controls the unlocking, and by setting the priority, the decision-making module selects the unlocking method according to the set rules. The priority rules are as follows: If the module detects that the unlocking time is within the regular unlocking time period, the unlocking method based on time regularity will be used first. If the module detects that the unlocking time is within the irregular unlocking time period, the unlocking method based on deep learning will be used. The feature extraction module works as follows:

1. Receive the original fingerprint image data from the fingerprint collection module, which includes the digitized fingerprint image and detailed fingerprint pattern information; 2. Preprocess the original fingerprint image; 3. Extract features from the preprocessed fingerprint image through deep learning technology; 4. The extracted key features include the endpoints and bifurcation points of the ridges and valleys, and are saved to form the feature template and feature vector of the fingerprint; 5. The feature extraction module uses a deep learning model and existing key features to infer and compensate for fingerprint images with missing lines.

6. The feature extraction module outputs the extracted feature information to the subsequent comparison module and decision module.

2. A biometric feature comparison system according to claim 1, characterized in that: The working steps of the time rule analysis module are as follows: Ⅰ. Collect the user's historical unlocking records, including the time and method of each unlocking; Ⅱ. Clean the collected raw data, remove duplicate, erroneous and incomplete records, and sort the data in chronological order; III. Use statistical methods to identify the regularity of users' unlocking time, analyze the distribution of users' unlocking time, and find out the time periods with strong regularity, including the time after get off work every day, the travel time on weekends, and the changing trend of users' unlocking time with the change of seasons and working days; IV. Based on the identified time pattern, construct the user's unlocking habit pattern, which includes unlocking behavior at fixed time points, periodic unlocking patterns, and possible abnormal unlocking behavior; V. The unlocking time pattern obtained by analysis is recorded and stored, and the record is shared with the decision-making module for determining the unlocking time; VI. Continuously collect new unlocking records, update users’ unlocking habit patterns, and regularly re-analyze users’ unlocking time patterns to adapt to changes in user habits and updates in security needs; The working steps of the biometric feature comparison system for fingerprint unlocking are divided into two situations, as follows: The first type: There is no abnormality in the fingerprint and the lock is unlocked normally; The first step is to collect fingerprint patterns through the fingerprint collection module and transmit them to the feature extraction module; In the second step, the feature extraction module uses deep learning technology to process the collected fingerprint images and accurately extract the key features of the ridges and valleys in the fingerprints; In the third step, the comparison module compares the extracted fingerprint features with the fingerprint template stored in the database; Step 4: The decision module determines the fingerprint matching situation based on the comparison results. If the match is successful, the decision module issues an unlock command. If the match is unsuccessful, the decision module does not issue an unlock command. If the match is unsuccessful for multiple times, the decision module issues an alarm command. The number of multiple unmatched times here can be freely set by the system. The second situation: the fingerprint contains water, dust and fingerprint fracture phenomenon; S10, transmitting the collected clear part of the image to the feature extraction module through the fingerprint collection module; S20, the feature extraction module determines that the information is a partially clear image, and sends the determination result to the decision module, and the decision module determines the time relationship between the unlocking time at this time and the regular unlocking time period through the time rule analysis module; S30, the decision module determines through the time rule analysis module that the unlocking time is within the regular unlocking time period, then selects to use the unlocking method based on the time rule, and sends the decision instruction to the feature extraction module; The decision module determines through the time rule analysis module that the unlocking time is not within the regular unlocking time period, and then chooses to use the unlocking method based on deep learning, and sends the decision instruction to the feature extraction module; S40, the unlocking method based on time rule includes the following steps: S401, based on the first and second steps above, the collected partially clear image is transmitted to the feature comparison module through the feature extraction module; S402 The feature comparison module compares the received partially clear image with the complete information image stored in the database, and sends the comparison result to the decision module; S403 If the comparison result of the clear image of the part matches the same part of the complete information image in the database, the decision module issues an unlocking instruction; if the comparison result of the clear image of the part does not match the same part of the complete information image in the database, the decision module does not issue an unlocking instruction; if the matching fails multiple times, the decision module issues an alarm instruction, and the number of multiple mismatches here can be freely set by the system; S50, the unlocking method based on deep learning includes the following steps: S501, based on the first and second steps above, the collected partially clear image is inferred and compensated by the feature extraction module to restore the complete fingerprint information image and transmit it to the feature comparison module; S502, the feature comparison module compares the restored complete fingerprint information image with the complete information image stored in the database, and sends the comparison result to the decision module; S503. If the comparison result between the restored complete fingerprint information image and the complete information image in the database matches, an unlocking command is issued through the decision module. If the comparison result between the restored complete fingerprint information image and the complete information image in the database does not match, no unlocking command is issued through the decision module. If multiple matches are unsuccessful, an alarm command is issued through the decision module. The number of multiple mismatches here can be freely set by the system.

3. A biometric feature comparison system according to claim 2, characterized in that: The image preprocessing steps in the two steps are as follows: Step 1: Use the weighted average method to calculate the final grayscale value based on the contribution of the three color channels of red, green and blue to the grayscale value, and convert the color fingerprint image into a grayscale image; Step 2: Adjust the fingerprint images collected from different devices and environments to a uniform size and orientation; Step 3: Use filtering algorithm to improve the clarity and contrast of fingerprint image; Step 4: Use the adaptive threshold method to convert the enhanced grayscale image into a binary image containing only black and white colors; Step 5: Use iterative stripping method to thin the lines in the binary fingerprint image into lines with a single pixel width; Step 6: Further eliminate existing noise points and false features.

4. A biometric feature comparison system according to claim 3, characterized in that: The deep learning feature extraction working steps in the three steps are as follows: S1. Establish a deep neural network model, which consists of multiple hidden layers, each hidden layer contains multiple neurons, and the neurons are interconnected through weights and biases. The weights and biases are parameters that need to be learned during the training process of the model; S2, the model randomly initializes weights and biases, and sends data to the neural network for forward propagation calculation. The input data will pass through each layer of the model, and generate output through weighted summation and activation function operations; S3. The model compares the output calculated by forward propagation with the true label and calculates the loss function. The loss function is used to measure the difference between the model's predicted results and the true results. The model calculates the gradient of the loss function with respect to each weight and bias, and updates the weights and biases based on these gradients. S4. Repeat the above steps S2 and S3 to form an iterative process. In each iteration, the model will perform forward propagation based on the current weights and biases, calculate the loss function, and then perform backpropagation and update the weights and biases. This process will continue until the preset number of iterations is reached and the loss function value drops below the preset threshold of 0.

1.

5. A biometric feature comparison system according to claim 4, characterized in that: The calculation formula of the weighted sum in step S2 is: ,in: It is The weight of the input, It is Inputs, is the bias term, is the result of weighted summation.

6. A biometric feature comparison system according to claim 5, characterized in that: The loss function calculation formula in the S3 step is as follows: ,in: is the value of the loss function, is the number of samples, It is The true label of a sample is a binary variable. The model is The probability that a sample is predicted to be the positive class.

7. A biometric feature comparison system according to claim 6, characterized in that: The specific implementation of the weighted average method in the step 1 is calculated by the formula gray value = 0.299×R+0.587×G+0.114×B.

8. A biometric feature comparison system according to claim 7, characterized in that: The working steps of the adaptive threshold method in step 4 to binarize the grayscale image are as follows: S221, reading a grayscale image to be processed, which includes a pre-processed grayscale image; S222, calculating local statistics of the image, including local mean and local standard deviation, by applying a rectangular region of fixed size to the image, and calculating statistical characteristics of the local neighborhood of each pixel in the rectangular region; S223, based on the above local statistics, determining a threshold for each pixel and each local area, where the threshold is a fixed offset based on the grayscale mean or a dynamic value based on the standard deviation; S224, comparing the gray value of each pixel in the image with the corresponding local threshold, setting the pixels whose gray value is greater than or equal to the threshold to white, and setting the pixels whose gray value is less than the threshold to black; S225, through the above steps, the original grayscale image is converted into a binary image, which contains only black and white colors and is used to represent fingerprint lines and other features; S226, removing noise and filling holes from the binary image to further improve the binarization result; S227. Output and save the generated binary image.

Images