An identity recognition method and system thereof
Feature vectors are extracted through capacitive screen acquisition and deep learning methods, which solves the problem of the need for specialized sensors in traditional biometric recognition technology, realizes sensorless user experience and widely used biometric recognition, significantly improving ease of use and security.
Patent Information
- Application Number
- CN202011601423.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-12-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-12-29
AI Technical Summary
Traditional biometric recognition technology requires special sensors, and user operations are interrupted during use, poor user experience, and limited application scenarios.
The capacitance screen collects time series of changes in capacitance value caused by user sliding, performs preprocessing and deep learning methods to extract feature vectors, compares them with the data entered in advance, and realizes biometric recognition. The capacitance screen device requires no additional sensors, and supports a wide range of application.
It realizes biometric recognition without special sensors, user operations are not disturbed, application scenarios are wide, and ease of use and security are significantly improved.
Smart Images

Figure CN112613019B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to an identity recognition method and system thereof. Background Art
[0002] With the continuous improvement of the level of social informatization, people generate more and more private data. Protecting the security of this information is an increasing demand. However, traditional identity authentication methods, such as keys, PINs, combination passwords, etc., are becoming less and less suitable for data protection applications due to the problems of being easily obtained by attackers and easily forgotten by users. Therefore, in recent years, biometric recognition has played an increasingly important role in the security field. From the fingerprint recognition and vein recognition technologies used in special places such as banks in the early days, to the relatively natural and non-sensory face recognition technology widely used in personal devices today, biometric recognition technology has made great progress. The so-called biometrics refers to the measurable physiological or behavioral characteristics that can uniquely identify an individual. Biometric recognition gets rid of passwords and media, and uses people themselves as passwords. It not only naturally solves the problem of forgotten passwords or lost media, but also greatly increases the difficulty of obtaining authentication information, making the ease of use and security significantly improved. The introduction of artificial intelligence technology has greatly enhanced its recognition accuracy and robustness. Among them, fingerprint recognition and face recognition are currently the most common. Fingerprint recognition is widely used in personal devices, banks, customs and other scenarios with extremely high security and recognition rate requirements. This method has excellent recognition effect, is safe and reliable, and the false recognition rate and rejection rate are less than one thousandth. Due to its robustness and high recognition effect, the face recognition system is mainly used in access control, security and other fields, and is also used in some personal devices.
[0003] Currently commonly used biometric recognition technologies, such as fingerprint recognition and face recognition, usually require the use of special sensors to collect data, which leads to limited use scenarios of these technologies. At the same time, during use, the user's operation will be completely interrupted by the authentication process, and they have to pay attention to the corresponding sensor and the authentication process itself. The whole process is highly involved and the user experience is relatively poor. Summary of the invention
[0004] The purpose of this application is to provide an identity recognition method and system, which have all the advantages of biometric recognition, without worrying about forgetting passwords or losing media, and have higher ease of use and security. At the same time, it has a wider range of applications and can be used on any device with a capacitive screen without relying on other sensors. In addition, the user's normal operation will not be interrupted during use, and a good user experience is taken into account.
[0005] To achieve the above object, the present application provides an identity recognition method, including the following steps: collecting original data, where the original data is a sequence of capacitance value matrices; preprocessing and encoding the original data to obtain multiple feature vectors; comparing the multiple feature vectors with a user template one by one, and obtaining an authentication result through a pre-trained discriminator model.
[0006] As described above, before collecting the original data, it is necessary to first log in to the system through a pre-registered account.
[0007] As described above, the sub-steps of pre-registering an account are as follows: obtaining multiple sets of sliding data according to an instruction; performing an input check on each set of sliding data. If the input is checked to be correct, process each set of sliding data to obtain a feature vector; if the input is checked to be incorrect, prompt to re-collect the incorrect sliding data; take the mean of all feature vectors to obtain a user template, and complete the account registration.
[0008] As described above, the sub-steps of preprocessing and encoding the original data to obtain multiple feature vectors are as follows: performing data preprocessing on the original data to obtain preprocessed data; inputting the preprocessed data into a pre-trained encoder model to obtain multiple feature vectors.
[0009] As described above, the sub-steps of performing data preprocessing on the original data to obtain preprocessed data are as follows: subtracting the background noise matrix from the original data to generate a subtraction result, and dividing the subtraction result by the capacitance value change range of the capacitive screen to obtain a denoised and normalized sequence of capacitance value matrices; finding all contacts greater than a threshold r in each frame of the normalized sequence of capacitance value matrices, and regarding the contact as a touch point; selecting a k*k range in each frame of the normalized sequence of capacitance value matrices such that the touch point is at a corner within the k*k range, and then saving the k*k matrix to obtain a time series of k*k matrices, and using the time series of k*k matrices as the preprocessed data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
[0010] As described above, the sub-steps of pre-training the encoder model are as follows: recruiting subject samples, and collecting all capacitance value sequences generated when the subject samples use a capacitive screen device or a biometric recognition device through multiple sliding modes; preprocessing the collected all capacitance value sequences to obtain processed data; constructing a three-dimensional convolutional neural network model, inputting the processed data of all subject samples for training, and outputting multiple feature vectors to complete the training; using an encoder optimization model to optimize the trained three-dimensional convolutional neural network model, and using the optimized three-dimensional convolutional neural network model as the pre-trained encoder model.
[0011] As above, the sub-steps of pre-training the discriminator model are as follows: input the processed data into the encoder model to obtain a standard feature vector; construct a twin neural network, use a fully connected network to encode the input newly collected data, obtain a new feature vector, input the standard feature vector and the new feature vector into the twin neural network, and the twin neural network outputs the difference between the standard feature vector and the new feature vector; use the discriminator optimization model to optimize the twin neural network, and use the optimized twin neural network as the trained discriminator model.
[0012] The present application also provides an identity recognition system, including: a capacitive screen device and a server; wherein the capacitive screen device is used to collect original data and send the original data to the server; the server is used to receive the original data and execute the above-mentioned identity recognition method to generate an authentication result.
[0013] As above, the server includes: an encoder and a discriminator; wherein the encoder is used to receive preprocessed data and obtain multiple feature vectors; the discriminator is used to receive multiple feature vectors and obtain authentication results.
[0014] As mentioned above, the capacitive screen device includes at least: an authentication system and a capacitive screen; wherein the capacitive screen is used to generate raw data; and the authentication system is used to collect raw data, process the raw data, and obtain pre-processed data.
[0015] The beneficial effects achieved by this application are as follows:
[0016] (1) The present invention collects the time series of capacitance value changes caused by the user sliding on the screen during the use of the device, extracts feature vectors after preprocessing using a deep learning method, and obtains authentication results by comparing them with the features of pre-entered data, thereby realizing a complete biometric recognition process.
[0017] (2) In terms of hardware equipment, the present invention only requires a device with a capacitive screen, and its application scenarios are more extensive than those of fingerprint recognition and face recognition. At the same time, in terms of data collection, the present invention silently collects data during the user operation process and will not interfere with the user's use, thereby greatly enhancing the user experience.
[0018] (3) The present invention can continuously collect data and perform authentication during user use, which significantly improves ease of use and security. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0020] Figure 1 A schematic diagram of the structure of an embodiment of an identity recognition system;
[0021] Figure 2 A schematic diagram of an embodiment of an identity recognition method. DETAILED DESCRIPTION
[0022] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.
[0023] like Figure 1 As shown, as an embodiment, the present application provides an identity recognition system, including: a capacitive screen device 1 and a server 2.
[0024] Among them, the capacitive screen device 1 is used to collect raw data and send the raw data to the server. Specifically, the raw data is a capacitance value matrix sequence generated by the touch operation.
[0025] Server 2: used to receive the original data and execute the following identity recognition method to generate the authentication result.
[0026] Furthermore, the capacitive screen device 1 at least includes: an authentication system 110 and a capacitive screen 120 .
[0027] The authentication system 110 is used to collect raw data, process the raw data, and obtain pre-processed data.
[0028] Capacitive screen 120: used to generate raw data.
[0029] Furthermore, the server 2 includes: an encoder 210 and a discriminator 220 .
[0030] The encoder 210 has an encoder model for receiving preprocessed data and obtaining a plurality of feature vectors.
[0031] Specifically, the encoder 210 is used to extract feature vectors that can be used to distinguish users from the preprocessed data.
[0032] Discriminator 220: having a discriminator model, used for receiving multiple feature vectors and obtaining an authentication result.
[0033] Specifically, the discriminator 220 is used to compare the standard feature vector of the capacitance matrix sequence previously input by the user with the feature vector of the newly collected data, and then obtain the authentication result.
[0034] As another embodiment, the present application provides a biometric identification device, including: a capacitive screen, an authentication system, an encoder and a discriminator.
[0035] The capacitive screen is used to generate raw data, which is a capacitance value matrix sequence generated by the touch operation.
[0036] Authentication system: used to collect raw data, process the raw data, and obtain pre-processed data.
[0037] Encoder: used to receive preprocessed data and obtain multiple feature vectors.
[0038] Discriminator: used to receive multiple feature vectors and obtain authentication results.
[0039] like Figure 2 As shown, the present application provides an identity recognition method, comprising the following steps:
[0040] S1: Collecting raw data, where the raw data is a capacitance value matrix sequence.
[0041] Specifically, the original data is a matrix sequence of capacitance values generated by sliding in various states that is continuously collected by the authentication system in the background when the user uses a capacitive screen in a scenario with high information security requirements.
[0042] Among them, assume that the capacitance sampling density of the capacitive screen is m*n (m>n), and the sampling frequency is f: the capacitance data acquisition mechanism starts recording from the time the user presses his finger on the screen until the user lifts his finger. Each frame of the collected raw data is an m*n matrix, where m represents the length of the capacitance matrix and n represents the width of the capacitance matrix.
[0043] Furthermore, before collecting raw data, you need to log in to the system through a pre-registered account.
[0044] Specifically, the user logs into the authentication system using a pre-registered account through the capacitive screen.
[0045] Furthermore, the sub-steps for pre-registering an account are as follows:
[0046] P1: Get multiple sets of sliding data according to the instructions.
[0047] Specifically, during the registration process, the user receives instructions through the capacitive screen and enters multiple groups of sliding operations in different states according to the instructions. The authentication system collects the data of multiple groups of sliding operations in the background and uses the collected data as sliding data to execute P2.
[0048] P2: Perform input checks on each group of sliding data. If the input is found to be correct, process each group of sliding data to obtain feature vectors; if an input error is detected, prompt the user to recollect the incorrect sliding data.
[0049] Furthermore, the sub-steps for processing each group of sliding data to obtain feature vectors are as follows:
[0050] P210: Preprocess each group of sliding data to obtain standard data.
[0051] Specifically, the sub-steps for preprocessing each group of sliding data to obtain standard data are as follows:
[0052] Q1: Subtract the background noise matrix from each group of sliding data to generate a subtraction result, and divide the subtraction result by the capacitance value change range of the capacitive screen to obtain a denoised and normalized capacitance value matrix sequence.
[0053] Specifically, the background noise matrix and the capacitance value change range are usually provided by the manufacturer of the capacitive screen.
[0054] Q2: Find all contacts greater than the threshold r in each frame of the normalized capacitance value matrix sequence, and regard the contact as a touch point.
[0055] Specifically, the threshold r is preset, and the specific value of the threshold r depends on the actual situation.
[0056] Q3: Select a k*k range in each frame of the normalized capacitance value matrix sequence such that after the touch point is located at a corner within the k*k range, save the k*k matrix to obtain a time series of k*k matrices, and use the time series of k*k matrices as standard data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
[0057] Specifically, the touch point can be located at the upper left corner, upper right corner, lower left corner, or lower right corner within the k*k range. Among them, the specific position of the touch point can be determined according to the actual situation.
[0058] P220: Input all the standard data into a pre-trained encoder model to obtain multiple feature vectors.
[0059] P3: Take the mean of all the feature vectors to obtain a user template and complete the account registration.
[0060] Specifically, when the number of multiple sets of sliding data obtained according to the instruction is w groups, all the preprocessed data is input into a pre-trained encoder model, and the number of multiple feature vectors obtained is w.
[0061] S2: Perform preprocessing and encoding on the original data to obtain multiple feature vectors.
[0062] Further, the sub-steps of performing preprocessing and encoding on the original data to obtain multiple feature vectors are as follows:
[0063] L1: Perform data preprocessing on the original data to obtain preprocessed data.
[0064] Specifically, the sub-steps of performing data preprocessing on the original data to obtain preprocessed data are as follows:
[0065] W1: Subtract the background noise matrix from the original data to generate a subtraction result, and divide the subtraction result by the capacitance value change range of the capacitive screen to obtain a denoised and normalized capacitance value matrix sequence.
[0066] Specifically, the background noise matrix and the capacitance value change range are usually provided by the manufacturer of the capacitive screen.
[0067] W2: Find all contacts greater than the threshold r in the normalized capacitance value matrix sequence of each frame, and regard the contact as a touch point.
[0068] Specifically, the threshold r is preset, and the specific value of the threshold r depends on the actual situation.
[0069] W3: Select a k*k range in the normalized capacitance value matrix sequence of each frame so that the touch point is located at a corner within the k*k range, then save the k*k matrix to obtain a time series of k*k matrices, and use the time series of k*k matrices as the preprocessed data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
[0070] Specifically, the touch point can be located at the upper left corner, upper right corner, lower left corner or lower right corner, etc. within the k*k range. Among them, the specific position of the touch point can be determined according to the actual situation.
[0071] L2: Input the preprocessed data into a pre-trained encoder model to obtain multiple feature vectors.
[0072] Further, the sub-steps of pre-training the encoder model are as follows:
[0073] F1: Recruit subject samples and collect all capacitance value sequences generated when the subject samples use a capacitive screen device or a biometric recognition device through various sliding modes.
[0074] Specifically, the multiple sliding modes mainly include: the sliding operations of the subject's finger in four directions of up, down, left, and right on the capacitive screen, and the zoom operation used when zooming in or out of a picture. The capacitance value sequence corresponds one-to-one with the zoom operation. Among them, the capacitance value sequence is a capacitance value matrix.
[0075] F2: Preprocess all the collected capacitance value sequences to obtain processed data.
[0076] Furthermore, the sub-steps of preprocessing all the collected capacitance value sequences to obtain processed data are as follows:
[0077] F210: Subtract the background noise matrix from each frame of the capacitance value sequence to generate a subtraction result, and divide the subtraction result by the capacitance value change range of the capacitive screen to obtain a denoised and normalized capacitance value matrix sequence.
[0078] Specifically, the background noise matrix and the capacitance value change range are usually provided by the manufacturer of the capacitive screen.
[0079] F220: Find all contacts greater than the threshold r in each frame of the normalized capacitance value matrix sequence, and regard the contact as a touch point.
[0080] Specifically, the threshold r is preset, and the specific value of the threshold r depends on the actual situation.
[0081] F230: Select a range of k*k in each frame of the normalized capacitance value matrix sequence. After the touch point is located at a corner within the k*k range, save the k*k matrix to obtain a time series of the k*k matrix, and use the time series of the k*k matrix as the processed data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
[0082] Specifically, the touch point can be located at the upper left corner, upper right corner, lower left corner, or lower right corner within the k*k range. Among them, the specific position of the touch point can be determined according to the actual situation.
[0083] F3: Construct a three-dimensional convolutional neural network (3DCNN) model, input the processed data of all subject samples for training, and output multiple feature vectors to complete the training.
[0084] F4: Use an encoder to optimize the trained three-dimensional convolutional neural network (3DCNN) model, and use the optimized three-dimensional convolutional neural network (3DCNN) model as the trained encoder model.
[0085] Specifically, as an embodiment, the encoder optimization model is a Triplet Loss optimization model, and the Triplet Loss optimization model is used for optimization. The three-dimensional convolutional neural network (3DCNN) model is optimized in the direction of reducing the feature vector distance of the same user data and increasing the feature vector distance of different user data, and the optimized three-dimensional convolutional neural network (3DCNN) model is used as the trained encoder model. Optimization is conducive to improving model performance.
[0086] S3: Compare multiple feature vectors with the user template one by one, and obtain the authentication result through the pre-trained discriminator model.
[0087] Specifically, as an embodiment, multiple feature vectors are compared with the user template one by one, and the authentication result is obtained through a pre-trained discriminator model to achieve a continuous authentication effect.
[0088] As another embodiment, several authentication results may be accumulated and voted to generate a final result, and the final structure may be used as the final authentication result, sacrificing a certain authentication efficiency to increase accuracy.
[0089] Furthermore, the sub-steps of pre-training the discriminator model are as follows:
[0090] K1: Input the processed data into the encoder model to obtain the standard feature vector.
[0091] Specifically, the processed data is obtained by preprocessing the capacitance value sequence of the test sample.
[0092] K2: Build a twin neural network (SNN architecture), use a fully connected network to encode the newly collected input data, obtain a new feature vector, input the standard feature vector and the new feature vector into the twin neural network, and the twin neural network outputs the difference between the standard feature vector and the new feature vector.
[0093] Specifically, the standard feature vector is a feature vector of pre-entered data, that is, a PPG data template of a user, which represents the average value of feature vectors of multiple PPG signals of the user. The new feature vector is a feature vector of newly collected data.
[0094] K3: Use the discriminator optimization model to optimize the twin neural network, and use the optimized twin neural network as the trained discriminator model.
[0095] Specifically, as an embodiment, the discriminator optimization model is a Contrastive Loss optimization model.
[0096] The beneficial effects achieved by this application are as follows:
[0097] (1) The present invention collects the time series of capacitance value changes caused by the user sliding on the screen during the use of the device, extracts feature vectors after preprocessing using a deep learning method, and obtains authentication results by comparing them with the features of pre-entered data, thereby realizing a complete biometric recognition process.
[0098] (2) In terms of hardware equipment, the present invention only requires a device with a capacitive screen, and its application scenarios are more extensive than those of fingerprint recognition and face recognition. At the same time, in terms of data collection, the present invention silently collects data during the user operation process and will not interfere with the user's use, thereby greatly enhancing the user experience.
[0099] (3) The present invention can continuously collect data and perform authentication during user use, which significantly improves ease of use and security.
[0100] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic creative concepts. Therefore, the scope of protection of the present application is intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application. Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the protection scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.
Claims
1. An identity recognition method, characterized in that: It includes the following steps: Collect original data, where the original data is a sequence of capacitance value matrices generated by the authentication system in the background during sliding under various states when the user uses a capacitive screen; Perform preprocessing and encoding on the original data to obtain multiple feature vectors; Compare each of the multiple feature vectors with the user template one by one, and obtain the authentication result through a pre-trained discriminator model, or vote to generate a final result after accumulating multiple authentication results, and use the final result as the authentication result; Among them, the sub-steps of performing preprocessing and encoding on the original data to obtain multiple feature vectors are as follows: Perform data preprocessing on the original data to obtain preprocessed data; Input the preprocessed data into a pre-trained encoder model to obtain multiple feature vectors; Among them, the sub-steps of pre-training the encoder model are as follows: Recruit subject samples, and collect all capacitance value sequences generated when the subject samples use a capacitive screen device or a biometric recognition device through various sliding modes; Perform preprocessing on all the collected capacitance value sequences to obtain processed data; Construct a three-dimensional convolutional neural network model, input the processed data of all subject samples for training, and output multiple feature vectors to complete the training; Use an encoder optimization model to optimize the trained three-dimensional convolutional neural network model, and use the optimized three-dimensional convolutional neural network model as the trained encoder model; Among them, the sub-steps of performing preprocessing on all the collected capacitance value sequences to obtain processed data are as follows: Subtract the background noise matrix from each frame of capacitance value sequence to generate a subtraction result, and divide the subtraction result by the capacitance value change range of the capacitive screen to obtain a noise-reduced and normalized capacitance value matrix sequence; Find all contacts greater than the threshold r in each frame of the normalized capacitance value matrix sequence, and regard the contact as a touch point; Select a k*k range in each frame of the normalized capacitance value matrix sequence so that the touch point is at a corner within the k*k range, then save the k*k matrix to obtain a time series of k*k matrices, and use the time series of k*k matrices as the processed data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
2. The identity recognition method according to claim 1, characterized in that: Before collecting the original data, it is necessary to first log in to the system through a pre-registered account.
3. The identity recognition method according to claim 2, characterized in that: The sub-steps of pre-registering an account are as follows: Obtain multiple groups of sliding data according to the instructions; Perform input checks on each group of sliding data. If the input is checked to be correct, process each group of sliding data to obtain feature vectors; if the input is checked to be incorrect, prompt to re-collect the incorrect sliding data; Take the mean of all feature vectors to obtain the user template and complete the account registration.
4. The identity recognition method according to claim 1, characterized in that: The sub-steps of performing data preprocessing on the original data to obtain preprocessed data are as follows: Subtract the background noise matrix from the original data to generate a subtraction result, and divide the subtraction result by the capacitance value change range of the capacitive screen to obtain a noise-reduced and normalized capacitance value matrix sequence; Find all contacts greater than the threshold r in each frame of the normalized capacitance value matrix sequence, and regard the contact as a touch point; In the normalized capacitance value matrix sequence of each frame, select a range of k*k, so that after the touch point is located at a corner within the k*k range, save the k*k matrix to obtain a time series of the k*k matrix, and use the time series of the k*k matrix as the preprocessed data, where k is a natural number, k < n, and n is the width of the capacitance matrix.
5. The identity recognition method according to claim 1, characterized in that: The sub-steps for pre-training the discriminator model are as follows: Input the processed data into the encoder model to obtain a standard feature vector; Construct a siamese neural network, use a fully connected network to encode the newly acquired input data to obtain a new feature vector, input the standard feature vector and the new feature vector into the siamese neural network, and the siamese neural network outputs the difference between the standard feature vector and the new feature vector; Use the discriminator optimization model to optimize the siamese neural network, and use the optimized siamese neural network as the trained discriminator model.
6. An identity recognition system, characterized in that: Including: A capacitive touch screen device and a server; Wherein, the capacitive touch screen device is used to collect raw data and send the raw data to the server; The server is used to receive the raw data and execute the identity recognition method as described in any one of claims 1-5 to generate an authentication result.
7. The identity recognition system according to claim 6, characterized in that: The server includes: an encoder and a discriminator; Wherein, the encoder is used to receive the preprocessed data and obtain a plurality of feature vectors; The discriminator is used to receive the plurality of feature vectors and obtain an authentication result.
8. The identity recognition system according to claim 7, characterized in that: The capacitive touch screen device at least includes: an authentication system and a capacitive screen; Wherein, the capacitive screen is used to generate raw data; The authentication system is used to collect the raw data, process the raw data, and obtain the preprocessed data.
