A mobile phone user touch screen behavior identity authentication method and related device

Abstract

A mobile phone user touch screen behavior identity authentication method and related device, comprising: data preprocessing on the original measurement data of a capacitive touch screen sensor; extracting the intra-gesture spatiotemporal features of touch gestures in the preprocessed data based on a 3D CNN architecture; modeling the patterns between touch gestures based on the intra-gesture spatiotemporal features of touch gestures using a long short-term memory network (LSTM); and training and testing the model. The present application extracts the intra-gesture spatiotemporal features of touch gestures based on a 3D CNN architecture and models the patterns between gestures using a long short-term memory network (LSTM), which can better fit the actual verification environment and bring greater improvement to the continuous authentication of mobile phone user identity.

Description

Technical Field

[0001] This invention belongs to the field of information technology, specifically relating to a method and related device for authenticating the identity of mobile phone users through touchscreen behavior. Background Technology

[0002] With the increasing popularity of mobile phones, more and more users are enjoying the convenience that mobile phones bring to their lives. However, at the same time, more and more privacy leaks are emerging. Mobile phones store a large amount of users' private and sensitive data, such as photos and emails. Preventing this data from being accessed by attackers is of great importance.

[0003] Currently, obtaining touchscreen biometrics from the underlying touch interface is a promising method for mobile phone user authentication. For example, when a part of a user's body (hand, ear, etc.) touches the touchscreen, the electrostatic field of the contact area on the capacitive touchscreen will be distorted, and the capacitance value of the contact area will also change accordingly, which can be used for user authentication.

[0004] However, the above methods are suitable for specific experimental scenarios, such as correctly placing one's hand on a touchscreen. In everyday, unconstrained conditions, it is difficult to perform accurate authentication because actual touch interaction involves the cross-use of various types of touch gestures, such as typing and swiping. However, existing technologies do not address the various interference scenarios in such real-world environments. Summary of the Invention

[0005] The purpose of this invention is to provide a method and related device for authenticating the identity of mobile phone users' touch screen behavior, so as to solve the problem that the existing technology cannot accurately verify the identity of users under various interference conditions in the actual touch interaction process, which involves the cross-use of various types of touch gestures.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for authenticating mobile phone user touchscreen behavior includes:

[0008] Perform data preprocessing on the raw measurement data of the capacitive touchscreen sensor;

[0009] This method uses a 3D CNN architecture to extract the spatiotemporal features of touch gestures from preprocessed data.

[0010] Based on the spatiotemporal features of touch gestures, a Long Short-Term Memory (LSTM) network is used to model the patterns between touch gestures.

[0011] Train and test the model.

[0012] Further data preprocessing includes: difference operations, threshold setting, and resampling;

[0013] Differential operation: Perform differential operation on each touch screen, specifically by reconstructing each touch frame by calculating the difference between the baseline measurement and the changed measurement of all cells in the grid;

[0014] Threshold setting: Set the threshold μ when the touch box x ij If cell x is greater than μ, keep the original measurement; otherwise, set the value to 0.

[0015] Resampling: For lengths less than For touch gestures, add a touch box of all zeros to obtain a touch box of a specific length; for lengths greater than... For touch gestures, downsample touch gesture X to obtain the resampled length. Touch gestures For a touch gesture X with a touch frame of length l, the start and end times are first determined based on the first and last touch frames of X; then, linear interpolation is performed between the start and end times to obtain... The sampling time for each touch frame; then, for an interpolated sampling time... Search for the touch frame with the closest timestamp in X. t and x t+1 ,in Finally, we obtain the result based on x. t and x t+1 Resampled touch frames

[0016] Furthermore, the spatiotemporal features of touch gestures within the preprocessed data are extracted:

[0017] Three-dimensional convolution operations are applied to the convolutional layers of the CNN model to simultaneously capture the appearance and outline of the fingertip of the touch point in the touch frame, as well as the dynamic sliding trajectory in the continuous touch frames; the input of the 3D CNN model is the touch gesture; the output of the model is the probability of the touch gesture belonging to a legitimate user or an intruder.

[0018] In each convolutional layer, 3D convolutional filtering is used to perform convolution operations on the feature maps. Each feature map is connected to multiple touch boxes in the previous layer to form a spatiotemporal cube. By applying 3D convolutional filters on these cubes, the 3D CNN model captures local features of different spatial regions of the touch frame and dynamic sliding information between touch frames.

[0019] After each convolution operation, a transformation operation is performed on each feature map. By performing convolution filtering and nonlinear transformation on the low-level feature maps, the number of feature maps in subsequent layers is gradually increased to generate high-level features.

[0020] For pooling layers, 3D pooling operations are used to capture features at different spatial scales by downsampling the feature map;

[0021] By alternately applying the above operations, the 3D CNN model converts the input touch gesture into a high-dimensional feature vector that captures the appearance and contour of the contact's fingertip, as well as dynamic swiping information of the fingertip movement encoded in consecutive touch frames.

[0022] Furthermore, the value at coordinate (x, y, z) in the j-th feature map of the i-th layer. The calculation formula is:

[0023]

[0024] Where H i W i D i Indicate i th The height, width, and temporal depth of the 3D convolutional filter layer; M is the number of feature maps; Indicates connection to the previous layer k th The values ​​at the filter (h,w,d) positions of the feature map.

[0025] Furthermore, the LSTM model consists of n identical networks R, which learn a nonlinear function. Touch gesture features X1…X n Transform into a hidden vector sequence h1…h n The hidden vectors store historical touch gesture information;

[0026] At each step i (i = 1, ..., n), the network will store the last hidden vector h. i-1 and the characteristics X of the current touch gesture i As input, and output a new hidden vector h i ;

[0027] Once the number of touch gestures reaches the predetermined number n, the recursive layer h... n The final output will be fed back to the output layer;

[0028] The output layer contains a fully connected layer and a Softmax layer for generating authentication results; the fully connected layer maps the learned touch gesture representations to a two-dimensional output; then, the Softmax layer uses the Sigmoid function to normalize the output to a value between 0 and 1, representing the probability that the touch gesture sequence belongs to a legitimate user.

[0029] Further, training and testing: The model is evaluated using a sequential training and testing strategy. By introducing the temporal relationship between the training dataset and the test dataset, an authentication model based on historical touch data is established, and new touch gestures are predicted.

[0030] The specific training and testing process is as follows:

[0031] Step 1: Use the touch gestures from the first six subsequences of legitimate users to construct positive training and test sets to train the CNN model, and use the eighth subsequence as the test set; divide the subsequences into touch trajectories of length n, where the length of the user's touch trace is determined by the number n of touch gestures used for decision-making.

[0032] Step 2: Construct negative training and test sets to train the CNN model; construct the negative training dataset by merging the first 6 subsequences of each remaining topic into a negative set, obtaining the same number of negative training touch trajectories as the positive training dataset; merge the eighth subsequence of the intruder into the negative verification dataset; maintain the same number of touch trajectories in the negative verification dataset and the positive verification dataset by random sampling.

[0033] Step 3: Construct a positive test dataset using the last two sub-sequences of the legitimate subject's gesture to test the recognition capability of legitimate subjects;

[0034] Step 4: Extract test data of the same size from the remaining subjects to construct a negative test dataset and test its ability to identify intruders.

[0035] Furthermore, three metrics were used to evaluate the certification quality of all methods.

[0036] Metric 1: False Acceptance Rate (FAR): Assesses the probability that an intruder is verified as a legitimate user; High FAR indicates that the authentication system will mistakenly allow an unauthorized user to use the smartphone;

[0037] Metric 2: False Rejection Rate (FRR): Assessing the probability that a legitimate user is verified as an intruder; a higher FRR indicates that authorized users will be denied access to the smartphone.

[0038] Indicator 3: Equal Error Rate (EER): This is the rate at which the FAR and FRR values ​​are equal; the lower the EER value, the higher the accuracy.

[0039] Furthermore, a mobile phone user touchscreen behavior authentication system includes:

[0040] The preprocessing module is used to preprocess the raw measurement data of the capacitive touch screen sensor.

[0041] The feature extraction module is used to extract the spatiotemporal features of touch gestures in the preprocessed data based on the 3D CNN architecture.

[0042] The modeling module is used to model the spatiotemporal features within touch gestures, employing a Long Short-Term Memory (LSTM) network to model the patterns between touch gestures.

[0043] The testing module is used for training and testing the model.

[0044] Furthermore, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a mobile phone user touchscreen behavior authentication method.

[0045] Furthermore, a computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a mobile phone user touchscreen behavior authentication method.

[0046] Compared with the prior art, the present invention has the following technical effects:

[0047] This invention provides a method for continuous authentication of mobile phone users based on a deep learning 3D convolutional neural network (3D CNN). It collects raw capacitance data from the touchscreens of smartphones used daily by 100 volunteers, ensuring the experimental data closely reflects the actual verification environment. The raw capacitance data undergoes preprocessing to accurately simulate user touch behavior, including differential operations, threshold setting, and resampling. Differential operations alleviate the inaccuracy of baseline capacitance values ​​in capacitive touchscreens, threshold setting removes noisy data, and resampling processes each touch gesture into a fixed and finite length. Based on the 3D CNN architecture, the spatiotemporal features within the touch gestures are extracted, representing the fine-grained appearance and contour of the touching fingertips and the dynamic sliding information within the touch gestures. A Long Short-Term Memory (LSTM) network is used to model the patterns between gestures to authenticate whether newly introduced touch gestures originate from legitimate users, resulting in a significant improvement in continuous authentication of mobile phone users. Attached Figure Description

[0048] Figure 1 This is a flowchart of a mobile phone user touch screen behavior authentication method based on a deep learning three-dimensional convolutional neural network according to the present invention. Detailed Implementation

[0049] This invention provides a method for continuous authentication of mobile phone users based on a deep learning 3D convolutional neural network. The method includes: preprocessing raw measurement data from a capacitive touchscreen sensor to accurately simulate user touch behavior; developing a 3D CNN-based architecture to extract spatiotemporal features within touch gestures; using Long Short-Term Memory (LSTM) to model inter-gesture patterns; and training and testing the model. (Reference) Figure 1 .

[0050] Specifically, the following steps are included:

[0051] Step one involves preprocessing the raw measurement data from the capacitive touchscreen sensor to accurately simulate the user's touch behavior.

[0052] Step 2: Develop a 3D CNN-based architecture to extract the spatiotemporal features within touch gestures.

[0053] Step 3: Use Long Short-Term Memory (LSTM) to model the patterns between gestures.

[0054] Step four: Train and test the model.

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings.

[0056] The raw measurement data of the capacitive touch screen sensor is preprocessed to accurately simulate the user's touch behavior. The data preprocessing consists of three steps: differential operation, threshold setting, and resampling.

[0057] Differential operation. Differential operation is performed on each touchscreen to alleviate the problem of inaccurate baseline capacitance values ​​in capacitive touchscreens. Specifically, each touch frame is reconstructed by calculating the difference between the baseline measurement and the changed measurement of all cells in the grid.

[0058] Threshold setting. Considering the influence of other factors on capacitive touchscreen measurements, such as sensor drift caused by changes in ambient temperature and humidity, it is necessary to eliminate unexpected interference from reconstructing capacitance values ​​at non-contact locations on the touchscreen. To remove noisy data, a threshold μ is set. When the touch frame x... ij If cell x is greater than μ, the original measurement is retained; otherwise, the value is set to 0. Empirically, μ is set to 50 in the implementation.

[0059] Resampling. Since the authentication model requires training with touch gestures of fixed length, fixed and finite lengths are used. Preprocess each touch gesture. For gestures shorter than [length missing], [they are processed separately]. For touch gestures, add a touch box of all zeros to obtain a touch box of a specific length. For lengths greater than... For touch gestures, downsample touch gesture X to obtain the resampled length. Touch gestures For length of

[0060] For a touch gesture X within a touch frame, the start and end times are first determined based on the first and last touch frames of X. Then, linear interpolation is performed between the start and end times to obtain... The sampling time for each touch frame. Then, for an interpolated sampling time... Search for the touch frame with the closest timestamp in X. t and x t+1 ,in Finally, we obtain the result based on x. t and x t+1 Resampled touch frames

[0061] Develop a 3D CNN-based architecture to extract the spatiotemporal features within touch gestures.

[0062] Data preprocessing consists of three steps: differential operation, threshold setting, and resampling. Differential operation is used to alleviate the problem of inaccurate baseline capacitance values ​​in capacitive touchscreens, threshold setting is used to remove noisy data, and resampling is used to process each touch gesture into a fixed and finite length.

[0063] Three-dimensional convolution operations were applied to the convolutional layers of the CNN model to capture both spatial and temporal information. This not only captured the appearance and outline of the fingertip in the touch frame, but also the dynamic sliding trajectory in consecutive touch frames.

[0064] The input to the 3D CNN model is a touch gesture. Since each element of the touch frame matrix represents a change in capacitance, the number of feature maps is set to 1, just like with grayscale images. The model's output is the probability of a touch gesture belonging to a legitimate user or an intruder.

[0065] In each convolutional layer, convolution operations are performed on the feature maps using 3D convolutional filters. Each feature map is connected to multiple touch boxes from the previous layer to form a spatiotemporal cube. By applying 3D convolutional filters to these cubes, the 3DCNN model not only captures local features of different spatial regions of the touch frame, but also captures dynamic sliding information between a series of touch frames.

[0066] i th layer j thThe (x,y,z) position in the feature map The calculation formula is:

[0067]

[0068] Where H i W i D i Indicate i th The height, width, and temporal depth of the 3D convolutional filter layer. M is the number of feature maps. Indicates connection to the previous layer k th The values ​​at the filter (h, w, d) positions of the feature map. Empirically, the same 3×3×3 convolutional filter is applied to the convolutional layers.

[0069] After each convolution operation, a transformation operation is performed on each feature map to add non-linear features. By performing convolutional filtering and non-linear transformations on the lower-level feature maps, the number of feature maps in subsequent layers is gradually increased to generate higher-level features.

[0070] For the pooling layers, 3D pooling operations are used to capture features at different spatial scales by downsampling the feature maps, enhancing invariance to touch frames and reducing model parameters. Based on experience, except for the 1×2×2 in the first pooling layer, a 2×2×2 3D pooling operation is performed on each feature map.

[0071] By alternately applying the above operations, the 3D CNN model converts the input touch gesture into a high-dimensional feature vector that captures the appearance and contour of the contact's fingertip, as well as dynamic swiping information of the fingertip movement encoded in consecutive touch frames.

[0072] An LSTM model consists of n identical networks R, which learn a nonlinear function. A series of touch gesture features X1…X n Transform into a hidden vector sequence h1…h n The hidden vectors store historical touch gesture information.

[0073] At each step i, the network will use the last hidden vector h i-1 and the characteristics X of the current touch gesture i As input, and output a new hidden vector h i .

[0074] Once the number of touch gestures reaches the predetermined number n, the recursive layer h... n The final output will be fed back to the output layer.

[0075] The output layer consists of a fully connected layer and a Softmax layer for generating authentication results. The fully connected layer maps the learned touch gesture representations to a two-dimensional output. The Softmax layer then uses a sigmoid function to normalize the output to values ​​between 0 and 1, representing the probability that the touch gesture sequence belongs to a legitimate user.

[0076] Training and Testing. Considering the practical application scenarios of continuous authentication, a sequential training and testing strategy is adopted to evaluate the model. By introducing the temporal relationship between the training and testing datasets, an authentication model based on historical touch data is established, and new touch gestures are further predicted. In this way, the evaluation process can more closely resemble real-world smartphone usage scenarios.

[0077] A 10,000-character touch gesture sequence was extracted from each subject and divided into K (K=10) non-overlapping subsequences. The first 7 subsequences were used as the training dataset to train the CNN model, the eighth subsequence was used as the validation dataset to verify the performance of different trained models, and the remaining subsequences were used as test data.

[0078] This evaluation process is repeated for each subject in turn. Given a user, he / she is designated as a legitimate user, and the remaining subjects are designated as intruders. The specific training and testing procedures are as follows:

[0079] Step 1: Using the touch gestures from the first six subsequences of legitimate users, construct positive training and test sets to train the CNN model, with the eighth subsequence used as the test set. Considering that model training depends on touch trajectories of fixed length, the subsequences are divided into touch trajectories of length n. The length of the user's touch trace is determined by the number of touch gestures n used for decision-making.

[0080] Step 2: Construct negative training and test sets to train the CNN model. We construct the negative training dataset by merging the first 6 subsequences of each remaining topic into a negative set, obtaining the same number of negative training touch trajectories as the positive training dataset. The eighth subsequence of the intruder is merged into the negative verification dataset. The same number of touch trajectories are maintained in both the negative and positive verification datasets through random sampling.

[0081] Step 3: Construct a positive test dataset using the last two sub-sequences of the legitimate subject's gesture to test the recognition capability of legitimate subjects.

[0082] Step 4: Extract test data of the same size from the remaining 99 subjects to construct a negative test dataset and test its ability to identify intruders.

[0083] The certification quality of all methods was evaluated using three metrics.

[0084] Metric 1: False Acceptance Rate (FAR). Assesses the probability that an intruder is verified as a legitimate user. High FAR indicates that the authentication system mistakenly allows unauthorized users to use the smartphone.

[0085] Metric 2: False Rejection Rate (FRR), which assesses the probability that a legitimate user is verified as an intruder. A higher FRR indicates that authorized users will be denied access to the smartphone.

[0086] Metric 3: Equal Error Rate (EER), which is the rate at which the FAR and FRR values ​​are equal. The lower the EER value, the higher the accuracy.

[0087] In another embodiment of the present invention, a mobile phone user touchscreen behavior authentication system is provided, which can be used to implement the above-mentioned mobile phone user touchscreen behavior authentication method. Specifically, the system includes:

[0088] The preprocessing module is used to preprocess the raw measurement data of the capacitive touch screen sensor.

[0089] The feature extraction module is used to extract the spatiotemporal features of touch gestures in the preprocessed data based on the 3D CNN architecture.

[0090] The modeling module is used to model the spatiotemporal features within touch gestures, employing a Long Short-Term Memory (LSTM) network to model the patterns between touch gestures.

[0091] The testing module is used for training and testing the model.

[0092] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0093] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a mobile phone user touchscreen behavior authentication method.

[0094] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the mobile phone user touchscreen behavior authentication method in the above embodiments.

[0095] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for authenticating the identity of mobile phone users based on touchscreen behavior, characterized in that, include: Perform data preprocessing on the raw measurement data of the capacitive touchscreen sensor; This method uses a 3D CNN architecture to extract the spatiotemporal features of touch gestures from preprocessed data. Based on the spatiotemporal features of touch gestures, a Long Short-Term Memory (LSTM) network is used to model the patterns between touch gestures. Train and test the model; Data preprocessing includes: difference operations, threshold setting, and resampling; Differential operation: Perform differential operation on each touch screen, specifically by reconstructing each touch frame by calculating the difference between the baseline measurement and the changed measurement of all cells in the grid; Threshold setting: Sets the threshold µ when the touch box is engaged. If cell x is greater than µ, keep the original measurement; otherwise, set the value to 0. Resampling: For a touch gesture with a length less than , add all-zero touch frames to obtain touch frames of a specific length; for a touch gesture with a length greater than , perform downsampling on the touch gesture X to obtain a resampled touch gesture with a length of ; for the touch gesture X with touch frames of length , first determine the start time and end time according to the first touch frame and the last touch frame of X; then, perform linear interpolation between the start time and the end time to obtain the sampling time of each touch frame in ; then, for an interpolated sampling time , search for the touch frame with the closest timestamp in X and , where t< <t+1; finally, obtain the resampled touch frame based on and , = +( ) x .​​ 2. The mobile phone user touchscreen behavior authentication method according to claim 1, characterized in that, Extracting the spatiotemporal features of touch gestures from the preprocessed data: Three-dimensional convolution operations are applied to the convolutional layers of the CNN model to capture the appearance and outline of the fingertip in the touch frame, as well as the dynamic sliding trajectory in the continuous touch frame; the input of the 3D CNN model is the touch gesture. The model outputs the probability of a touch gesture belonging to a legitimate user or an intruder; In each convolutional layer, 3D convolutional filtering is used to perform convolution operations on the feature maps. Each feature map is connected to multiple touch boxes in the previous layer to form a spatiotemporal cube. By applying 3D convolutional filters on these cubes, the 3D CNN model captures local features of different spatial regions of the touch frame and dynamic sliding information between touch frames. After each convolution operation, a transformation operation is performed on each feature map. By performing convolution filtering and nonlinear transformation on the low-level feature maps, the number of feature maps in subsequent layers is gradually increased to generate high-level features. For pooling layers, 3D pooling operations are used to capture features at different spatial scales by downsampling the feature map; By alternately applying the above operations, the 3D CNN model converts the input touch gesture into a high-dimensional feature vector that captures the appearance and contour of the contact's fingertip, as well as dynamic swiping information of the fingertip movement encoded in consecutive touch frames.

3. The mobile phone user touchscreen behavior authentication method according to claim 2, characterized in that, The value of the coordinates (x, y, z) in the j-th feature map of the i-th layer. The calculation formula is: in express The height, width, and temporal depth of the 3D convolutional filter layer; M is the number of feature maps; Indicates connection to the previous layer Feature map filters The value of the position.

4. The mobile phone user touchscreen behavior authentication method according to claim 1, characterized in that, An LSTM model consists of n identical networks R, which learn a nonlinear function. Touch gesture features … Transform into a hidden vector sequence … The hidden vectors store historical touch gesture information; At each step i (i = 1,…,n), the network will store the last hidden vector. and the characteristics of the current touch gesture As input, and output a new hidden vector. ; Once the number of touch gestures reaches the predetermined number n, the recursive layer... The final output will be fed back to the output layer; The output layer contains a fully connected layer and a Softmax layer for generating authentication results; the fully connected layer maps the learned touch gesture representations to a two-dimensional output; then, the Softmax layer uses the Sigmoid function to normalize the output to a value between 0 and 1, representing the probability that the touch gesture sequence belongs to a legitimate user.

5. The mobile phone user touchscreen behavior authentication method according to claim 1, characterized in that, Training and Testing: The model is evaluated using a sequential training and testing strategy. By introducing the temporal relationship between the training and testing datasets, an authentication model based on historical touch data is established, and new touch gestures are predicted. The specific training and testing process is as follows: Step 1: Using the touch gestures from the first six subsequences of legitimate users, construct positive training and test sets to train the CNN model, with the eighth subsequence used as the test set; divide the subsequences into touch trajectories of length n, where the length of the user's touch trace is determined by the number n of touch gestures used for decision-making; Step 2: Construct negative training and test sets to train the CNN model; The negative training dataset is constructed by merging the first 6 subsequences of each remaining topic into a negative set, and obtaining the same number of negative training touch trajectories as the positive training dataset from the negative set. The eighth subsequence of the intruder is merged into a negative verification dataset; By randomly sampling, the same number of touch trajectories are maintained in both the negative and positive verification datasets; Step 3: Construct a positive test dataset using the last two sub-sequences of the legitimate subject's gesture to test the recognition capability of legitimate subjects; Step 4: Extract test data of the same size from the remaining subjects to construct a negative test dataset and test its ability to identify intruders.

6. The mobile phone user touchscreen behavior authentication method according to claim 5, characterized in that, The certification quality of all methods was evaluated using three metrics; Metric 1: False Acceptance Rate (FAR): Assesses the probability that an intruder is verified as a legitimate user; High FAR indicates that the authentication system will mistakenly allow an unauthorized user to use the smartphone; Metric 2: False Rejection Rate (FRR): Assesses the probability that a legitimate user is verified as an intruder; a higher FRR indicates that authorized users will be denied access to the smartphone. Indicator 3: Equal Error Rate (EER): This is the rate at which the FAR and FRR values ​​are equal; the lower the EER value, the higher the accuracy.

7. A mobile phone user touchscreen behavior authentication system, characterized in that, include: The preprocessing module is used to preprocess the raw measurement data of the capacitive touch screen sensor. The feature extraction module is used to extract the spatiotemporal features of touch gestures in the preprocessed data based on the 3D CNN architecture. The modeling module is used to model the spatiotemporal features within touch gestures, employing a Long Short-Term Memory (LSTM) network to model the patterns between touch gestures. The testing module is used for training and testing the model; Data preprocessing includes: difference operations, threshold setting, and resampling; Differential operation: Perform differential operation on each touch screen, specifically by reconstructing each touch frame by calculating the difference between the baseline measurement and the changed measurement of all cells in the grid; Threshold setting: Sets the threshold µ when the touch box is engaged. If cell x is greater than µ, keep the original measurement; otherwise, set the value to 0. Resampling: For a length less than touch gesture, add all-zero touch frames to obtain a touch sequence of a specific length; for a length greater than touch gesture, perform downsampling on the touch gesture X to obtain a resampled length of touch gesture ; for a touch gesture X with a length of touch frames, first determine the start time and end time according to the first touch frame and the last touch frame of X; then, perform linear interpolation between the start time and the end time to obtain the sampling time of each touch frame in; then, for an interpolated sampling time , search for the touch frame with the closest timestamp in X and , where t< <t+1; finally, obtain the resampled touch frame based on and , = +( ) x .​ 8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the mobile phone user touch screen behavior authentication method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the mobile phone user touch screen behavior authentication method as described in any one of claims 1 to 6.

Images