A biometric fingerprint detection method based on dual fingerprint features

Abstract

This invention discloses a biometric fingerprint detection method based on dual fingerprint features. The method includes: (1) constructing a dataset of resting fingerprint image sequences and a dataset of resting time intervals of a legitimate user actively and continuously pressing two different fingers in a resting state; (2) extracting multiple features from the legitimate user's fingerprint images and normalizing them, and forming a fusion feature vector with the normalized time interval between the two continuous presses; (3) performing clustering calculations on the fusion feature vector using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in a resting state, so as to perform fingerprint detection based on the cluster boundaries; (4) if the user is identified as being subjected to a presentation attack, then preventing unlocking. This invention supplements the vulnerabilities of previous fingerprint recognition devices in the face of biometric fingerprint presentation attack identification attacks, providing stronger protection for biometric fingerprint recognition.

Description

Technical Field

[0001] This invention relates to the field of biometric fingerprint recognition technology, and in particular to a biometric fingerprint detection method based on dual fingerprint features. Background Technology

[0002] ISO / IEC 30107 defines a presentation attack as presenting anomalous biometric data to a biometric data acquisition subsystem with the aim of interfering with the system's judgment. This includes fake fingerprint attacks and coercion attacks. Currently, research on fingerprint security focuses on fake fingerprint attacks, where attackers attempt to gain authorization using counterfeit fingerprints. To address this type of attack, research teams have developed next-generation multimodal biometric acquisition technologies and corresponding instruments. Unlike fake fingerprint attacks, coercion attacks involve an attacker forcing an unwilling but legitimate victim to press their fingerprint onto the fingerprint sensor. In a coercion attack, the fingerprint sensor still contacts the legitimate person's biometric fingerprint, making this attack difficult to detect with both traditional fingerprint recognition methods and recent counterfeit fingerprint detection methods.

[0003] To address the difficulty in detecting coercive attacks, Chinese patent CN114550225A discloses a method for detecting forced fingerprint pressing. This method collects both the resting fingerprint of a legitimate user and the fingerprint of a user under forced pressure, extracts four different forced unlock features from the two types of fingerprint images, and uses these extracted features for security identification, returning a judgment result. This method provides a way to detect forced fingerprint pressing, achieving the detection of coercive attacks and is more secure than existing fingerprint recognition methods. However, this method relies solely on a single fingerprint image from a single press, resulting in a high false positive rate and a limited scope of attack detection, as it only targets coercive attacks.

[0004] To address the aforementioned practical problems and technical challenges, this paper proposes a biometric fingerprint detection method based on dual fingerprint features. When subjected to an attack, legitimate users cannot press their fingers twice consecutively and rapidly as they would in a normal resting state. Users will resist to some extent, resulting in distortion of the fingerprint image sequence from consecutive presses and a significant increase in the time interval between presses, leading to features that are significantly different from those from resting presses. Summary of the Invention

[0005] To address the aforementioned issues, this paper proposes a biometric fingerprint detection method based on dual fingerprint features, specifically targeting changes in user pressing characteristics under the presented attack scenarios. The fingerprint detection results of this method rely on the fingerprint image sequences of two different fingers pressed consecutively by the user and the interval between the consecutive fingerprint acquisition by the fingerprint acquisition module, thus improving the recognition accuracy. This fusion feature vector, together with the normalized time interval between the two consecutive presses, forms a fusion feature vector with a wider attack detection range, demonstrating practical significance and promising application prospects.

[0006] A biometric fingerprint detection method based on dual fingerprint features specifically includes the following steps:

[0007] Step (1) Construct a dataset of resting fingerprint image sequences of two different fingers that were actively and continuously pressed by a legitimate user A in a resting state to enter the fingerprint module;

[0008] Step (2) Construct a dataset of resting time intervals of legitimate user A actively and continuously pressing the two different fingers described in step (1) in a resting state, and normalize the dataset.

[0009] Step (3) Use the image feature extraction algorithm to extract multiple features from the fingerprint image sequence of legitimate user A, normalize them, and combine them with the normalized time interval between the two consecutive presses of legitimate user A to form a fusion feature vector in the form of [[image feature 1], [image feature 2], ..., interval time].

[0010] Step (4) The fused feature vector is clustered using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in the resting state, so as to perform fingerprint detection based on the cluster boundaries;

[0011] Step (5) The legitimate user A presses the fingerprints of the two different fingers described in step (1) on the fingerprint module in a resting state, a state using a fake fingerprint, and a state under duress attack, respectively, to perform identity recognition.

[0012] Further, in step (1), the legitimate user A, in a resting state, actively and continuously presses two different fingers onto the fingerprint module to record a fingerprint image sequence that can be used for fingerprint device recognition and unlocking, with no obvious pause or waiting between the two presses. The smallest data unit in the dataset is the fingerprint image sequence, which consists of two fingerprint images of different fingers with a sequential pressing order. The fingerprint image sequence is used for user identification. The image sequence data consisting of different pressing orders of two different fingers, i.e., pressing finger one first and then finger two, and pressing finger two first and then finger one, respectively constitute 1 / 2 of the resting fingerprint image sequence dataset.

[0013] Furthermore, the time interval data collected in step (2) is the time interval between the continuous pressing of two different fingers to collect fingerprint images when the legitimate user A actively performs biometric fingerprint recognition. Specifically, it is the time difference between the fingerprint acquisition module completing the fingerprint image acquisition of finger two and completing the fingerprint image acquisition of finger one.

[0014] Furthermore, the image feature extraction algorithms used in step (3) include HOG feature extraction algorithm, LBP feature extraction algorithm, SIFT and convolutional neural network, to extract fingerprint image features used to distinguish between fingerprints subjected to presentation attacks and those pressed at rest. The features formed together with the time interval between two consecutive presses by the user are as follows:

[0015] The fingerprint pressure intensity feature in step (3a) is represented by the average gray intensity of the fingerprint image, and its expression is as follows:

[0016]

[0017] Where ω represents the fingerprint image being divided into ω×ω blocks, (u,v) is a pixel in the fingerprint image, and g(u,v) represents the gray value of a single pixel;

[0018] The fingerprint pressing angle feature in step (3b) is specifically the angle between the user's pressing and the fingerprint acquisition device, expressed as the grayscale intensity variance, and its expression is as follows:

[0019]

[0020] The fingerprint pressing position feature in step (3c) is specifically the center of the user's fingerprint pressing, represented by the grayscale difference value, and its expression is as follows:

[0021] Δg(x,y)=g(x,y)-g(x+Δx,y+Δy)

[0022] The fingerprint texture features in step (3d) include contrast (con), entropy (ent), and gradient; the contrast expression is shown below:

[0023]

[0024] Where p(i) is the probability of moving point (x,y) across the entire image to obtain Δg;

[0025] The entropy expression is shown below:

[0026]

[0027] The gradient expression is shown below:

[0028] G(x,y)=|G x (x,y)|+|Gy (x,y)|

[0029] in r represents the pixel block width.

[0030] The time interval between the two consecutive presses in step (3e) is specifically the time difference between the second successful fingerprint image entry and the first successful fingerprint image entry for legitimate user A.

[0031] Further, in step (4), the fused feature vector is clustered using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in the resting state, so as to perform fingerprint detection based on the cluster boundaries.

[0032] Furthermore, in step (5), in order to complete identity recognition, user A needs to press the fingerprint of different fingers twice on the fingerprint module to authenticate the identity; the conventional fingerprint recognition classification and the fingerprint recognition under presentation attack are processed in parallel. While detecting whether the fingerprint image sequence is in the database, it is also detected whether the fingerprint is under presentation attack; if the fingerprint is in a resting state, the fingerprint recognition is performed normally; otherwise, the fingerprint authentication is rejected.

[0033] The beneficial effects of this invention: Compared with the prior art, the significant advantages of this invention are:

[0034] 1. This invention addresses the practical problems and technical challenges of existing technologies. Specifically targeting presentation attack scenarios, this invention utilizes fingerprint image sequences from two different fingers of a user and the time interval between consecutive presses to achieve a more secure biometric fingerprint detection method compared to existing technologies. Test results show that this method's fingerprint detection results, relying on the fingerprints of two different fingers and the interval between two consecutive presses, can more accurately detect whether a user's fingerprint recognition is under presentation attack, improving recognition accuracy and demonstrating practical significance and promising application prospects.

[0035] 2. This invention uses feature engineering, taking the user's fingerprint image and the time interval between continuous presses as features, and uses a clustering algorithm to distinguish between resting fingerprints and fingerprints subjected to presentation attacks. It is data-driven and relies solely on the user's own physiological characteristics and behavioral habits, thereby enhancing the security and accuracy of identification. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of a dataset construction process according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the training process of a detection model according to an embodiment of the present invention;

[0038] Figure 3This is a schematic flowchart of a biometric fingerprint detection method based on dual fingerprint features according to an embodiment of the present invention;

[0039] Figure 4 This invention presents a comparison of attack detection performance with previous technologies. Detailed Implementation

[0040] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, and the terms "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0041] The resting fingerprint image dataset and the resting pressure time interval dataset were jointly constructed by multiple legitimate users. The data entry process for each legitimate user is as follows: Figure 1 As shown, it includes the following steps:

[0042] Step (1) Constructs a dataset of resting fingerprint image sequences of two different fingers actively and continuously pressed by a legitimate user in a resting state to enter the fingerprint module. The legitimate user actively and continuously presses two different fingers in a resting state to enter the fingerprint image sequence dataset that can be used for fingerprint device recognition and unlocking. There is no obvious pause or waiting between the two presses; the entry process is smooth and natural. The two fingerprint images of different fingers entered sequentially together constitute the smallest data unit of the database, namely the user fingerprint image sequence, used for user identification. When a legitimate user actively performs fingerprint recognition, identity authentication can be achieved. The image sequence data composed of different pressing orders of the two different fingers—that is, pressing finger one first and then finger two, and pressing finger two first and then finger one—each constitute 1 / 2 of the resting fingerprint image sequence dataset.

[0043] Step (2) Construct a dataset of resting time intervals when a legitimate user actively and continuously presses two different fingers in step (1) in a resting state. The time interval data is the time difference when a legitimate user actively performs biometric fingerprint recognition and continuously presses two different fingers to collect fingerprint images. Specifically, it is the time difference between the fingerprint acquisition module completing the fingerprint image acquisition of finger two and completing the fingerprint image acquisition of finger one. The dataset is then normalized.

[0044] The fingerprint image sequence database stores fingerprint image sequence records of multiple legitimate users. During the database establishment process, each legitimate user actively and continuously presses the fingerprint module with two different fingers in a resting state to record fingerprint image sequence data that can be used for fingerprint device recognition and unlocking. There is no obvious pause or waiting between the two presses, and the recording process is smooth and natural. The two fingerprint images of different fingers recorded in succession together constitute the smallest data unit of the database, namely the user fingerprint image sequence, which is used for user identification. When a legitimate user actively performs fingerprint recognition, identity authentication can be achieved.

[0045] The training process diagram of the detection model in this embodiment is shown below. Figure 2 As shown, it includes the following steps:

[0046] Step (1) Construct the resting fingerprint image dataset and resting press time interval dataset of multiple legitimate users as described above;

[0047] Step (2) uses an image feature extraction algorithm to extract various features from the fingerprint image sequence of a legitimate user, which, together with the time interval between two consecutive presses by the legitimate user, form a feature vector; the expression for the average gray intensity of the fingerprint image is as follows:

[0048]

[0049] Where ω represents the fingerprint image being divided into ω×ω blocks, and g(u,v) represents the gray value of a single pixel. The expression for the variance of gray intensity is as follows:

[0050]

[0051] The expression for grayscale difference is as follows:

[0052] Δg(x,y)=g(x,y)-g(x+Δx,y+Δy)

[0053] Texture features include contrast (con), entropy (ent), and gradient; the contrast expression is shown below:

[0054]

[0055] Where p(i) is the probability of moving point (x,y) across the entire image to obtain Δg; and the entropy expression is as follows:

[0056]

[0057] The gradient expression is shown below:

[0058] G(x,y)=|G x (x,y)|+|G y (x,y)|

[0059] in r represents the pixel block width.

[0060] Step (3) Cluster the fused feature vector using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in the resting state, so as to perform fingerprint detection based on the cluster boundaries;

[0061] A schematic flowchart of the biometric fingerprint detection method based on dual fingerprint features according to an embodiment of the present invention is shown below. Figure 3 As shown:

[0062] Example 1

[0063] A legitimate user, A, performs fingerprint recognition while at rest. A presses two pre-registered fingerprints onto the fingerprint acquisition module twice in any order, with the entire pressing process being in a natural state. The "any order" means either pressing and registering finger one first, then pressing and registering finger two, or pressing and registering finger two first, then pressing and registering finger one.

[0064] After user A completes the first fingerprint image acquisition, the fingerprint acquisition module caches the first fingerprint image and records the successful acquisition time. When the legitimate user A completes the second fingerprint image acquisition within the specified time interval, the fingerprint acquisition module obtains the successful acquisition time of the second fingerprint image and calculates the time difference between the two successful fingerprint acquisitions, which is used to construct the fusion feature vector. The fingerprint acquisition module combines the second fingerprint image with the cached first fingerprint image to form a fingerprint image sequence, which is used to construct the fusion feature vector and complete identity recognition.

[0065] The attack-indicating features extracted from the fingerprint image sequence and the normalized time difference between two successful image acquisitions together constitute a fused feature vector. Based on the cluster boundaries formed by the clustering algorithm, it is determined whether User A is in a resting state. If User A is not in a resting state, authentication is directly rejected. If the determination result indicates that User A is in a resting state, further, identity recognition features are extracted from the acquired fingerprint image sequence and compared with multiple legitimate user identity identifiers existing in the database to return the identity recognition result.

[0066] Example 2

[0067] A legitimate user, A, performs fingerprint recognition while at rest. A presses one fingerprint that was already registered during registration onto the fingerprint sensor in any order, followed by another fingerprint that was not registered during registration. Alternatively, A presses one fingerprint that was not registered during registration onto the fingerprint sensor in any order, followed by another fingerprint that was not registered during registration. The entire pressing process is performed in a natural state.

[0068] After User A completes the first fingerprint image acquisition, the fingerprint acquisition module caches the first fingerprint image and records the successful acquisition time. When User A completes the second fingerprint image acquisition within a specified time interval, the fingerprint acquisition module obtains the successful acquisition time of the second fingerprint image and calculates the time difference between the two successful fingerprint acquisitions, which is used to construct the fusion feature vector. The fingerprint acquisition module combines the second fingerprint image with the cached first fingerprint image to form a fingerprint image sequence, which is used to construct the fusion feature vector and complete identity recognition.

[0069] The attack features extracted from the fingerprint image sequence and the time difference between the two successful image acquisitions, after normalization, together constitute a fused feature vector. Based on the cluster boundaries formed by the clustering algorithm, it can be determined that User A is in a resting state. However, furthermore, since the fingerprint sequence formed by User A's pressing is not in the database, the identity authentication fails.

[0070] Example 3

[0071] User A performs fingerprint recognition while at rest. User A uses a replica fingerprint, such as a fingerprint film, to press the fingerprint module, completing two fingerprint image acquisitions. The fingerprint image sequence and the time difference between successful fingerprint image acquisitions are then constructed. After User A completes the first fingerprint image acquisition using the replica fingerprint, the fingerprint acquisition module caches the first fingerprint image and records the first successful acquisition time. When User A completes the second fingerprint image acquisition within a specified time interval using the replica fingerprint, the fingerprint acquisition module obtains the successful acquisition time of the second fingerprint image and calculates the time difference between the two successful fingerprint acquisitions, which is used to construct a fusion feature vector. The fingerprint acquisition module combines the second fingerprint image with the cached first fingerprint image to form a fingerprint image sequence, which is used to construct the fusion feature vector and complete identity verification.

[0072] The attack features extracted from the fingerprint image sequence and the time difference between two successful image acquisitions after normalization together constitute the fused feature vector. Based on the cluster boundaries formed by the clustering algorithm, fingerprints exceeding the cluster boundaries due to differences in fingerprint film features and fine-grained differences in time intervals cannot pass authentication.

[0073] Example 4

[0074] User A is forced to press their fingerprint during fingerprint recognition by attacker B; that is, legitimate user A is subjected to a coercive attack while pressing their fingerprint. User A completes the first fingerprint image acquisition under duress. The fingerprint acquisition module caches the first fingerprint image and records the time of the first successful acquisition. The attacker then coerces user A to complete the second fingerprint image acquisition within a specified time interval. The fingerprint acquisition module obtains the time of the successful acquisition of the second fingerprint image and calculates the time difference between the two successful fingerprint acquisitions, which is used to construct a fused feature vector.

[0075] Due to the coercive attack, the legitimate user A pressed their fingerprint with greater force and speed. Because of the longer time interval between the resistance, the legitimate user could not press their finger twice in a continuous and rapid manner as in a normal resting state. The user would resist to a certain extent, and the fingerprint image sequence of continuous pressing would be deformed. The time interval between pressing would be significantly increased, resulting in features that are significantly different from the fused feature vector generated by resting pressing. These features would exceed the cluster boundary, indicating that the user was not in a resting state and would fail the identity authentication.

[0076] Figure 4 The graph shows a comparison of attack detection performance between the prior art (Southeast University. A method for detecting forced pressure on biometric fingerprints: CN202210170301.3[P]. 2022-05-27) and this embodiment.

[0077] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features.

Claims

1. A biometric fingerprint detection method based on dual fingerprint features, characterized in that: Specifically, the following steps are included: Step (1) Construct a dataset of resting fingerprint image sequences of two different fingers that were actively and continuously pressed by a legitimate user A in a resting state to enter the fingerprint module; Step (2) Construct a dataset of resting time intervals of two different fingers actively and continuously pressing in step (1) by legitimate user A in a resting state, and normalize the dataset; Step (3) Use an image feature extraction algorithm to extract multiple features of the fingerprint image sequence of legitimate user A, normalize them, and combine them with the normalized time interval of the two consecutive presses of legitimate user A to form a fused feature vector; The features composed of the two consecutive press time intervals of the user in step (3) are as follows: The fingerprint pressure intensity feature in step (3a) is represented by the average gray intensity of the fingerprint image, and its expression is as follows: Where ω represents the fingerprint image being divided into ω×ω blocks, (u,v) is a pixel in the fingerprint image, and g(u,v) represents the gray value of a single pixel; The fingerprint pressing angle feature in step (3b) is specifically the angle between the user's pressing and the fingerprint acquisition device, expressed as the grayscale intensity variance, and its expression is as follows: The fingerprint pressing position feature in step (3c) is specifically the center of the user's fingerprint pressing, represented by the grayscale difference value, and its expression is as follows: Δg(x,y)=g(x,y)-g(x+Δx,y+Δy) The fingerprint texture features in step (3d) include contrast (con), entropy (ent), and gradient; the contrast expression is shown below: Where p(i) is the probability of moving point (x,y) across the entire image to obtain Δg; The entropy expression is shown below: The gradient expression is shown below: G(x,y)=|G x (x,y)|+|G y (x,y)| in r represents the pixel block width; The time interval between the two consecutive presses in step (3e) is specifically the time difference between the second successful fingerprint image entry and the first successful fingerprint image entry by the legitimate user A. Step (4) Cluster the fused feature vectors using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in the resting state, so as to perform fingerprint detection based on the cluster boundaries; Step (5) Legitimate user A presses the fingerprints of the two different fingers described in step (1) on the fingerprint module in different pressing sequences to perform identity recognition.

2. The biometric fingerprint detection method based on dual fingerprint features according to claim 1, characterized in that: In step (1), the legitimate user A actively and continuously presses the fingerprint module with two different fingers in a resting state to record a fingerprint image sequence that can be used for fingerprint device recognition and unlocking. There is no obvious pause or waiting between the two presses. The smallest data unit in the dataset is the fingerprint image sequence, which consists of two fingerprint images of different fingers with a sequential pressing order. The fingerprint image sequence is used for user identification. The image sequence data consisting of different pressing orders of two different fingers, i.e., pressing finger one first and then pressing finger two, and pressing finger two first and then pressing finger one, respectively constitute 1 / 2 of the resting fingerprint image sequence dataset.

3. The biometric fingerprint detection method based on dual fingerprint features according to claim 1, characterized in that: The time interval data collected in step (2) is the time interval between the fingerprint image acquisition of two different fingers when the legitimate user A actively performs biometric fingerprint recognition. Specifically, it is the time difference between the fingerprint acquisition module completing the fingerprint image acquisition of finger two and the fingerprint image acquisition of finger one.

4. The biometric fingerprint detection method based on dual fingerprint features according to claim 1, characterized in that: The image feature extraction algorithms used in step (3) include gray-level co-occurrence matrix method, Gabor decomposition method, SIFT and convolutional neural network, to extract fingerprint image sequence features used to distinguish between those subjected to presentation attacks and those subjected to resting pressure. After normalization, these features are combined with the normalized time interval between two consecutive presses by the user to form a fusion feature vector in the form of [[image feature 1], [image feature 2], ..., interval time].

5. The biometric fingerprint detection method based on dual fingerprint features according to claim 1, characterized in that: In step (4), the fused feature vector is clustered using a clustering algorithm to determine the cluster boundaries of continuous fingerprint presses in the resting state, so as to perform fingerprint detection based on the cluster boundaries.

6. The biometric fingerprint detection method based on dual fingerprint features according to claim 1, characterized in that: In step (5), in order to complete the identity recognition, user A needs to press the fingerprint of different fingers twice on the fingerprint module to authenticate the identity; the conventional fingerprint recognition classification and the fingerprint recognition under presentation attack are processed in parallel. While checking whether the fingerprint image sequence is in the database, it is also checked whether the fingerprint is under presentation attack; if the fingerprint is in a resting state, the fingerprint recognition is performed normally; otherwise, the fingerprint authentication is rejected.

Images