Method, apparatus and electronic device for fingerprint liveness detection

Abstract

The application discloses a method and device for fingerprint living body detection and electronic equipment. The method comprises the following steps: acquiring direction field information corresponding to a to-be-detected fingerprint image, a fingerprint enhanced image and a minutia point list; dividing the to-be-detected fingerprint image into a ridge line area image and an edge area image according to the fingerprint enhanced image; performing a direction field guided center difference convolution operation on the ridge line area image according to the direction field information to obtain a first initial feature map; performing a random rotation operation on the minutia points corresponding to the edge area image in the minutia point list to obtain a second initial feature map; and determining feature vectors corresponding to the first initial feature map and the second initial feature map respectively by using a fingerprint living body detection network, and determining a detection result according to the feature vectors, wherein the fingerprint living body detection network is used for detecting whether the fingerprint image is a real fingerprint image. The application solves the technical problem that the generalization ability of the fingerprint recognition method in the related art is poor and the counterfeit fingerprint cannot be distinguished.

Description

Technical Field

[0001] This application relates to the field of fingerprint recognition, and more specifically, to a method, apparatus, and electronic device for fingerprint liveness detection. Background Technology

[0002] Fingerprint authentication has been widely used in security systems and applications due to its convenience and reliability. However, tools for creating fingerprints using synthetic materials have become a prominent means of carrying out presentation attacks (PAs). These deception techniques pose a significant threat to automated fingerprint authentication systems. Current attack techniques are capable of reconstructing details such as the pores of fingerprints, making it increasingly challenging to directly detect forged information in the original fingerprint image. Furthermore, existing fingerprint recognition methods suffer from poor generalization ability and are unable to distinguish between forged fingerprints.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, and electronic device for fingerprint liveness detection, which at least solves the technical problem that fingerprint recognition methods in the related art have poor generalization ability and cannot distinguish forged fingerprints.

[0005] According to one aspect of the embodiments of this application, a method for fingerprint liveness detection is provided, comprising: acquiring orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected, wherein the orientation field information is used to represent the orientation of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located, the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected determined based on the orientation field information, and the list of minutiae is used to represent an image composed of minutiae in the fingerprint image to be detected; dividing the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; performing orientation field-guided center difference convolution operation on the ridge region image based on the orientation field information to obtain a first initial feature map; performing random rotation operation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; using a fingerprint liveness detection network to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and determining the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0006] Optionally, based on the enhanced fingerprint image, the fingerprint image to be detected is divided into a ridge region image and an edge region image, including: performing a binarization operation on the enhanced fingerprint image to obtain an edge mask; determining the ridge region image based on the edge mask and the fingerprint image to be detected; and determining the edge region image based on the ridge region image and the fingerprint image to be detected.

[0007] Optionally, a first initial feature map is obtained by performing a direction field-guided center difference convolution operation on the ridge region image based on the direction field information, including: obtaining the convolution kernel and candidate mask corresponding to the direction field-guided center difference convolution operation on the ridge region image, and obtaining the center pixel and the pixels in the receptive field of the ridge region image, wherein the candidate mask is used to extract gradient information along the ridge direction in the ridge region image; obtaining a multi-channel feature map based on the convolution kernel, candidate mask, center pixel and pixels in the receptive field; and determining the feature map of the direction field information corresponding to each pixel in the ridge region image from the multi-channel feature map based on the direction field information, thereby obtaining the first initial feature map.

[0008] Optionally, a second initial feature map is obtained by randomly rotating the detail points corresponding to the edge region image in the detail point list. This includes: extracting the foreground region of the edge region image using an edge extraction algorithm; determining the side length of the sub-blocks corresponding to the detail points in the edge region image based on the side length of the longer side in the foreground region; determining the sub-block to which each detail point in the edge region image belongs based on the side length of the sub-blocks, thus obtaining a set of sub-blocks; and randomly rotating the sub-blocks in the set of sub-blocks to obtain the second initial feature map.

[0009] Optionally, a fingerprint liveness detection network is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map, respectively, and the detection result is determined based on the feature vectors. This includes: using a first encoder to transform the first initial feature map to obtain a first feature map; using a second encoder to transform the second initial feature map to obtain a second feature map; splicing the first feature map and the second feature map to obtain a spliced ​​feature map; using a decoder to convert the spliced ​​feature map into a feature vector, and outputting the detection result corresponding to the feature vector.

[0010] Optionally, the loss function corresponding to the fingerprint liveness detection network during training includes: a first loss function and a second loss function, wherein the first loss function is a binary classification cross-entropy loss, and the second loss function is used to adjust the feature vector output by the decoder of the fingerprint liveness detection network according to the fingerprint material.

[0011] Optionally, the second loss function is determined as follows: Anchor vectors are determined from the feature vectors output by the decoder of the fingerprint liveness detection network, where the anchor vector is the feature vector corresponding to the anchor sample, and the anchor sample is any class of sample composed of positive and negative samples; a first feature vector is determined for samples belonging to the same class as the anchor sample, and a second feature vector is determined for samples belonging to a different class than the anchor sample; the second loss function is determined based on the anchor vector, the first feature vector, the second feature vector, and a vector threshold.

[0012] According to another aspect of the embodiments of this application, a fingerprint liveness detection apparatus is also provided, comprising: an acquisition module, configured to acquire orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected, wherein the orientation field information is used to represent the fingerprint ridge direction of the region where each pixel in the fingerprint image to be detected is located, the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected determined based on the orientation field information, and the list of minutiae is used to represent an image composed of minutiae in the fingerprint image to be detected; a segmentation module, configured to segment the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; a convolution module, configured to perform orientation field-guided center difference convolution operation on the ridge region image based on the orientation field information to obtain a first initial feature map; a rotation module, configured to perform random rotation operation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; and a determination module, configured to use a fingerprint liveness detection network to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and determine the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0013] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory for storing program instructions; and a processor connected to the memory for executing program instructions to perform the following functions: acquiring orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected, wherein the orientation field information is used to represent the fingerprint ridge direction of the region where each pixel in the fingerprint image to be detected is located, the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected determined based on the orientation field information, and the list of minutiae is used to represent an image composed of minutiae in the fingerprint image to be detected; dividing the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; performing orientation field-guided center difference convolution operation on the ridge region image based on the orientation field information to obtain a first initial feature map; performing random rotation operation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; using a fingerprint liveness detection network to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and determining the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0014] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-described fingerprint liveness detection method by running the computer program.

[0015] In this embodiment, the process involves acquiring orientation field information, a fingerprint enhancement image, and a minutiae list corresponding to the fingerprint image to be detected. The orientation field information represents the direction of the fingerprint ridges in the region where each pixel in the fingerprint image is located. The fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, determined based on the orientation field information. The minutiae list represents an image composed of minutiae in the fingerprint image to be detected. Based on the fingerprint enhancement image, the fingerprint image to be detected is divided into ridge region images and edge region images. Orientation field-guided center difference convolution is performed on the ridge region images based on the orientation field information to obtain a first initial feature map. The minutiae list is then processed... The minutiae corresponding to the edge region image are randomly rotated to obtain a second initial feature map. A fingerprint liveness detection network is used to determine the feature vectors corresponding to the first and second initial feature maps, and the detection result is determined based on the feature vectors. The fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image. This achieves the purpose of processing the ridge region image and the edge region image of the fingerprint image to be detected to obtain the corresponding feature maps, thereby realizing the technical effect of identifying whether the fingerprint image to be detected is a real fingerprint image based on the feature map. This solves the technical problem of poor generalization ability and inability to distinguish forged fingerprints in related fingerprint recognition methods. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a hardware structure block diagram of a computer terminal for implementing a fingerprint liveness detection method according to an embodiment of this application;

[0018] Figure 2 This is a flowchart of a fingerprint liveness detection method according to an embodiment of this application;

[0019] Figure 3a This is a schematic diagram illustrating a fingerprint image preprocessing method according to an embodiment of this application;

[0020] Figure 3b This is a comparison image of a fake fingerprint and a real fingerprint created using a current attack technique according to an embodiment of this application;

[0021] Figure 3c This is a schematic diagram of a fingerprint image decomposition process according to an embodiment of this application;

[0022] Figure 3d This is a schematic diagram of a candidate mask according to an embodiment of this application;

[0023] Figure 3e This is a schematic diagram of a conventional convolution, CDC, and OB-CDC convolution operation according to an embodiment of this application;

[0024] Figure 3f This is a schematic diagram illustrating a random rotation operation on detail points according to an embodiment of this application;

[0025] Figure 3g This is a network structure diagram of a PADRD-Net according to an embodiment of this application;

[0026] Figure 4 This is a structural diagram of a fingerprint liveness detection device according to an embodiment of this application. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] The fingerprint liveness detection method provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal for implementing a method for fingerprint liveness detection is shown. Figure 1As shown, the computer terminal 10 may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission module 106 for communication functions connected via wired and / or wireless networks. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0030] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be implemented wholly or partially as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be wholly or partially integrated into any other element in the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0031] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the fingerprint liveness detection method in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby implementing the aforementioned fingerprint liveness detection method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0032] The transmission module 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission module 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0033] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0034] It should be noted here that, in some optional embodiments, the above... Figure 1 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0035] In the above operating environment, this application provides a method embodiment for fingerprint liveness detection. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] Figure 2 This is a flowchart of a fingerprint liveness detection method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0037] Step S202: Obtain the orientation field information, fingerprint enhancement image, and minutiae list corresponding to the fingerprint image to be detected. The orientation field information is used to represent the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located. The fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, determined based on the orientation field information. The minutiae list is used to represent an image composed of minutiae in the fingerprint image to be detected.

[0038] In step S202 above, the fingerprint image to be detected needs to be preprocessed to obtain the orientation field information, fingerprint enhancement image, and minutiae list corresponding to the fingerprint image to be detected. For example... Figure 3aAs shown, the above preprocessing operations can be performed using FingerNet, converting functions such as orientation field, segmentation, enhancement, and detail extraction into convolutional methods and integrating them into a unified planar network. Figure 3a The output of FingerNet is shown.

[0039] exist Figure 3a In the orientation field information, FingerNet uses the ASPP layer after basic feature extraction as a multi-scale feature extraction module, and then performs parallel orientation regression on each scale feature map before fusing them to obtain the final orientation field estimate. For each input pixel, FingerNet predicts N discrete angles. The probability of the i-th element p ori (i) indicates the position The probability of the ridge direction value. By choosing p ori The maximum value of (i) determines the final direction field estimate θ(x,y), as shown in the following formula:

[0040]

[0041] for Figure 3a For fingerprint enhancement images, FingerNet directly employs the Gabor enhancement method and proposes a selective convolution method to address the issue of Gabor filters sharing weights on image patches with the same w and θ, rather than sharing weights across the entire image. The specific calculation formula is as follows:

[0042]

[0043] In formula (2) above, E(x,y) represents the value of each pixel after enhancement processing at position (x,y), and I represents the original image. This represents the result of Gabor convolution of the original image and the group. Let M(x, y, i) represent the group phase obtained by extracting parameters from the group Gabor filtered image. M(x, y, i) represents the mask generated to select appropriate enhancement blocks from the group phase. The i-th value at pixel (x, y) in mask M (i.e., the value corresponding to the Gabor filter in the i-th direction) is defined as follows:

[0044]

[0045] for Figure 3a In terms of details, FingerNet uses multilayer perceptrons to predict the probability that each input pixel is a fingerprint region, with an output size of [missing information]. The segmentation score map is generated. The detail extraction part takes the enhanced image and the segmentation score map as input, performs three convolutional pooling blocks as feature extraction, and generates a detail score map, an X probability map, a Y probability map, and a detail distribution map. By filtering the detail score maps with appropriate thresholds, a detail list can be obtained. The precise location of detail points is obtained by adding an offset, where the parameter of the offset is the maximum probability in the X and Y probability maps.

[0046] Step S204: Based on the enhanced fingerprint image, the fingerprint image to be detected is divided into ridge region image and edge region image.

[0047] In step S204 of the above fingerprint liveness detection method, the fingerprint image to be detected is divided into a ridge region image and an edge region image based on the fingerprint enhancement image. Specifically, this includes the following steps: performing a binarization operation on the fingerprint enhancement image to obtain an edge mask; determining the ridge region image based on the edge mask and the fingerprint image to be detected; and determining the edge region image based on the ridge region image and the fingerprint image to be detected.

[0048] In this embodiment of the application, by detecting a large number of forged fingerprints, such as Figure 3b As shown, current attack techniques can reconstruct fingerprint details, such as sweat pores, making it increasingly challenging to directly detect forged evidence in the original fingerprint image. Furthermore, due to the physical properties of the deceptive materials and the limitations of current attack techniques' precision, anomalous spurs are frequently observed along the edges of the ridges. Therefore, since the factors contributing to the formation of artificial evidence differ in different regions of the fingerprint, this paper proposes a method to effectively extract subtle fingerprint forgery evidence from each region by decomposing the fingerprint image into different regions—namely, the aforementioned ridge region image and edge region image—and employing different processing techniques. Specifically, for the ridge region image, processing is performed using central difference convolution based on the orientation field; for the edge region image, a minutiae random rotation algorithm is used.

[0049] The ridge region contains the main area of ​​the fingerprint ridges, while the edge region contains the edge areas of the ridges and valleys; their combination forms the original fingerprint image. To extract the ridge region image and the edge region image, [the process is described in the original text]. Figure 3a Fingerprint enhancement images are obtained using FingerNet and converted into binary {0,1} edge masks EM(x,y). Ridge region images and edge region images can be obtained using the following formula:

[0050] I ridge (x, y) = I raw (x, y)·EM(x, y), (4)

[0051] I edge (x, y) = I raw (x, y) - I ridge(x, y), (5)

[0052] Among them, I raw (x, y) represents the original fingerprint image, i.e., the fingerprint image to be detected mentioned above. ridge (x, y) represents the ridge region image, I edge (x, y) represents the edge region image. The specific decomposition process is as follows: Figure 3c As shown, in Figure 3c First, the enhanced fingerprint image of the fingerprint image to be detected is binarized to obtain the edge mask, which is the above EM(x,y). Then, it is obtained through formula (4). Figure 3c The edge region image is obtained from the ridge region image using formula (5).

[0053] Step S206: Perform orientation field-guided center difference convolution on the ridge region image based on the orientation field information to obtain the first initial feature map.

[0054] In step S206 of the above fingerprint liveness detection method, a center difference convolution operation guided by the orientation field is performed on the ridge region image based on the orientation field information to obtain a first initial feature map. Specifically, this includes the following steps: obtaining the convolution kernel and candidate mask corresponding to the center difference convolution operation guided by the orientation field on the ridge region image, and obtaining the center pixel and the pixels in the receptive field of the ridge region image, wherein the candidate mask is used to extract gradient information along the ridge direction in the ridge region image; obtaining a multi-channel feature map based on the convolution kernel, candidate mask, center pixel and pixels in the receptive field; and determining the feature map of the orientation field information corresponding to each pixel in the ridge region image from the multi-channel feature map based on the orientation field information to obtain the first initial feature map.

[0055] In this embodiment, since it is difficult to detect obvious forgery evidence in the ridge region, the focus is shifted from the spatial domain to the gradient field to extract fine-grained features. Central Difference Convolution (CDC) is an operator widely used for extracting invariants and detailed features. It uses learnable filters to capture rich gradient information. Unlike conventional convolution, CDC introduces central gradient features to enhance representation and generalization capabilities. Its description is as follows:

[0056]

[0057]

[0058] Wherein, formula (6) represents the convolution operation formula corresponding to regular convolution, and formula (7) represents the convolution operation formula corresponding to CDC. x and y represent the input feature map and output feature map, respectively, and w is the learnable convolution kernel. When enumerating position p n In the local receptive field R, p0 represents the current position on the input and output feature maps.

[0059] In the Fingerprint Presentation Attack Detection (PAD) task, due to the influence of sweat secretion and local damage, live fingerprints and forged fingerprints exhibit inconsistent brightness change patterns along the ridge direction during contact with the capture device screen. Therefore, gradient information along the ridge direction is of greater interest. This application employs Orientation-Based Central Difference Convolution (OB-CDC) to obtain gradient information along the ridge direction. Specifically, since the width of a ridge is approximately 8 pixels at 500 dpi, the size of the convolution kernel is expanded from 3×3 to 7×7, and a series of 7×7 candidate masks related to the gradient of the orientation field are designed, such as... Figure 3d As shown, five weights are used in these candidate masks. To distinguish the importance of gradient information at different locations, higher weights are assigned to locations along the ridge direction, where i∈[0,4]. OB-CDC takes the feature map x (i.e., the ridge region image mentioned above) and the orientation field information O (which can also refer to the orientation field information corresponding to the ridge region image) as input, and calculates the output feature map y. Each component of y is determined by the following formula:

[0060]

[0061] Where O(p0) represents the gradient of the orientation field at position p0, and M(O(p0)) represents the candidate mask selected based on the gradient of the orientation field, when enumerating positions p n When the pixel (i.e., the pixel in the receptive field mentioned above) is in the local receptive field R, p0 represents the current position on the input feature map and the output feature map (i.e., the center pixel of the ridge region image mentioned above).

[0062] Figure 3eA comparison of conventional convolution, CDC, and OB-CDC is given. The OB-CDC algorithm used in this embodiment includes two steps: 1) calculating the complete output feature map using each candidate mask; 2) determining the final value at each position in the output feature map based on the gradient of the orientation field, thereby obtaining the first initial feature map. Therefore, the complexity of OB-CDC is related to the number of candidate masks. For each candidate mask, the complexity is O(HWhwC). i C o ), where H represents the height of the input feature map, W represents the width of the input feature map, h represents the height of the convolution kernel, w represents the width of the convolution kernel, and C i and C o These represent the number of input channels and the number of output channels, respectively. Therefore, the overall complexity is O(NHWhwC). i C o N represents the number of candidate masks.

[0063] Step S208: Perform random rotation on the detail points corresponding to the edge region images in the detail point list to obtain the second initial feature map.

[0064] In step S208 of the above fingerprint liveness detection method, the minutiae corresponding to the edge region image in the minutiae list are randomly rotated to obtain a second initial feature map. Specifically, this includes the following steps: extracting the foreground region of the edge region image using an edge extraction algorithm; determining the side length of the sub-blocks corresponding to the minutiae in the edge region image based on the side length of the longer side in the foreground region; determining the sub-block to which each minutiae in the edge region image belongs based on the side length of the sub-blocks to obtain a set of sub-blocks; and performing a random rotation operation on the sub-blocks in the set of sub-blocks to obtain the second initial feature map.

[0065] In the embodiments of this application, for Figure 3a The obtained list of detail points needs to be randomly rotated, such as... Figure 3fAs shown, the first image is the original image containing minutiae, and the following three images are images obtained after randomly rotating the sub-blocks corresponding to the minutiae at different frequencies (e.g., different iteration numbers). In this embodiment, since it is necessary to randomly rotate the minutiae corresponding to the edge region image, such as using the MinutiaPatches Random Rotation (MPRR) algorithm, the foreground region of the edge region image is first extracted using an edge extraction algorithm. 1 / 10 of the side length of the longer side in the foreground region is determined as the side length of the sub-block corresponding to the minutiae in the edge region image. Based on the side length of this sub-block, the sub-blocks to which each minutiae in the edge region image belongs are divided, resulting in a set of sub-blocks. Then, a random rotation operation is performed on each sub-block in the set of sub-blocks to obtain a second initial feature map.

[0066] It should be noted that the diversity of fingerprint textures is caused by the different types and locations of minutiae. During the MPRR operation, the rotation degree of each small block is set to 90°×i, i∈[0,3], to destroy the identity information represented by the fingerprint. Simultaneously, MPRR is performed on the minutiae in each iteration of the training process, thereby achieving data augmentation. Choosing a 90° step rotation avoids the loss of forged evidence due to interpolation in rotations of other degrees. Furthermore, the side length of the sub-block can be 1 / 10 of the side length of the longer side in the foreground region, or it can be determined according to other proportions. The 1 / 10 mentioned above is merely an example and does not represent a limitation.

[0067] Step S210: The fingerprint liveness detection network is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and the detection result is determined based on the feature vectors. The fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0068] In step S210 of the above fingerprint liveness detection method, a fingerprint liveness detection network is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map, and the detection result is determined based on the feature vectors. Specifically, the steps include: using a first encoder to transform the first initial feature map to obtain a first feature map; using a second encoder to transform the second initial feature map to obtain a second feature map; splicing the first feature map and the second feature map to obtain a spliced ​​feature map; using a decoder to convert the spliced ​​feature map into a feature vector, and outputting the detection result corresponding to the feature vector.

[0069] In the above-mentioned fingerprint liveness detection method, the loss function corresponding to the fingerprint liveness detection network during the training process includes: a first loss function and a second loss function. The first loss function is the cross-entropy loss for binary classification, and the second loss function is used to adjust the feature vector output by the decoder of the fingerprint liveness detection network according to the fingerprint material.

[0070] In the aforementioned fingerprint liveness detection method, the second loss function is determined as follows: Anchor vectors are determined from the feature vectors output by the decoder of the fingerprint liveness detection network. Anchor vectors are feature vectors corresponding to anchor samples, and anchor samples are any class of samples composed of positive and negative samples. First feature vectors are determined for samples belonging to the same class as the anchor samples, and second feature vectors are determined for samples belonging to different classes than the anchor samples. The second loss function is determined based on the anchor vectors, the first feature vector, the second feature vector, and a vector threshold. It should be noted that samples of different classes refer to samples from different fingerprint materials, while samples of the same class refer to samples from the same fingerprint material.

[0071] In step S210 above, this application uses a fingerprint liveness detection network to detect whether the fingerprint image to be detected is a real fingerprint image. This fingerprint liveness detection network can also be a Presentation-Attack-Detection-with-Region-Decomposition-Network (PADRD-Net), with a specific network architecture as follows: Figure 3g As shown, for ridge branches, OB-CDC layers are used instead of regular convolutions in the stem layer. For edge branches, edge regions with MPRR are used for training, and the original edge regions are used for inference. Since rotation operations truncate the ridges and destroy gradient field information, MPRR is only used on edge branches.

[0072] exist Figure 3g In the image, the ridge region is processed by the OB-CDC algorithm to obtain the first initial feature map. Then, it is processed by the Encoder of the ResNet-18 network to obtain the first feature map, which is... Figure 3g In Feature Map_1, the edge region image undergoes MPRR operation to obtain the second initial feature map. This second initial feature map is then processed by the ReaNet-18 network's encoder to obtain the second feature map, which is the final feature map. Figure 3gFeature Map_2 is used in this example. It should be noted that two ReaNet-18 networks are deployed for processing the ridge region and edge region images, respectively. The encoders in these two ReaNet-18 networks (i.e., the first encoder and the second encoder mentioned above) are used to process the first and second initial feature maps, respectively, resulting in the first feature map (Feature Map_1) and the second feature map (Feature Map_2). After concatenating the first and second feature maps, a concatenated feature map is obtained. This concatenated feature map is then input into the decoder to obtain the corresponding feature vector. After passing through a fully connected layer (FC layer), the 512-dimensional feature vector is transformed into a 2-dimensional vector. Based on the 2-dimensional vector, the classification of the fingerprint image to be detected—whether it is a real fingerprint image (Live) or a fake fingerprint image (Spoof)—is determined.

[0073] In the process of determining the spliced ​​feature map, a mid-way fusion structure is adopted to fuse the features of the two processed branches and connect the intermediate feature maps after the first two blocks of the ResNet-18 backbone. This fusion function ensures that the fused features contain high-level information while retaining spatial information, enabling the network to pay more attention to regions containing forged evidence.

[0074] In the training process of the fingerprint liveness detection network described above, two loss functions are required: the first loss function and the second loss function. The first loss function is the cross-entropy loss function, which can be expressed as L. cls The cross-entropy loss function is a commonly used loss function in image processing and will not be elaborated further here. The second loss function mentioned above is used to adjust the feature vector output by the decoder of the fingerprint liveness detection network based on the fingerprint material. Specifically, since the differences between forged fingerprints can be much larger than those between real fingerprints, an adaptive PA loss is used to further align the feature space, bringing all real fingerprints closer together and separating fake fingerprints of different attack types, thereby changing the feature distribution. For example, the PA adaptive loss defines the label of a sample as the material of the fingerprint, while the material of a real fingerprint can be considered as skin, and other non-skin materials are considered as labels for fake fingerprints, thus distinguishing fake fingerprints of different materials.

[0075] The PA adaptive loss is also known as the second loss mentioned above, and its corresponding second loss function L padp The formula is as follows:

[0076]

[0077] Among them, f af represents the feature vector corresponding to the anchor sample in the mini-batch, that is, the anchor vector in the feature vector output by the decoder of the fingerprint liveness detection network mentioned above. p f represents the first feature vector corresponding to the positive samples in the mini-batch. n Let α represent the second feature vector corresponding to the negative sample in the mini-batch, and let α represent the vector threshold between positive and negative sample pairs.

[0078] The final loss function of the above fingerprint liveness detection network during training is a weighted sum of the two losses, as shown in the following formula:

[0079]

[0080] Among them, L cls The cross-entropy loss, L, represents the binary classification of real or forged fingerprints. padp Let λ represent the PA adaptive loss, and λ be the balance coefficient.

[0081] By using steps S202 to S208 in the above-mentioned fingerprint liveness detection method, the purpose of processing the ridge region image and the edge region image of the fingerprint image to be detected to obtain the corresponding feature maps is achieved. This realizes the technical effect of identifying whether the fingerprint image to be detected is a real fingerprint image based on the feature map, thereby solving the technical problem of poor generalization ability and inability to distinguish forged fingerprints in related fingerprint recognition methods.

[0082] To verify the effectiveness of the method provided in this application, the following experiments were conducted. The LivDet2021 and LivDet2019 datasets were selected, in which the materials of the fake fingerprints in the training and test sets are completely different. Regarding the selection of metrics, the ISO / IEC SC37 standard was followed, using the true presentation error rate (BPCER@APCER = 1%) with a fixed attack presentation error rate and the true detection rate (TDR@FDR = 1%) with a fixed fake detection rate to evaluate the performance of the method. APCER describes the proportion of presentation attacks that are incorrectly classified as true presentations in a specific scenario, and BPCER represents the proportion of true presentations that are incorrectly classified as presentation attacks in a specific scenario. In a scenario where live samples are considered positive samples and spoofed samples are considered negative samples, the calculation formula is as follows:

[0083]

[0084]

[0085] The accuracy of the model is measured using ACER (Average Classification Error Rate), as shown in the following formula:

[0086]

[0087] In addition to the ACER metric mentioned above, the ACC metric is also used on the entire dataset to directly reflect the accuracy of the model's classification. Specifically, the smaller the ACER and BPCER@APCER = 1%, the better the performance of the PAD model; the higher the TDR@FDR = 1% and the higher the ACC, the better the performance of the PAD model.

[0088] In the experiments, the PA adaptive α was set to 0.35, and the weighted sum λ of the two losses was set to 0.2. During training, the Adam optimizer was used, and the learning rate was reduced from 5e-4 to 5e-5 using the CosineAnnealingLR scheduler. By comparing the proposed PADRD-Net with the FSB and CFD-PAD methods, the results are shown in Table 1. FSB utilizes local patch centering and fingerprint detail alignment to complete the PAD task, while CFD-PAD weights the importance of each channel in the feature map and suppresses "noisy" channels to minimize interference.

[0089] Table 1

[0090]

[0091] As can be seen from Table 1, the PADRD-Net provided in this application embodiment shows good results on all evaluation metrics of the entire dataset. Compared with LivDet2021, BPCER@APCER=1% decreased from 36.09% to 20.39%, and TDR@FDR=1% increased from 68.01% to 87.12%.

[0092] This application proposes a novel fingerprint PAD (Patent Object Defacement) fake evidence detection method from the perspective of region decomposition. Since fake evidence in different fingerprint regions is affected by various factors (such as manufacturing process precision and blurring caused by sweat secretion), exhibiting different patterns, the fingerprint image is decomposed into ridge regions and edge regions. Furthermore, because fake fingerprint evidence mainly exists in the gradient field and spatial domain, this application uses a center difference convolution (OB-CDC) layer based on the directional field to highlight gradient changes along the ridge direction. It also uses the MPRR (Multi-Level Reduction) operation to preserve fake evidence while destroying identity information, thereby preventing the model from favoring topic classification. Feature vectors are obtained by fusing the feature maps corresponding to the ridge and edge branches using PADRD-Net, and then used to classify real or fake fingerprints. Experiments on the LivDet dataset demonstrate the effectiveness and robustness of this method.

[0093] Figure 4 This is a structural diagram of a fingerprint liveness detection device according to an embodiment of this application, as shown below. Figure 4As shown, the device includes:

[0094] The acquisition module 40 is used to acquire the orientation field information, fingerprint enhancement image and minutiae list corresponding to the fingerprint image to be detected. The orientation field information is used to represent the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located. The fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, which is determined based on the orientation field information. The minutiae list is used to represent an image composed of minutiae in the fingerprint image to be detected.

[0095] The segmentation module 42 is used to segment the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image;

[0096] Convolution module 44 is used to perform orientation field-guided center difference convolution operation on the ridge region image based on orientation field information to obtain the first initial feature map;

[0097] Rotation module 46 is used to perform random rotation operations on the detail points corresponding to the edge region images in the detail point list to obtain a second initial feature map.

[0098] The determination module 48 is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively using the fingerprint liveness detection network, and to determine the detection result based on the feature vectors. The fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0099] By using the acquisition module 40, segmentation module 42, convolution module 44, rotation module 46, and determination module 48 in the aforementioned fingerprint liveness detection device, the purpose of processing the ridge region image and edge region image of the fingerprint image to be detected to obtain corresponding feature maps is achieved. This enables the technical effect of identifying whether the fingerprint image to be detected is a real fingerprint image based on the feature map, thereby solving the technical problem of poor generalization ability and inability to distinguish forged fingerprints in related fingerprint recognition methods.

[0100] In the segmentation module of the aforementioned fingerprint liveness detection device, the segmentation module is used to perform a binarization operation on the fingerprint enhancement image to obtain an edge mask; determine the ridge region image based on the edge mask and the fingerprint image to be detected; and determine the edge region image based on the ridge region image and the fingerprint image to be detected.

[0101] In the convolution module of the aforementioned fingerprint liveness detection device, the convolution module is used to acquire the convolution kernel and candidate mask corresponding to the orientation field-guided center difference convolution operation on the ridge region image, as well as to acquire the center pixel of the ridge region image and the pixels in the receptive field of the ridge region image. The candidate mask is used to extract gradient information along the ridge direction in the ridge region image. Based on the convolution kernel, candidate mask, center pixel, and pixels in the receptive field, a multi-channel feature map is obtained. Based on the orientation field information, the feature map of the orientation field information corresponding to each pixel in the ridge region image is determined from the multi-channel feature map to obtain the first initial feature map.

[0102] In the rotation module of the aforementioned fingerprint liveness detection device, the rotation module is used to extract the foreground region of the edge region image through an edge extraction algorithm; determine the side length of the sub-block corresponding to the detail point in the edge region image based on the side length value of the longer side in the foreground region; determine the sub-block to which each detail point in the edge region image belongs based on the side length of the sub-block, and obtain a set of sub-blocks; and perform a random rotation operation on the sub-blocks in the set of sub-blocks to obtain a second initial feature map.

[0103] In the determination module of the aforementioned fingerprint liveness detection device, the determination module is used to convert the first initial feature map using a first encoder to obtain a first feature map; to convert the second initial feature map using a second encoder to obtain a second feature map; to stitch the first feature map and the second feature map together to obtain a stitched feature map; to convert the stitched feature map into a feature vector according to the decoder, and to output the detection result corresponding to the feature vector.

[0104] In the determination module of the aforementioned fingerprint liveness detection device, the loss function corresponding to the fingerprint liveness detection network during training includes: a first loss function and a second loss function, wherein the first loss function is a binary classification cross-entropy loss, and the second loss function is used to adjust the feature vector output by the decoder of the fingerprint liveness detection network according to the fingerprint material.

[0105] In the determination module of the aforementioned fingerprint liveness detection device, the second loss function is determined as follows: Anchor vectors are determined from the feature vectors output by the decoder of the fingerprint liveness detection network, where the anchor vector is the feature vector corresponding to the anchor sample, and the anchor sample is any class of sample composed of positive and negative samples; a first feature vector is determined corresponding to a sample belonging to the same class as the anchor sample, and a second feature vector is determined corresponding to a sample belonging to a different class than the anchor sample; the second loss function is determined based on the anchor vector, the first feature vector, the second feature vector, and a vector threshold.

[0106] It should be noted that, Figure 4 The fingerprint liveness detection device shown is used to perform... Figure 2The fingerprint liveness detection method shown above also applies to the fingerprint liveness detection device, and will not be repeated here.

[0107] This application also provides an electronic device, including: a memory for storing program instructions; and a processor connected to the memory for executing program instructions to perform the following functions: acquiring orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected, wherein the orientation field information is used to represent the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located, the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected determined based on the orientation field information, and the list of minutiae is used to represent an image composed of minutiae in the fingerprint image to be detected; dividing the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; performing orientation field-guided center difference convolution on the ridge region image based on the orientation field information to obtain a first initial feature map; performing random rotation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; using a fingerprint liveness detection network to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and determining the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0108] It should be noted that the aforementioned electronic equipment is used to perform Figure 2 The fingerprint liveness detection method shown above also applies to this electronic device, and will not be repeated here.

[0109] This application embodiment also provides a non-volatile storage medium, which includes a stored computer program. The device containing the non-volatile storage medium executes the following fingerprint liveness detection method by running the computer program: acquiring orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected; wherein the orientation field information represents the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located; the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, determined based on the orientation field information; and the list of minutiae represents an image composed of minutiae in the fingerprint image to be detected; dividing the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; performing orientation field-guided center difference convolution on the ridge region image based on the orientation field information to obtain a first initial feature map; performing random rotation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; and using a fingerprint liveness detection network to determine the feature vectors corresponding to the first and second initial feature maps respectively, and determining the detection result based on the feature vectors. The fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

[0110] It should be noted that the aforementioned non-volatile storage media is used for execution. Figure 2 The fingerprint liveness detection method shown above also applies to this non-volatile storage medium, and will not be repeated here.

[0111] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0112] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0113] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0117] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for fingerprint liveness detection, characterized in that, include: The orientation field information, fingerprint enhancement image, and minutiae list corresponding to the fingerprint image to be detected are obtained. The orientation field information is used to represent the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located. The fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, determined based on the orientation field information. The minutiae list is used to represent an image composed of minutiae in the fingerprint image to be detected. Based on the enhanced fingerprint image, the fingerprint image to be detected is divided into a ridge region image and an edge region image; Based on the orientation field information, the ridge region image is subjected to orientation field-guided center difference convolution operation to obtain a first initial feature map; Randomly rotate the detail points corresponding to the edge region images in the detail point list to obtain a second initial feature map; A fingerprint liveness detection network is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map, respectively, and the detection result is determined based on the feature vectors. The fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

2. The method according to claim 1, characterized in that, Based on the enhanced fingerprint image, the fingerprint image to be detected is divided into a ridge region image and an edge region image, including: The enhanced fingerprint image is binarized to obtain the edge mask; Based on the edge mask and the fingerprint image to be detected, the ridge region image is determined; The edge region image is determined based on the ridge region image and the fingerprint image to be detected.

3. The method according to claim 1, characterized in that, Based on the orientation field information, a orientation field-guided center difference convolution operation is performed on the ridge region image to obtain a first initial feature map, including: Obtain the convolution kernel and candidate mask corresponding to the orientation field-guided center difference convolution operation on the ridge region image, and obtain the center pixel of the ridge region image and the pixel in the receptive field of the ridge region image, wherein the candidate mask is used to extract gradient information along the ridge direction in the ridge region image; A multi-channel feature map is obtained based on the convolution kernel, the candidate mask, the center pixel, and the pixels in the receptive field; Based on the orientation field information, a feature map of the orientation field information corresponding to each pixel in the ridge region image is determined from the multi-channel feature map to obtain the first initial feature map.

4. The method according to claim 1, characterized in that, Randomly rotate the minutiae corresponding to the edge region images in the minutiae list to obtain a second initial feature map, including: The foreground region of the edge region image is extracted using an edge extraction algorithm; Based on the side length value of the longer side in the foreground region, the side length of the sub-block corresponding to the detail point in the edge region image is determined; Based on the side length of the sub-block, determine the sub-block to which each detail point of the edge region image belongs, and obtain a set of sub-blocks; The sub-blocks in the sub-block set are randomly rotated to obtain the second initial feature map.

5. The method according to claim 1, characterized in that, A fingerprint liveness detection network is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map, respectively, and the detection result is determined based on the feature vectors, including: The first initial feature map is transformed using a first encoder to obtain a first feature map; The second initial feature map is transformed using a second encoder to obtain a second feature map; The first feature map and the second feature map are concatenated to obtain a concatenated feature map. The decoder converts the spliced ​​feature map into a feature vector and outputs the detection result corresponding to the feature vector.

6. The method according to claim 1, characterized in that, The loss function corresponding to the fingerprint liveness detection network during training includes a first loss function and a second loss function, wherein the first loss function is a binary classification cross-entropy loss, and the second loss function is used to adjust the feature vector output by the decoder of the fingerprint liveness detection network according to the fingerprint material.

7. The method according to claim 6, characterized in that, The second loss function is determined in the following way: Determine the anchor vector in the feature vector output by the decoder of the fingerprint liveness detection network, wherein the anchor vector is the feature vector corresponding to the anchor sample, and the anchor sample is any one of the samples composed of positive and negative samples; Determine the first feature vector corresponding to the sample belonging to the same class as the anchor sample, and determine the second feature vector corresponding to the sample belonging to a different class from the anchor sample; The second loss function is determined based on the anchor vector, the first feature vector, the second feature vector, and the vector threshold.

8. A device for fingerprint liveness detection, characterized in that, include: The acquisition module is used to acquire orientation field information, fingerprint enhancement image and minutiae list corresponding to the fingerprint image to be detected. The orientation field information is used to represent the direction of the fingerprint ridge in the region where each pixel in the fingerprint image to be detected is located. The fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected, determined based on the orientation field information. The minutiae list is used to represent an image composed of minutiae in the fingerprint image to be detected. The segmentation module is used to segment the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; The convolution module is used to perform orientation field-guided center difference convolution operation on the ridge region image based on the orientation field information to obtain a first initial feature map; The rotation module is used to perform random rotation operations on the detail points corresponding to the edge region images in the detail point list to obtain a second initial feature map. The determination module is used to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively using a fingerprint liveness detection network, and to determine the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

9. An electronic device, characterized in that, include: Memory, used to store program instructions; A processor, connected to the memory, is configured to execute program instructions to perform the following functions: acquiring orientation field information, a fingerprint enhancement image, and a list of minutiae corresponding to a fingerprint image to be detected, wherein the orientation field information represents the direction of the fingerprint ridges in the region where each pixel in the fingerprint image to be detected is located, the fingerprint enhancement image is a feature map corresponding to each pixel in the fingerprint image to be detected determined based on the orientation field information, and the list of minutiae represents an image composed of minutiae in the fingerprint image to be detected; dividing the fingerprint image to be detected into a ridge region image and an edge region image based on the fingerprint enhancement image; performing orientation field-guided center difference convolution on the ridge region image based on the orientation field information to obtain a first initial feature map; performing a random rotation operation on the minutiae corresponding to the edge region image in the list of minutiae to obtain a second initial feature map; using a fingerprint liveness detection network to determine the feature vectors corresponding to the first initial feature map and the second initial feature map respectively, and determining the detection result based on the feature vectors, wherein the fingerprint liveness detection network is used to detect whether the fingerprint image is a real fingerprint image.

10. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored computer program, wherein the device containing the non-volatile storage medium executes the fingerprint liveness detection method according to any one of claims 1 to 7 by running the computer program.

Images