Fingerprint Anti-counterfeiting method and apparatus

Abstract

Provided are a fingerprint anti-counterfeiting method and apparatus. The fingerprint anti-counterfeiting method comprises: acquiring a number n of fingerprint images collected by means of an ultrasonic fingerprint sensor, n≥2 and n being an integer (S00); and, by means of a neural network model, determining the authenticity of a fingerprint corresponding to the n fingerprint images (S10). The described solution can improve a fingerprint anti-counterfeiting effect.

Description

Fingerprint anti-counterfeiting method and device

[0001] The present application claims priority to the Chinese patent application with the application date of June 21, 2024, the application number of "202410817383.5", and the patent name of "Fingerprint Anti-counterfeiting Method and Device", the entire content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] Embodiments of the present application relate to the field of fingerprint anti-counterfeiting technology, in particular to a fingerprint anti-counterfeiting method and device. BACKGROUND

[0003] Fingerprint recognition is a commonly used biometric technology, which plays an important role in identity verification in many fields such as unlocking of smart phones, financial transactions, and access control. Common fingerprint recognition technologies include optical fingerprint recognition technology and ultrasonic fingerprint recognition technology. Among them, ultrasonic fingerprint recognition technology is increasingly becoming one of the mainstream fingerprint recognition technologies due to its fast unlocking, high accuracy, ease of use, and wide applicability. At present, fingerprint recognition technologies such as ultrasonic fingerprint recognition do not perform well in identifying 3D real and fake fingerprints. If a fake fingerprint is mistaken for a real fingerprint during fingerprint recognition, it may lead to security risks such as fraud attacks, privacy leaks, and property losses. Therefore, in order to reduce the possibility of such security risks, some technical solutions are needed to accurately determine the authenticity of fingerprints and improve the effectiveness of fingerprint anti-counterfeiting. SUMMARY

[0004] In view of the above, the embodiments of the present application provide a fingerprint anti-counterfeiting scheme to at least partially solve the above problems.

[0005] According to a first aspect of the embodiments of the present application, a fingerprint anti-counterfeiting method is provided, comprising: acquiring n frames of fingerprint images collected by an ultrasonic fingerprint sensor, wherein n≥2 and n is an integer; determining the authenticity of a fingerprint corresponding to the n frames of fingerprint images by a neural network model.

[0006] According to a second aspect of the embodiments of the present application, a fingerprint anti-counterfeiting device is provided, comprising: an acquisition module configured to acquire n frames of fingerprint images collected by an ultrasonic fingerprint sensor, wherein n≥2 and n is an integer; and a determination module configured to determine the authenticity of a fingerprint corresponding to the n frames of fingerprint images by a neural network model.

[0007] According to a third aspect of the embodiments of the present application, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus; the memory is configured to store a computer program; and the processor is configured to execute the method of the first aspect by running the computer program stored on the memory.

[0008] According to a fourth aspect of the embodiments of the present application, a computer storage medium is provided, and the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method of the first aspect.

[0009] According to a fifth aspect of the embodiments of the present application, a computer program product is provided, comprising a computer program, and the computer program is executed by a processor to implement the method of the first aspect.

[0010] According to the fingerprint anti-counterfeiting scheme provided by the embodiments of the present application, n frames of fingerprint images (i.e. ultrasonic fingerprint images) collected by an ultrasonic fingerprint sensor can be obtained, wherein n≥2 and n is an integer, and then a neural network model is used to determine the authenticity of the fingerprint corresponding to the n frames of fingerprint images. Therefore, the fingerprint anti-counterfeiting scheme can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images when performing fingerprint anti-counterfeiting. Compared with the recognition mode based on a single frame of fingerprint image, the neural network model can utilize more information increments beneficial to the fingerprint anti-counterfeiting task to perform fingerprint anti-counterfeiting, thereby effectively ensuring the accuracy. Moreover, since the neural network model is used to perform fingerprint anti-counterfeiting on n frames of fingerprint images, the accuracy can be further ensured, so that the fingerprint anti-counterfeiting scheme of the embodiments of the present application can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improve the fingerprint anti-counterfeiting effect, and improve the processing efficiency of fingerprint anti-counterfeiting. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the embodiments of the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0012] FIG. 1A is a step flowchart of a fingerprint anti-counterfeiting method according to an embodiment of the present application.

[0013] FIG. 1B is a step flowchart of some optional steps of step S10 according to an embodiment of the present application.

[0014] FIG. 2 is a step flowchart of some optional sub-steps of obtaining fused feature information in step S104 according to an optional implementation of an embodiment of the present application.

[0015] FIG. 3 is a schematic diagram of an overall implementation process of an optional implementation of the embodiment of the present application.

[0016] FIG. 4 is an optional flowchart of obtaining a first feature vector corresponding to each group of fingerprint images in the embodiment of the present application.

[0017] FIG. 5 is a schematic diagram of a first encoding network of some optional neural network models in the embodiment of the present application.

[0018] FIG. 6 is an optional sub-step flowchart of obtaining a first fusion feature vector in the embodiment of the present application.

[0019] FIG. 7 is some optional sub-step flowcharts of sub-step S1042B in the embodiment of the present application.

[0020] FIG. 8 is a schematic diagram of a second encoding network of some optional neural network models in the embodiment of the present application.

[0021] FIG. 9 is some optional sub-step flowcharts of obtaining fusion feature information in step S104 of the optional implementation two in the embodiment of the present application.

[0022] FIG. 10 is a schematic diagram of an overall implementation process of the optional implementation two in the embodiment of the present application.

[0023] FIG. 11 is some optional sub-step flowcharts of sub-step S202 in the embodiment of the present application.

[0024] FIG. 12 is a schematic diagram of a third encoding network of some optional neural network models in the embodiment of the present application.

[0025] FIG. 13 is a schematic diagram of a fourth encoding network of some optional neural network models in the embodiment of the present application.

[0026] FIG. 14 is some optional sub-step flowcharts of step S106 in the embodiment of the present application.

[0027] FIG. 15 is a schematic diagram of a fingerprint anti-counterfeiting device in the embodiment of the present application.

[0028] FIG. 16 is a structural schematic diagram of an electronic device in the embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the person skilled in the art better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the embodiments of the present application, all other embodiments obtained by the person skilled in the art should belong to the scope of protection of the embodiments of the present application.

[0030] The embodiments of the present application are further illustrated below with reference to the accompanying drawings of the embodiments of the present application.

[0031] FIG. 1A is a flowchart of steps of a fingerprint anti-counterfeiting method according to an embodiment of the present application. According to a first aspect in the embodiments of the present application, a fingerprint anti-counterfeiting method is provided, referring to FIG. 1A, the method includes steps S00 and S10, specifically:

[0032] S00: acquiring n frames of fingerprint images collected by an ultrasonic fingerprint sensor, where n≥2 and n is an integer.

[0033] S10: determining the authenticity of the fingerprint corresponding to the n frames of fingerprint images by a neural network model.

[0034] Based on this, the optional fingerprint anti-counterfeiting scheme of steps S00-S10 in the present application can acquire n frames of fingerprint images (i.e. ultrasonic fingerprint images) collected by an ultrasonic fingerprint sensor, where n≥2 and n is an integer, and then determine the authenticity of the fingerprint corresponding to the n frames of fingerprint images by a neural network model. Therefore, this fingerprint anti-counterfeiting scheme can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images when performing fingerprint anti-counterfeiting. Compared with the recognition method using a single frame of fingerprint image, the neural network model can utilize more information increments beneficial to the fingerprint anti-counterfeiting task to perform fingerprint anti-counterfeiting, thereby effectively ensuring the accuracy. Moreover, since the neural network model is used to perform fingerprint anti-counterfeiting on n frames of fingerprint images, the accuracy can be further ensured, so that the fingerprint anti-counterfeiting scheme of the embodiments of the present application can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improve the fingerprint anti-counterfeiting effect, and improve the processing efficiency of fingerprint anti-counterfeiting.

[0035] Some optional details and embodiments of the fingerprint anti-counterfeiting method of the present application are exemplarily described below.

[0036] It should be noted that the fingerprint anti-counterfeiting method in the embodiments of the present application can be applied to ultrasonic fingerprint anti-counterfeiting, and the above-mentioned fingerprint images can be ultrasonic fingerprint images.

[0037] The application examples of the present application can be exemplarily described in the application scenario of ultrasonic fingerprint anti-counterfeiting. The ultrasonic fingerprint image refers to the fingerprint image obtained by collecting the fingerprint through the ultrasonic fingerprint sensor. For example, when collecting the fingerprint through the ultrasonic fingerprint sensor, the fingerprint carrier (for example, the real finger of a user, a 3D fake finger (for example, a stereoscopic fingerprint mold), a 2.5D fake finger (for example, a planar fingerprint mold), etc.) can be placed above the fingerprint identification area of the ultrasonic fingerprint sensor for identification. The ultrasonic fingerprint sensor can include an acoustic layer, which can emit and receive ultrasonic signals (for example, the acoustic layer can be a piezoelectric material layer, which can emit and receive ultrasonic signals through the piezoelectric effect, and some example acoustic layers can be made of PVDF (polyvinylidene difluoride) material). When the acoustic layer emits ultrasonic signals outward, the fingerprint carrier above the fingerprint identification area can reflect at least part of the ultrasonic signals to form an ultrasonic echo signal carrying fingerprint information, and the acoustic layer can receive the ultrasonic echo signal reflected by the fingerprint carrier. The ultrasonic fingerprint sensor can generate an ultrasonic fingerprint image according to the ultrasonic echo signal, or a processing unit (which can include any processor such as a CPU, an MCU, etc.) electrically connected to the ultrasonic fingerprint sensor can generate an ultrasonic fingerprint image according to the ultrasonic echo signal. It should be understood that different ultrasonic echo signals can be received at different times of fingerprint collection, and multiple frames of ultrasonic fingerprint images can be generated. In the present application example, n frames of ultrasonic fingerprint images can be obtained from the multiple frames of ultrasonic fingerprint images for the data processing process of fingerprint anti-counterfeiting.

[0038] It should be understood that, since the fingerprint image used in the fingerprint anti-counterfeiting scheme of the present application example is an ultrasonic fingerprint image, the fingerprint anti-counterfeiting scheme of the present application example can be effectively applied to ultrasonic fingerprint anti-counterfeiting to effectively improve the accuracy of ultrasonic fingerprint anti-counterfeiting, more accurately judge the authenticity of the ultrasonic fingerprint, and improve the effect of ultrasonic fingerprint anti-counterfeiting.

[0039] It should be noted that in the present application example, the collection time interval of adjacent 2 frames of fingerprint images in step S00 can be different. For example, taking n as 3, that is, taking 3 frames of fingerprint images as an example, the first frame of fingerprint image is the fingerprint image collected at 5 ms, the second frame of fingerprint image can be the fingerprint image collected at 10 ms, and the third frame of fingerprint image can be the fingerprint image collected at 20 ms. The collection time interval of the first frame and the second frame of fingerprint images is 10 ms-5 ms=5 ms, and the collection time interval of the second frame and the third frame of fingerprint images is 20 ms-10 ms=10 ms. It can be seen that the collection time interval of adjacent two frames can be different. Corresponding to the case that the n frames of fingerprint images are n frames of ultrasonic fingerprint images, that is, if F iultrasound fingerprint images (0≤i≤n, i∈Z) representing different flight times, the following equation "flytime(F i+2 )-flytime(F i+1 )=flytime(F i+1 )-flytime(F i )" does not have to be satisfied, where flytime(F i ) refers to the flight time of the ultrasound echo signal corresponding to the ultrasound fingerprint image F i , and the time corresponding to the flight time can be taken as the time at which the ultrasound fingerprint image is collected. It can be seen that the time interval between adjacent two frames can be different.

[0040] Of course, in the n frames of fingerprint images used in step S00 in the embodiments of the present application, the time interval between adjacent two frames can also be the same, and the embodiments of the present application do not particularly limit this.

[0041] It should be understood that the n frames of fingerprint images used in the embodiments of the present application can be a plurality of frames of fingerprint images collected continuously (for example, four frames of fingerprint images 1, 2, 3 and 4 are collected continuously, and if three frames of fingerprint images collected continuously are used, they can be fingerprint images 1, 2 and 3 or fingerprint images 2, 3 and 4, which should be understood to be only a convenient example for understanding), or a plurality of frames of fingerprint images collected not completely continuously (for example, four frames of fingerprint images 1, 2, 3 and 4 are collected continuously, and if three frames of fingerprint images collected not completely continuously are used, they can be fingerprint images 1, 2 and 4 or fingerprint images 1, 3 and 4, which should be understood to be only a convenient example for understanding), which can be selected as needed, and the embodiments of the present application do not limit this.

[0042] In some optional embodiments, the n frames of fingerprint images used in the embodiments of the present application can be n frames of fingerprint images collected after a single ultrasound signal is emitted by an ultrasound fingerprint sensor.

[0043] Different from other types of fingerprint sensors, since the ultrasonic fingerprint sensor reflects the ultrasonic signal after emitting one ultrasonic signal (i.e., single ultrasonic signal), the ultrasonic signal can be reflected by different medium layers, for example, the real finger can include epidermis layer, dermis layer and other medium layers, the ultrasonic signal can be reflected by the epidermis layer of the real finger to form an ultrasonic echo signal, and the ultrasonic signal can also be reflected by the dermis layer of the real finger to form an ultrasonic echo signal, and the ultrasonic echo signals reflected by the two times can be collected by the ultrasonic fingerprint sensor to form multiple frames of fingerprint images. Therefore, in the embodiments of the present application, the n frames of fingerprint images are n frames of fingerprint images collected by the ultrasonic fingerprint sensor after emitting a single ultrasonic signal, the n frames of fingerprint images can effectively carry the information reflected by different medium layers, the frame difference and change information between the n frames of fingerprint images can be better utilized in the present application, and more information increments beneficial to the fingerprint anti-counterfeiting task can be used for fingerprint anti-counterfeiting to improve the fingerprint anti-counterfeiting effect. For example, in the optional embodiments below, the fingerprint feature information of such n frames of fingerprint images can be fused by a neural network model, the frame difference and change information between the n frames of fingerprint images can be more effectively utilized, compared with the recognition mode of a single frame of fingerprint image, the fusion feature information for determining the authenticity of the fingerprint has more information increments beneficial to the fingerprint anti-counterfeiting task, and thus the accuracy of fingerprint anti-counterfeiting can be effectively improved, and therefore the fingerprint anti-counterfeiting scheme of the present application can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor and improve the fingerprint anti-counterfeiting effect. Moreover, since the n frames of fingerprint images can be collected by the ultrasonic fingerprint sensor after emitting a single ultrasonic signal in the above optional embodiments, the time-consuming can be effectively reduced, and the efficiency of fingerprint anti-counterfeiting can be effectively improved.

[0044] In other optional embodiments, the n frames of fingerprint images used in the embodiments of the present application can be n frames of fingerprint images collected by the ultrasonic fingerprint sensor after emitting multiple ultrasonic signals.

[0045] In the scenario where the time consumption is acceptable, the n frames of fingerprint images used in the embodiments of the present application are n frames of fingerprint images collected by the ultrasonic fingerprint sensor after emitting multiple ultrasonic signals, which is beneficial to making the n frames of fingerprint images carry more information, and thus the frame difference and change information between the fingerprint images corresponding to the multiple ultrasonic signals can be better utilized, more information increments beneficial to the fingerprint anti-counterfeiting task can be used for fingerprint anti-counterfeiting, which is beneficial to improving the fingerprint anti-counterfeiting effect and improving the security.

[0046] In the embodiments of the present application, the specific number of the multiple ultrasonic signals can be selected according to the requirements, which is not limited herein. For example, 2 times, 3 times, 5 times, 10 times, etc.

[0047] Optionally, at least two of the multiple ultrasonic signals are ultrasonic signals of different frequencies. Since ultrasonic signals of different frequencies have different penetrability, the multiple ultrasonic signals including at least two ultrasonic signals of different frequencies are used to acquire n frames of fingerprint images, the interframe difference and change information between the n frames of fingerprint images are more significant and more convenient to use, and more information increments beneficial to the fingerprint anti-counterfeiting task can be obtained, so that the use of the n frames of fingerprint images for fingerprint anti-counterfeiting is more conducive to improving the fingerprint anti-counterfeiting effect and better improving the security.

[0048] Optionally, the multiple ultrasonic signals can be ultrasonic signals of different frequencies in pairs, so as to achieve a better fingerprint anti-counterfeiting effect.

[0049] It should be noted that the neural network model in the embodiments of the present application can be a pre-trained neural network model, which can be implemented based on any suitable neural network model framework, and the embodiments of the present application do not make special limitations.

[0050] In some optional embodiments, referring to the flowchart in FIG. 1B, the step S10 can be implemented by the following steps S104 and S106, specifically:

[0051] S104: performing fingerprint feature fusion processing on the n frames of fingerprint images by the neural network model to obtain fusion feature information.

[0052] Since the n frames of fingerprint images exist fingerprints, fingerprint feature extraction can be performed on the n frames of fingerprint images to extract corresponding feature information. Then the extracted feature information can be fused to obtain fusion feature information. It should be understood that the pre-trained neural network model used in step S104 of the embodiments of the present application to perform fingerprint feature fusion processing on the n frames of fingerprint images can effectively improve the accuracy of the obtained fusion feature information and improve the processing efficiency of fingerprint anti-counterfeiting.

[0053] Since the n frames of fingerprint images are fingerprint images collected at different times, there is certain change information between different frames of fingerprint images, and for different fingerprint carriers, the change information is generally different, which can be caused by the deformation difference of different fingerprint carriers (such as real fingers of users, 3D fake fingers, 2.5D fake fingers, etc.) at different times during fingerprint collection. Therefore, compared with the recognition mode of single frame of fingerprint image, the fusion feature information obtained by the neural network model to fuse the fingerprint feature information of multiple frames of fingerprint images in the embodiments of the present application for determining the authenticity of the fingerprint has more information increments beneficial to the fingerprint anti-counterfeiting task, and thus in the case of using the above change information and these beneficial information increments, the accuracy of fingerprint anti-counterfeiting can be effectively improved, and the effect of fingerprint anti-counterfeiting can be improved.

[0054] In some optional embodiments, step S104 comprises: performing fingerprint feature extraction on the n frames of fingerprint images by the neural network model, and performing fusion processing on the extracted feature information to obtain the fusion feature information. Based on this, the n frames of fingerprint images can be effectively subjected to fingerprint feature fusion processing, and accurate fusion feature information can be obtained, so as to improve the fingerprint anti-counterfeiting effect.

[0055] In the embodiments of the present application, the step S104 of "performing fingerprint feature extraction on the n frames of fingerprint images, and performing fusion processing on the extracted feature information to obtain the fusion feature information" can be implemented through various optional implementation manners, which will be described in detail below.

[0056] <Optional implementation manner one>

[0057] In optional implementation manner one, the n frames of fingerprint images need to be grouped for processing. Optionally, the fingerprint anti-counterfeiting method of the embodiments of the present application further comprises, before performing fingerprint feature extraction on the n frames of fingerprint images by the neural network model:

[0058] Step S103: dividing the n frames of fingerprint images into g groups of fingerprint images, and each group of fingerprint images includes a frames of fingerprint images, wherein 2≤g≤n, n / g≤a<n, and g and a are integers.

[0059] It should be understood that the above n, g and a can be set as needed. Optionally, a≥2, that is, each group of fingerprint images includes at least 2 frames of fingerprint images, so that the inter-frame differences and change information of multiple frames of fingerprint images can be better utilized, and the fingerprint anti-counterfeiting effect can be improved. For example, in an example, n=3, g=2, and a=2. For the convenience of describing the embodiments of the present application, the following can be described with this example.

[0060] It should be understood that in other optional embodiments, a can also be 1, that is, each group of fingerprint images includes 1 frame of fingerprint image, when meeting the needs of fingerprint anti-counterfeiting.

[0061] In some optional embodiments, in the embodiments of the present application, when the n frames of fingerprint images are divided into g groups of fingerprint images, the a frames of fingerprint images in each group of fingerprint images are a frames of continuously collected fingerprint images, and there is an intersection between fingerprint images of at least two groups of fingerprint images in the g groups of fingerprint images.

[0062] Based on this, on the one hand, since there is an intersection between the fingerprint images of at least two groups of fingerprint images in the g groups of fingerprint images, when the g groups of fingerprint images are obtained, it is not necessary to collect more frames of fingerprint images, thereby reducing the fingerprint image collection time of the ultrasonic fingerprint sensor, and further improving the efficiency of fingerprint anti-counterfeiting; on the other hand, when the number of n frames of fingerprint images is fixed, since there is an intersection between the fingerprint images of at least two groups of fingerprint images in the g groups of fingerprint images, the number of groups of g groups of fingerprint images can be more, so that the amount of calculation information is larger when the g groups of fingerprint images are subsequently processed, the inter-frame differences and change information between the fingerprint images can be better utilized, and the effect of fingerprint anti-counterfeiting is improved; on the other hand, by this way, the n frames of continuously collected fingerprint images are divided into g groups of fingerprint images, which can also better model the relationship between the inter-frame change information of the fingerprint images and the authenticity, effectively adapt to the demand of fingerprint anti-counterfeiting, and further improve the effect of fingerprint anti-counterfeiting.

[0063] For example, taking the aforementioned example in which n=3, g=2, and a=2 as an example. The n=3 frames of fingerprint images are denoted as F0, F1, and F2, respectively, and the 3 frames of fingerprint images F0, F1, and F2 are 3 frames of continuously collected fingerprint images (for ultrasonic fingerprint images, the 3 frames of fingerprint images F0, F1, and F2 are collected with adjacent time of flight). After the n=3 frames of fingerprint images are divided, the g=2 groups of fingerprint images (for example, denoted as g0 group and g1 group, respectively) can be: the a=2 frames of the first group, i.e., the g0 group, are fingerprint images F0 and F1, and the a=2 frames of the second group, i.e., the g1 group, are fingerprint images F1 and F2 (for example, can be understood in combination with FIG. 3). As can be seen, there is an intersection between the first group, i.e., the g0 group, and the second group, i.e., the g1 group, which is the fingerprint image F1, and the a=2 frames of fingerprint images in the g0 group and the g1 group are both 2 frames of continuously collected fingerprint images. It should be understood that other cases can be similarly deduced, and will not be described herein again.

[0064] In some other optional embodiments, in the application, when the n frames of fingerprint images are divided into g groups of fingerprint images, the a frames of fingerprint images in each group of fingerprint images are all a frames of continuously collected fingerprint images, and there is no intersection between the fingerprint images of any two groups of fingerprint images in the g groups of fingerprint images.

[0065] Based on this, on the one hand, since there is no intersection between the fingerprint images of any two groups of fingerprint images in the g groups of fingerprint images, when the number of n frames of fingerprint images is fixed, a smaller number of g groups of fingerprint images can be formed, thereby the calculation amount can be reduced and the processing efficiency can be improved while ensuring the effect of fingerprint anti-counterfeiting; on the other hand, by this way, the n frames of continuously collected fingerprint images are divided into g groups of fingerprint images, which can also better model the relationship between the inter-frame change information of the fingerprint images and the authenticity, effectively adapt to the demand of fingerprint anti-counterfeiting, and further improve the effect of fingerprint anti-counterfeiting.

[0066] For example, taking another example where n = 4, g = 2, a = 2. Denote the n = 4 fingerprint images as F0, F1, F2, F3 respectively, then the 4 fingerprint images F0, F1, F2, F3 are 4 fingerprint images continuously captured (for the case of ultrasonic fingerprint images, the 4 fingerprint images F0, F1, F2, F3 are captured with adjacent time of flight). Then after dividing the n = 4 fingerprint images, the g = 2 groups of fingerprint images (for example, denoted as g0 group and g1 group respectively) can be: the a = 2 fingerprint images of the first group g0 group are F0, F1, and the a = 2 fingerprint images of the second group g1 group are F2, F3. It can be seen that there is no intersection between the first group g0 group and the second group g1 group, and the a = 2 fingerprint images in the g0 group and the g1 group are all 2 fingerprint images continuously captured. It should be understood that other cases can be similarly deduced, which will not be described here.

[0067] On the basis of the above step S103, referring to the flow chart shown in FIG. 2, the step S104 of "performing fingerprint feature extraction on the n fingerprint images, and performing fusion processing on the extracted feature information to obtain fused feature information" includes the following sub-steps S1041 and S1042, specifically:

[0068] S1041: respectively performing first fingerprint feature extraction processing on the g groups of fingerprint images to obtain g first feature vectors.

[0069] The embodiment of the present application extracts g first feature vectors which can respectively reflect the fingerprint features of the g groups of fingerprint images by respectively performing first fingerprint feature extraction processing on the g groups of fingerprint images.

[0070] It should be understood that for the case of a ≥ 2 (i.e. the case where each group of fingerprint images in the g groups of fingerprint images includes at least 2 fingerprint images), the inter-frame differences and change information of the multiple fingerprint images in the same group can be analyzed during the process of obtaining the g first feature vectors by performing first fingerprint feature extraction processing on the g groups of fingerprint images, and therefore the inter-frame differences and change information of the multiple fingerprint images in different groups can be further analyzed by subsequently performing feature fusion processing on the g first feature vectors to obtain the first fused feature vector (i.e. the fused feature information), which further realizes the fusion processing of the feature information, can obtain better fused feature information, thereby enabling the inter-frame differences and change information between the n fingerprint images to be more effectively utilized, and enabling the fused feature information to have more beneficial information increment for the fingerprint anti-counterfeiting task, and further enabling the accuracy of fingerprint anti-counterfeiting to be more effectively improved, enabling the technical solution of the present application to more accurately judge the authenticity of the fingerprint captured by the ultrasonic fingerprint sensor, and improving the effect of fingerprint anti-counterfeiting.

[0071] Optionally, the fingerprint feature extraction and fusion in step S104 in any of the embodiments can be implemented by using a pre-trained neural network model. Optionally, referring to FIG. 3, the neural network model used in optional implementation scheme one can include a first encoding network N1 and a second encoding network N2. The structure and function of the first encoding network N1 and the second encoding network N2 can complete the fingerprint feature extraction and fusion. This part will be described in combination with the following.

[0072] It should be understood that, from the following embodiments, it can also be known that, for the case of a≥2, the inter-frame difference and change information of the multiple frames of fingerprint images in the same group can be analyzed during the processing of the first encoding network N1, and the inter-frame difference and change information of the multiple frames of fingerprint images in different groups can be further analyzed during the processing of the second encoding network N2, and better fusion feature information can be obtained, so that the inter-frame difference and change information between the n frames of fingerprint images can be more effectively utilized, and the fusion feature information has more beneficial information increment for the fingerprint anti-counterfeiting task, thereby the accuracy of fingerprint anti-counterfeiting can be more effectively improved, and the technical solution of the present application can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, and improve the effect of fingerprint anti-counterfeiting.

[0073] In some optional embodiments, referring to the flowchart shown in FIG. 4, the first feature vector of each group of fingerprint images in the g groups of fingerprint images can be calculated by the following sub-steps S1041A-S1041E in sub-step S1041 to obtain g groups of first feature vectors. Specifically, for each group of fingerprint images in the g groups of fingerprint images:

[0074] S1041A: The group of fingerprint images is subjected to convolution processing to obtain a first processing result.

[0075] Optionally, referring to FIG. 5, which shows a schematic diagram of the first encoding network of some optional neural network models in the embodiments of the present application. As shown in FIG. 5, the first encoding network N1 can include a convolution layer (which can be referred to as a Conv layer for short), and the input of the convolution layer is the input of the first encoding network N1. When the first group of g0 groups of fingerprint images or the second group of g1 groups of fingerprint images is input, the group of fingerprint images can be input into the convolution layer, and the group of fingerprint images is subjected to convolution processing by the convolution layer, and the result obtained is the first processing result.

[0076] S1041B: The first processing result is processed by i processing layers to obtain a second processing result, where i≥2 and i is an integer, and each of the i processing layers includes a convolution layer and a linear rectifier layer.

[0077] It should be understood that the value of i of the i processing layers described above can be set as required, and is not limited in the embodiments of the present application. For example, in an example, i = 3 can be described.

[0078] Optionally, as shown in FIG. 5, the first encoding network N1 can include i processing layers (in the example of FIG. 5, i = 3), each of which is composed of a convolution layer and a rectified linear units layer (which can be referred to as a ReLU layer for short). The convolution layer (Conv layer) can perform convolution processing on input data. The rectified linear units layer (ReLU layer) uses a rectified linear function (ReLU function) as an excitation function, enhances the nonlinear characteristics of the neural network, and can improve the training speed of the neural network without significantly affecting the generalization accuracy of the model.

[0079] The first processing result is input data of a first processing layer in the i processing layers, and is processed through the first to i processing layers. The result output by the i processing layer is the second processing result.

[0080] In some optional embodiments, the substep S1041B can be implemented through the following substeps S1041B1-S1041B3:

[0081] S1041B1: The first processing result is input data of a first processing layer in the i processing layers, so that the first processing layer processes the first processing result to obtain an output result of the first processing layer.

[0082] S1041B2: For a bth processing layer in the i processing layers: the input data of a (b-1)th processing layer and the output result of the (b-1)th processing layer are connected and processed, and the connection result is taken as input data of the bth processing layer, so that the bth processing layer processes the input data of the bth processing layer to obtain an output result of the bth processing layer, where 2≤b≤i and b is an integer.

[0083] For example, referring to the example shown in FIG. 5, where i = 3, the input data of the bth = 2 processing layer is the input data of the first processing layer (i.e., the (b-1)th processing layer) and the output result of the first processing layer, which are connected (Concat) to obtain a connection result. The second processing layer processes the connection result through its convolution layer and rectified linear units layer to obtain an output result of the second processing layer. The input data of the bth = 3 processing layer is the input data of the second processing layer (i.e., the (b-1)th processing layer) and the output result of the second processing layer, which are connected (Concat) to obtain a connection result. The third processing layer processes the connection result through its convolution layer and rectified linear units layer to obtain an output result of the third processing layer.

[0084] S1041B3: determining the output result of the i-th processing layer in the i processing layers as the second processing result.

[0085] For example, referring to the example shown in FIG. 5, the output result of the third processing layer in the i=3 processing layers is determined as the second processing result.

[0086] Based on this, in the embodiments of the present application, the first processing result can be effectively processed by the i processing layers through the optional implementation of the above sub-steps S1041B1-S1041B3 to obtain the second processing result, which is conducive to making the subsequently obtained first feature vector more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0087] S1041C: performing down-sampling processing on the second processing result to obtain a third processing result.

[0088] The second processing result can be down-sampled in any manner. For example, in an optional embodiment, the sub-step S1041C includes: performing addition processing on the input data of the i-th processing layer and the second processing result, and performing average pooling processing on the second processing result after the addition processing to obtain the third processing result.

[0089] It should be understood that the second processing result can be effectively down-sampled by such an average pooling manner to obtain the third processing result, which is conducive to making the subsequently obtained first feature vector more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0090] For example, referring to the first encoding network example shown in FIG. 5, in the i (i=3) processing layers, the input data of the third processing layer and the second processing result (i.e. the output result of the third processing layer) are subjected to addition (Add) processing. The first encoding network N1 includes a first average pooling layer (average pooling, Average Pooling, abbreviated as Avg Pool), which performs average pooling processing on the second processing result after the addition processing to achieve down-sampling processing and obtain the third processing result.

[0091] S1041D: processing the third processing result through j processing layers to obtain a fourth processing result, wherein j≥2 and j is an integer, and the j processing layers each include a convolution layer and a linear rectifier layer.

[0092] It should be understood that the value of j of the above j processing layers can be set according to actual needs, which is not limited in the embodiments of the present application. For example, in an example, j=3 can be described.

[0093] The structure of the processing layer in sub-step S1041D and the processing layer in sub-step S1041B is similar. Optionally, as shown in FIG. 5, the first encoding network N1 can further include j processing layers (in the example of FIG. 5, j = 3) after the first average pooling layer, each of which is composed of a convolution layer and a linear rectification layer. The convolution layer and the linear rectification layer have been described above and will not be repeated here.

[0094] The third processing result is taken as the input data of the first processing layer in the j processing layers, and is processed by the first to j processing layers. The result output by the jth processing layer is the fourth processing result.

[0095] In some optional embodiments, sub-step S1041D can be implemented by sub-steps S1041D1-S1041D3 as follows:

[0096] S1041D1: taking the third processing result as the input data of the first processing layer in the j processing layers, so that the first processing layer processes the third processing result to obtain the output result of the first processing layer.

[0097] S1041D2: for the cth processing layer in the j processing layers: connecting the input data of the (c-1)th processing layer and the output result of the (c-1)th processing layer, taking the connection result as the input data of the cth processing layer, so that the cth processing layer processes the input data of the cth processing layer to obtain the output result of the cth processing layer, where 2≤c≤j and c is an integer.

[0098] For example, referring to the example shown in FIG. 5, where j = 3, the input data of the c = 2th processing layer is the connection (Concat) result of the input data of the first processing layer (i.e., the (c-1)th processing layer) and the output result of the first processing layer, and the second processing layer processes the connection result through its convolution layer and linear rectification layer to obtain the output result of the second processing layer. The input data of the c = 3th processing layer is the connection (Concat) result of the input data of the second processing layer (i.e., the (c-1)th processing layer) and the output result of the second processing layer, and the third processing layer processes the connection result through its convolution layer and linear rectification layer to obtain the output result of the third processing layer.

[0099] S1041D3: determining the output result of the jth processing layer in the j processing layers as the fourth processing result.

[0100] For example, referring to the example shown in FIG. 5, the output result of the third processing layer in the j = 3 processing layers is determined as the fourth processing result.

[0101] Based on this, in the embodiments of the present application, through the optional implementation of the above sub-steps S1041D1-S1041D3, the third processing result can be effectively processed by the j processing layers to obtain the fourth processing result, which is conducive to making the subsequently obtained first feature vector more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0102] S1041E: performing down-sampling processing on the fourth processing result to obtain the first feature vector corresponding to the group of fingerprint images.

[0103] The fourth processing result can be down-sampled in any manner. For example, in an optional embodiment, sub-step S1041E includes: performing average pooling processing on the fourth processing result to obtain the first feature vector corresponding to the group of fingerprint images.

[0104] It should be understood that through such an average pooling manner, the fourth processing result can be effectively down-sampled to accurately and effectively obtain the first feature vector corresponding to the group of fingerprint images, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0105] For example, referring to the first encoding network example shown in FIG. 5, the first encoding network N1 further includes a second average pooling layer (Avg Pool). In the j (j=3) processing layers, the output result of the third processing layer (i.e., the fourth processing result) is input into the second average pooling layer, and the fourth processing result is subjected to average pooling processing by the second average pooling layer to achieve down-sampling processing, thereby obtaining the first feature vector.

[0106] Based on this, in the embodiments of the present application, through the optional implementation of the above sub-steps S1041A-S1041E, the first fingerprint feature extraction processing can be effectively performed on each group of fingerprint images, and the corresponding first feature vector can be accurately obtained, so as to facilitate the subsequent data processing of fingerprint anti-counterfeiting.

[0107] S1042: performing feature fusion processing on the g first feature vectors to obtain a first fused feature vector, and determining the first fused feature vector as the fused feature information.

[0108] Based on this, in the embodiments of the present application, through the optional implementation of steps S1041-S1042, on the one hand, by dividing the n frames of fingerprint images into g groups for processing, the relationship between the inter-frame differences and change information of the fingerprint images and the authenticity can be more conveniently modeled, effectively adapting to the demand for fingerprint anti-counterfeiting and improving the fingerprint anti-counterfeiting effect; on the other hand, by dividing the n frames of fingerprint images into g groups and then respectively performing first fingerprint feature extraction processing, and then performing feature fusion processing on the g first feature vectors respectively obtained, a first fusion feature vector carrying the fingerprint feature information of the g first feature vectors is obtained, and the first fusion feature vector is determined as the fusion feature information, which can ensure that the fusion feature information can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images, compared with the recognition mode of a single frame of fingerprint image, effectively enabling the fusion feature information to have an information increment more beneficial to the fingerprint anti-counterfeiting task, thereby effectively improving the accuracy of fingerprint anti-counterfeiting, so as to facilitate the present scheme to more accurately judge the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor and improve the fingerprint anti-counterfeiting effect.

[0109] In some optional embodiments, referring to the flowchart shown in FIG. 6, the manner of obtaining the first fusion feature vector in sub-step S1042 includes the following sub-steps S1042A-S1042C, specifically:

[0110] S1042A: Concatenate the g first feature vectors to obtain a second feature vector.

[0111] After obtaining the g first feature vectors, the g first feature vectors can be concatenated (Concat) to obtain a second feature vector. For example, taking the g=2 first feature vectors E0 and E1 obtained from the g=2 groups of fingerprint images g0 and g1 as an example, if the first feature vectors E0 and E1 are both 24×H×W feature vectors (where 24 represents the number of channels, H represents the height of the feature vector, and W represents the width of the feature vector), after the concatenation (Concat) processing, the result obtained is a 48×H×W feature vector.

[0112] S1042B: Perform second fingerprint feature extraction processing on the second feature vector to obtain a third feature vector.

[0113] In the embodiments of the present application, the second feature vector obtained can be further subjected to fingerprint feature extraction to obtain a third feature vector.

[0114] In some optional embodiments, referring to the flowchart shown in FIG. 7, sub-step S1042B includes the following sub-steps S1042B1-S1042B4, specifically:

[0115] S1042B1: processing the second feature vector through k processing layers to obtain a fifth processing result, where k>2 and k is an integer, and each of the k processing layers comprises a convolution layer and a linear rectifier layer.

[0116] It should be understood that the value of k of the k processing layers described above can be set according to actual needs, and is not limited in the embodiments of the present application. For example, in an example, k=3 can be described.

[0117] Optionally, referring to FIG. 8, a schematic diagram of a second encoding network of some optional neural network model according to an embodiment of the present application is shown. As shown in FIG. 8, the second encoding network N2 can comprise k processing layers (in the example of FIG. 8, k=3), and each of the k processing layers is composed of a convolution layer and a linear rectifier layer. The processing layer here is similar in structure to the processing layer described above.

[0118] The second feature vector is input data of the first processing layer in the k processing layers, and is processed through the first to k processing layers. The result output by the kth processing layer is the fifth processing result.

[0119] In some optional embodiments, the sub-step S1042B1 can be implemented through the following sub-steps SA1-S A3:

[0120] SA1: taking the second feature vector as input data of the first processing layer in the k processing layers, so that the first processing layer processes the second feature vector to obtain an output result of the first processing layer.

[0121] SA2: for the dth processing layer in the k processing layers, connecting the input data of the (d-1)th processing layer and the output result of the (d-1)th processing layer, taking the connection result as input data of the dth processing layer, so that the dth processing layer processes the input data of the dth processing layer to obtain an output result of the dth processing layer, where 2

[0122] For example, referring to the example shown in FIG. 8, where k=3, the input data of the dth=2 processing layer is the connection (Concat) result of the input data of the first processing layer (i.e., the (d-1)th processing layer) and the output result of the first processing layer, and the second processing layer processes the connection result through its convolution layer and linear rectifier layer to obtain an output result of the second processing layer. The input data of the dth=3 processing layer is the connection (Concat) result of the input data of the second processing layer (i.e., the (d-1)th processing layer) and the output result of the second processing layer, and the third processing layer processes the connection result through its convolution layer and linear rectifier layer to obtain an output result of the third processing layer.

[0123] SA3: determine the output result of the kth processing layer in the k processing layers as the fifth processing result.

[0124] For example, referring to the example shown in FIG. 8, the output result of the 3rd processing layer in the k=3 processing layers is determined as the fifth processing result.

[0125] Based on this, in the embodiments of the present application, the second feature vector can be effectively processed by the k processing layers through the optional implementation of the above-mentioned sub-steps SA1-SA3 to obtain the fifth processing result, which is conducive to making the third feature vector obtained subsequently more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0126] S1042B2: down-sampling processing is performed on the fifth processing result to obtain a sixth processing result.

[0127] The fifth processing result can be down-sampled in any manner. For example, in an optional embodiment, the sub-step S1042B2 includes: performing addition processing on the input data of the kth processing layer and the fifth processing result, and performing average pooling processing on the fifth processing result after the addition processing to obtain the sixth processing result.

[0128] It should be understood that the fifth processing result can be effectively down-sampled by such an average pooling manner to obtain the sixth processing result, which is conducive to making the third feature vector obtained subsequently more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0129] For example, referring to the second encoding network example shown in FIG. 8, in the k=3 processing layers, the input data of the 3rd processing layer and the second processing result (i.e. the output result of the 3rd processing layer) are subjected to addition (Add) processing. The second encoding network N2 includes a third average pooling layer (Avg Pool), which performs average pooling processing on the fifth processing result after the addition processing to achieve down-sampling processing and obtain the sixth processing result.

[0130] S1042B3: the sixth processing result is processed by h processing layers to obtain a seventh processing result, wherein h≥2 and h is an integer, and the h processing layers each include a convolution layer and a linear rectifier layer.

[0131] It should be understood that the value of h of the above-mentioned h processing layers can be set according to actual needs, which is not limited in the embodiments of the present application. For example, in an example, h=3 can be described.

[0132] The processing layer in sub-step S1042B3 and the processing layer in the foregoing are similar in structure. Optionally, as shown in FIG. 8, the second encoding network N2 can further include h processing layers (h=3 in the example shown in FIG. 8) after the third average pooling layer, each of which is composed of a convolution layer and a linear rectification layer, which have been described in the foregoing and will not be described herein again.

[0133] The sixth processing result is taken as input data of a first processing layer in the h processing layers, and is processed by the first to h processing layers. The result output by the h processing layer is the seventh processing result.

[0134] In some optional embodiments, sub-step S1042B3 can be implemented through sub-steps SB1 to SB3 as follows.

[0135] SB1: The sixth processing result is taken as input data of a first processing layer in the h processing layers, so that the first processing layer processes the sixth processing result to obtain an output result of the first processing layer.

[0136] SB2: For an e-th processing layer in the h processing layers, the input data of the e-1-th processing layer and the output result of the e-1-th processing layer are connected and processed, and the connection result is taken as input data of the e-th processing layer, so that the e-th processing layer processes the input data of the e-th processing layer to obtain an output result of the e-th processing layer, where 2≤e≤h and e is an integer.

[0137] For example, referring to the example shown in FIG. 8, where h=3, the input data of the e=2-th processing layer is the input data of the first processing layer (namely, the e-1-th processing layer) and the output result of the first processing layer, which are connected (Concat) to obtain a connection result, and the second processing layer processes the connection result through the convolution layer and the linear rectification layer to obtain an output result of the second processing layer. The input data of the e=3-th processing layer is the input data of the second processing layer (namely, the e-1-th processing layer) and the output result of the second processing layer, which are connected (Concat) to obtain a connection result, and the third processing layer processes the connection result through the convolution layer and the linear rectification layer to obtain an output result of the third processing layer.

[0138] SB3: The output result of the h-th processing layer in the h processing layers is determined as the seventh processing result.

[0139] For example, referring to the example shown in FIG. 8, the output result of the third processing layer in the h=3 processing layers is determined as the seventh processing result.

[0140] Based on this, in the embodiments of the present application, through the optional implementation of the above sub-steps SB1-SB3, the sixth processing result can be effectively processed by the h processing layers to obtain the seventh processing result, which is conducive to making the third feature vector obtained subsequently more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0141] S1042B4: performing downsampling processing on the seventh processing result to obtain the third feature vector.

[0142] The seventh processing result can be processed by any means. For example, in an optional embodiment, the sub-step S1042B4 includes: performing average pooling processing on the seventh processing result to obtain the third feature vector corresponding to the group of fingerprint images.

[0143] It should be understood that through such an average pooling manner, the seventh processing result can be effectively downsampled to accurately and effectively obtain the third feature vector, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0144] For example, referring to the second encoding network example shown in FIG. 8, the second encoding network N2 further includes a fourth average pooling layer (Avg Pool). In the h=3 processing layers, the output result of the third processing layer is input into the fourth average pooling layer, and the seventh processing result is processed by the fourth average pooling layer to achieve downsampling processing and obtain the third feature vector.

[0145] S1042C: performing depth convolution processing on the third feature vector to obtain a first fusion feature vector.

[0146] Optionally, after obtaining the third feature vector, the third feature vector is further processed by depth convolution processing to complete the fusion of fingerprint features and obtain the first fusion feature vector, so as to use the first fusion feature vector as the fusion feature information for further data processing of fingerprint anti-counterfeiting. Depth convolution (Depthwise Conv) processing is a convolution processing that does not change the number of input and output channels.

[0147] For example, referring to FIG. 8, the second encoding network N2 further includes a depth convolution layer after the second average pooling layer, and the third feature vector can be input into the depth convolution layer for depth convolution processing to obtain a processing result.

[0148] Based on this, in the embodiments of the present application, through the optional implementation of the above sub-steps S1042A-S1042C, the g first feature vectors can be processed by feature fusion processing, and the first fusion feature vector can be more accurately and effectively obtained, so as to determine the first fusion feature vector as the fusion feature information for further data processing of fingerprint anti-counterfeiting, and improve the effect of fingerprint anti-counterfeiting.

[0149] <Optional implementation two>

[0150] In optional implementation two, the n frames of fingerprint images are not grouped, but directly processed as a group of fingerprint images.

[0151] In some optional embodiments, referring to the flowchart shown in FIG. 9, the step S104 of “performing fingerprint feature extraction on the n frames of fingerprint images, and performing fusion processing on the extracted feature information to obtain fused feature information” includes the following sub-steps S202 and S204, specifically:

[0152] S202: performing third fingerprint feature extraction processing on the n frames of fingerprint images to obtain a fourth feature vector.

[0153] Optionally, in optional implementation two, the n frames of fingerprint images can be n frames of continuously collected fingerprint images. For example, taking n = 3 as an example, the n = 3 frames of fingerprint images can be denoted as fingerprint images F0, F1, and F2, respectively (which can be understood in combination with FIG. 10).

[0154] The embodiments of the present application extract the fourth feature vector that can reflect the fingerprint features of the n frames of fingerprint images respectively by performing third fingerprint feature extraction processing on the n frames of fingerprint images.

[0155] Optionally, the neural network model used in optional implementation two can include a third encoding network N3 and a fourth encoding network N4, which can be used to complete the fingerprint feature extraction and fusion of step S104 in any embodiment. The structure and function of the third encoding network N3 and the fourth encoding network N4 can complete the fingerprint feature extraction and fusion. This part will be described in combination below.

[0156] In some optional embodiments, referring to the flowchart shown in FIG. 11, the sub-step S202 includes the following sub-steps S2021 and S2022, specifically:

[0157] S2021: taking the n frames of fingerprint images as a group of fingerprint images, and performing convolution processing on the group of fingerprint images to obtain an eighth processing result.

[0158] For the convenience of describing the embodiments of the present application, the group of fingerprint images taken by the n frames of fingerprint images can be denoted as g2 group of fingerprint images.

[0159] Optionally, referring to FIG. 12, a schematic diagram of a third encoding network of some optional neural network model according to an embodiment of the present application is shown. As shown in FIG. 12, the third encoding network N3 can include a convolutional layer (which can be referred to as a Conv layer for short), and the input of the convolutional layer is the input of the third encoding network N3. When the g2 group of fingerprint images is input, the group of fingerprint images can be input into the convolutional layer, and the g2 group of fingerprint images is processed by the convolutional layer to obtain an eighth processing result.

[0160] S2022: processing the eighth processing result by m processing networks to obtain a fourth feature vector, wherein the processing network includes: t processing layers arranged in sequence, a first down-sampling processing layer, r processing layers, and a second down-sampling processing layer, each processing layer includes a convolutional layer and a rectified linear unit layer, m≥2, t≥2, r≥2, and m, t, and r are all integers.

[0161] It should be understood that the value of m of the m processing networks, the value of t of the t processing layers, and the value of r of the r processing layers described above can be set as required, and are not limited in the embodiments of the present application. For example, in an example, m=2, t=3, and r=3 can be described.

[0162] It should be understood that the processing layer here is similar to the processing layer described in the optional implementation manner one above, that is, each processing layer is composed of a convolutional layer and a rectified linear unit layer, wherein the convolutional layer (Conv layer) can perform convolution processing on the input data, the rectified linear unit layer (ReLU layer) uses a rectified linear function (ReLU function) as an excitation function to enhance the nonlinear characteristics of the neural network, which can improve the training speed of the neural network without significantly affecting the generalization accuracy of the model.

[0163] It should be understood that the first down-sampling processing layer and the second down-sampling processing layer can both realize down-sampling processing on the input data, which can be realized by any suitable manner. For example, the average pooling processing can be used to realize the down-sampling processing.

[0164] Optionally, referring to FIG. 12, a schematic diagram of a first processing network in the m=2 processing networks (for example, respectively referred to as a first processing network and a second processing network) is shown. The first processing network can be included in the third encoding network N3, and the first processing network includes t=3 processing layers arranged in sequence, a first down-sampling processing layer, r=3 processing layers, and a second down-sampling processing layer.

[0165] The eighth processing result is taken as input data of a first processing network in the m processing networks (such as the first processing network in FIG. 12), and after processing of each component structure in the first to m processing networks, the output result of the m processing network is the fourth feature vector.

[0166] Based on this, in the optional implementation of the above sub-steps S2021-S2022 in the embodiments of the present application, the n frames of fingerprint images are taken as a group of fingerprint images for third fingerprint feature extraction processing, and the eighth processing result is further processed through the m special structure processing networks, so that the fourth feature vector can be effectively obtained, thereby facilitating data processing for fingerprint anti-counterfeiting.

[0167] In some optional embodiments, the sub-step S2022 includes the following sub-steps S2022A-S2022C, specifically:

[0168] S2022A: for the first processing network in the m processing networks: processing the eighth processing result through the t processing layers of the first processing network to obtain a ninth processing result; performing down-sampling processing on the ninth processing result through the first down-sampling processing layer to obtain a first down-sampling processing result; processing the first down-sampling processing result through the r processing layers of the first processing network to obtain a tenth processing result; performing down-sampling processing on the tenth processing result through the second down-sampling processing layer to obtain a second down-sampling processing result; and taking the second down-sampling processing result as the output result of the first processing network.

[0169] In some optional embodiments, the sub-step S2022A includes the following sub-steps SC1-SC3:

[0170] SC1: taking the eighth processing result as input data of a first processing layer in the t processing layers of the first processing network, so that the first processing layer processes the eighth processing result to obtain an output result of the first processing layer.

[0171] It should be understood that the eighth processing result is taken as input data of a first processing layer in the t processing layers of the first processing network, and after processing of the first to t processing layers, the output result of the t processing layer is the ninth processing result.

[0172] SC2: for an fth processing layer in the t processing layers: performing connection processing on the input data of the f-1th processing layer and the output result of the f-1th processing layer, taking the connection result as input data of the fth processing layer, so that the fth processing layer processes the input data of the fth processing layer to obtain an output result of the fth processing layer, where 2≤f≤t and f is an integer.

[0173] For example, referring to the example shown in FIG12, in the first processing network (i.e., the first processing network), t=3, where the input data of the f=2nd processing layer is the input data of the first processing layer (i.e., the (f-1)th processing layer) and the output result of the first processing layer are concatenated. The second processing layer processes this concatenated result through its convolutional layer and linear rectifier layer to obtain the output result of the second processing layer. The input data of the f=3rd processing layer is the input data of the second processing layer (i.e., the (f-1)th processing layer) and the output result of the second processing layer are concatenated. The third processing layer processes this concatenated result through its convolutional layer and linear rectifier layer to obtain the output result of the third processing layer.

[0174] SC3: Determine the output of the t-th processing layer out of the t processing layers as the ninth processing result.

[0175] For example, referring to the example shown in FIG12, the output result of the third processing layer in the t=3 processing layers is determined as the ninth processing result.

[0176] Based on this, in the embodiments of this application, the optional implementation of the above sub-steps SC1 to SC3 can effectively process the eighth processing result through the t processing layers of the first processing network to obtain the ninth processing result, which is beneficial to make the subsequently obtained fourth feature vector more accurate and effective, so as to facilitate fingerprint anti-counterfeiting data processing.

[0177] The ninth processing result can be downsampled in any way. In some optional embodiments, the sub-step S2022A of "downsampling the ninth processing result through the first downsampling processing layer to obtain the first downsampling processing result" includes: adding the input data of the t-th processing layer and the ninth processing result, and then performing average pooling on the ninth processing result after addition through the first downsampling processing layer to obtain the first downsampling processing result.

[0178] It should be understood that this average pooling method can effectively downsample the ninth processing result to obtain the first downsampling result, which helps to make the subsequent fourth feature vector more accurate and effective, so as to facilitate fingerprint anti-counterfeiting data processing.

[0179] For example, referring to the third encoding network example shown in FIG. 12, in the t (t = 3) processing layers, the input data of the third processing layer and the ninth processing result (i.e., the output result of the third processing layer) are subjected to addition (Add) processing. The third encoding network N3 includes a first down-sampling processing layer in the form of an average pooling layer (Avg Pool), and the ninth processing result subjected to the addition processing is subjected to average pooling processing by the first down-sampling processing layer to achieve down-sampling processing, thereby obtaining a first down-sampling processing result.

[0180] In some optional embodiments, the "processing, by the r processing layers of the first processing network, the first down-sampling processing result to obtain a tenth processing result" in substep S2022A includes the following substeps SD1-SD3:

[0181] SD1: determining the first down-sampling processing result as the input data of the first processing layer among the r processing layers of the first processing network, so that the first processing layer processes the first down-sampling processing result to obtain the output result of the first processing layer.

[0182] It should be understood that the first down-sampling processing result is taken as the input data of the first processing layer among the r processing layers of the first processing network, and is processed by the first to rth processing layers, and the result output by the rth processing layer is the tenth processing result.

[0183] SD2: performing connection processing on the input data of the q-1th processing layer and the output result of the q-1th processing layer, and taking the connection result as the input data of the qth processing layer, so that the qth processing layer processes the input data of the qth processing layer to obtain the output result of the qth processing layer, wherein 2≤q≤r and q is an integer.

[0184] For example, referring to the example shown in FIG. 12, in the first processing network (i.e., the first processing network), r = 3, wherein the input data of the q = 2th processing layer is the input data of the first processing layer (i.e., the q-1th processing layer) and the output result of the first processing layer subjected to connection (Concat) processing, and the connection result is processed by the convolution layer and the linear rectifier layer of the second processing layer to obtain the output result of the second processing layer. The input data of the q = 3th processing layer is the input data of the second processing layer (i.e., the q-1th processing layer) and the output result of the second processing layer subjected to connection (Concat) processing, and the connection result is processed by the convolution layer and the linear rectifier layer of the third processing layer to obtain the output result of the third processing layer.

[0185] SD3: determining the output result of the rth processing layer among the r processing layers as the tenth processing result.

[0186] For example, referring to the example shown in FIG. 12, the output result of the third processing layer in the r = 3 processing layers is determined as the tenth processing result.

[0187] Based on this, in the embodiments of the present application, the first down-sampling processing result can be effectively processed by the r processing layers of the first processing network through the optional implementation manners of the above-mentioned sub-steps SD1-SD3 to obtain the tenth processing result, which is conducive to making the fourth feature vector obtained subsequently more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0188] The tenth processing result can be down-sampled in any manner. In some optional embodiments, the "down-sampling processing the tenth processing result through the second down-sampling processing layer to obtain the second down-sampling processing result" in the sub-step S2022A includes: performing average pooling processing on the tenth processing result through the second down-sampling processing layer to obtain the second down-sampling processing result.

[0189] It should be understood that the tenth processing result can be effectively down-sampled through such an average pooling manner to accurately and effectively obtain the second down-sampling processing result, which is conducive to making the fourth feature vector obtained subsequently more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0190] For example, referring to the third encoding network example shown in FIG. 12, the third encoding network N3 includes a second down-sampling processing layer in the form of an average pooling layer (Avg Pool). The output result of the third processing layer in the r (r = 3) processing layers is input into the second down-sampling processing layer in the form of an average pooling layer (Avg Pool), and the tenth processing result is average-pooled through the second down-sampling processing layer to achieve down-sampling processing to obtain the second down-sampling processing result.

[0191] S2022B: For the yth processing network in the m processing networks: the output result of the y-1th processing network is determined as the input data of the first processing layer in the t processing layers of the yth processing network, and the yth processing network is processed until the corresponding second down-sampling processing result is calculated as the output result of the yth processing network, wherein 2≤y≤m and y is an integer.

[0192] It should be understood that for the y = 2-mth processing network in the m processing networks, the related optional processing manners of the t processing layers, the first down-sampling processing layer, and the r processing layers inside are similar to the related contents of the first processing network in the m processing networks, and can be understood with reference to the related contents of the first processing network in the m processing networks, which will not be described here.

[0193] Optionally, the y=2~mth processing network in the m processing networks can be arranged in the fourth encoding network N4. For example, referring to FIG. 13, a schematic diagram of the fourth encoding network of some optional neural network models in the embodiments of the present application is shown. In combination with FIG. 12 and FIG. 13, there are m=2 processing networks in the third encoding network N3 and the fourth encoding network N4, wherein the third encoding network N3 is arranged with the 1st processing network (i.e., the first processing network), and the fourth encoding network N4 is arranged with the 2nd processing network (i.e., the second processing network). The output result of the first processing network is input to the second processing network as input data for further processing. It should be understood that FIG. 12 and FIG. 13 are examples and are not any limitation on the embodiments of the present application.

[0194] S2022C: determining the output result of the mth processing network in the m processing networks as the fourth feature vector.

[0195] For example, referring to the example shown in FIG. 13, after obtaining the output result of the m=2th processing network, it can be determined as the fourth feature vector.

[0196] Based on this, through the optional implementation manner of the above sub-steps S2022A-S2022C in the embodiments of the present application, the eighth processing result can be effectively processed by the m processing networks to obtain the fourth feature vector, which is beneficial to make the subsequently obtained second fusion feature vector more accurate and effective, so as to facilitate the data processing of fingerprint anti-counterfeiting.

[0197] S204: performing depth convolution processing on the fourth feature vector to obtain a second fusion feature vector, and determining the second fusion feature vector as the fusion feature information.

[0198] Optionally, after obtaining the fourth feature vector, the fourth feature vector is further subjected to depth convolution processing to complete the fusion of the fingerprint features, so as to obtain the second fusion feature vector, which is used as the fusion feature information for further data processing of fingerprint anti-counterfeiting. Depth convolution (Depthwise Conv) processing is a convolution processing that does not change the number of input and output channels.

[0199] For example, referring to FIG. 13, after the second down-sampling processing layer in the form of the average pooling layer (Avg Pool) in the fourth encoding network N4, a depth convolution layer is further included, and the fourth feature vector can be input to the depth convolution layer for depth convolution processing to obtain a processing result.

[0200] Based on this, in the embodiments of the present application, through the optional implementation of the above sub-steps S202-S204, the n frames of fingerprint images are taken as a group of fingerprint images for third fingerprint feature extraction processing, and then the fourth feature vector obtained by processing is subjected to deep convolution processing, so as to effectively obtain the second fusion feature vector fusing the fingerprint feature information carried by the n frames of fingerprint images, and determine the second fusion feature vector as the fusion feature information, which can ensure that the fusion feature information can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images, compared with the recognition mode of single frame of fingerprint image, the fusion feature information can effectively have more beneficial information increment for the fingerprint anti-counterfeiting task, thereby effectively improving the accuracy of fingerprint anti-counterfeiting, so as to facilitate the present scheme to more accurately judge the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor and improve the effect of fingerprint anti-counterfeiting.

[0201] S106: determining the authenticity of the fingerprint corresponding to the n frames of fingerprint images based on the fusion feature information.

[0202] After obtaining the fusion feature information in step S104, the authenticity of the fingerprint can be determined based on the fusion feature information to realize the function of fingerprint anti-counterfeiting. It should be understood that the implementation of step S106 is not specifically limited in the embodiments of the present application.

[0203] Based on this, in the embodiments of the present application, the n frames of fingerprint images collected by the ultrasonic fingerprint sensor can be obtained through the optional implementation including the above steps S104-S106, wherein n≥2 and n is an integer, then the fusion feature information is obtained by performing fingerprint feature fusion processing on the n frames of fingerprint images through the neural network model, and then the authenticity of the fingerprint corresponding to the n frames of fingerprint images can be determined based on the fusion feature information. Therefore, the technical scheme of the embodiments of the present application fuses the feature information of multiple frames of fingerprint images through the neural network model, can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images, compared with the recognition mode of single frame of fingerprint image, the fusion feature information used to determine the authenticity of the fingerprint in the technical scheme of the present application has more beneficial information increment for the fingerprint anti-counterfeiting task, thereby effectively improving the accuracy of fingerprint anti-counterfeiting, so that the fingerprint anti-counterfeiting scheme of the present application can more accurately judge the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor and improve the effect of fingerprint anti-counterfeiting, and since the neural network model is used for fingerprint feature fusion, the accuracy of the obtained fusion feature information can be effectively improved, and the processing efficiency of fingerprint anti-counterfeiting can be improved.

[0204] In some optional embodiments, referring to FIG. 14, step S106 includes the following sub-steps S1061 and S1062, specifically:

[0205] S1061: Determine score information for determining the authenticity of the n-frame fingerprint image corresponding fingerprint based on the fusion feature information.

[0206] The sub-step S1061 can calculate the score information based on the fusion feature information in any suitable manner. Optionally, the neural network model in the embodiments of the present application can further include a classifier. Optionally, the fusion feature information (for example, the first fusion feature vector in the above-mentioned "optional implementation one" or the second fusion feature vector in the above-mentioned "optional implementation two") can be input into the classifier in the pre-trained neural network model, and the score information (for example, which can be understood in combination with the schematic diagrams shown in FIG. 3 and FIG. 10) can be output by the classifier. Optionally, the classifier can perform a binary classification task of true / false fingerprints.

[0207] Optionally, the sigmoid function can be used in the classifier in the neural network model, and the fusion feature information can be processed into a floating-point number between 0 and 1 by the sigmoid function, and the obtained floating-point number can be determined as the score information.

[0208] S1062: Determine the classification result for indicating the authenticity of the n-frame fingerprint image corresponding fingerprint according to the score information and the preset threshold condition.

[0209] After obtaining the score information, the score information can be combined with the preset threshold condition for judgment, and according to the judgment, the n-frame fingerprint image corresponding fingerprint can be classified as a true / false fingerprint, and the classification result for indicating the authenticity of the n-frame fingerprint image corresponding fingerprint can be obtained.

[0210] The preset threshold condition can be preset as needed. For example, in an optional embodiment, after it is determined that the score information is greater than or equal to a certain preset threshold, the obtained classification result indicates that the n-frame fingerprint image corresponding fingerprint is a true fingerprint, and after it is determined that the score information is less than the preset threshold, the obtained classification result indicates that the n-frame fingerprint image corresponding fingerprint is a false fingerprint. For example, the above-mentioned preset threshold can be set as needed, for example, taking the score information as a floating-point number between 0 and 1 as an example, the preset threshold can be simply understood as 0.7, 0.8, etc. It should be understood that for different application scenarios, the above-mentioned preset threshold can be set to different values to adapt to the needs of different application scenarios.

[0211] Optionally, during the training of the foregoing neural network model, the training data set can be composed of the collected multiple frames of fingerprint images and their corresponding true / false finger labels. If the neural network model is trained with the label of true finger data being 1 and the label of false finger data being 0, the score information in the form of a floating point number output by the classifier can represent the confidence of the model that the fingerprint corresponding to the n frames of fingerprint images is a true finger fingerprint, and then the fingerprint authenticity can be determined according to the score information, the true / false fingerprint classification is realized, and the classification result indicating the authenticity of the fingerprint corresponding to the n frames of fingerprint images is obtained.

[0212] For example, Table 1 below shows the performance gain comparison of the technical solution in the embodiment of the application (which can be referred to as a multi-frame input model) that uses multiple frames of fingerprint images to extract features and fusion for fingerprint anti-counterfeiting, compared with the technical solution (which can be referred to as a single-frame input model) in the related art that uses a single frame of fingerprint image for fingerprint anti-counterfeiting in an example experiment:

[0213] Table 1

[0214] It can be seen that in the example experiment: in terms of the false live reject ratio (FLRR), the technical solution of the embodiment of the application is 1.25%, the technical solution of the related art is 4.68%, and the false live reject ratio of the technical solution of the embodiment of the application is lower than that of the technical solution of the related art; in terms of the spoof accept ratio (SAR), the technical solution of the embodiment of the application is 5.75%, the technical solution of the related art is 9.65%, and the false live reject ratio of the technical solution of the embodiment of the application is lower than that of the technical solution of the related art; the F1 score is an index combining the false live reject ratio FLRR and the spoof accept ratio SAR, and can be directly used for the overall performance of the classifier of the fingerprint anti-counterfeiting neural network model (which can also be understood as the overall fingerprint anti-counterfeiting performance). It can be seen that the F1 score of the technical solution of the embodiment of the application is 0.9645, while the F1 score of the technical solution of the related art is 0.9277, the F1 score of the technical solution of the embodiment of the application is higher than that of the technical solution of the related art, and the overall performance of the technical solution of the embodiment of the application is superior to that of the related art. It can be seen that the technical solution of the embodiment of the application effectively improves the accuracy of fingerprint anti-counterfeiting, can more accurately determine true / false fingerprints, and improves the effect of fingerprint anti-counterfeiting.

[0215] It can be understood that the foregoing description of the fingerprint anti-counterfeiting method is only as some exemplary description of the embodiment of the application and does not limit the embodiment of the application in any way.

[0216] In summary, the fingerprint anti-counterfeiting scheme provided in the embodiments of the present application can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images when performing fingerprint anti-counterfeiting. Compared with the identification mode through a single frame of fingerprint image, the fingerprint anti-counterfeiting scheme can utilize more information increments beneficial to the fingerprint anti-counterfeiting task through the neural network model to perform fingerprint anti-counterfeiting, thereby effectively ensuring the accuracy. Moreover, since the fingerprint anti-counterfeiting of n frames of fingerprint images is realized through the neural network model, the accuracy can be further ensured, so that the fingerprint anti-counterfeiting scheme can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improve the fingerprint anti-counterfeiting effect, and improve the processing efficiency of fingerprint anti-counterfeiting. In some optional implementation manners, the feature information of multiple frames of fingerprint images can be fused through the neural network model, the inter-frame differences and change information between multiple frames of fingerprint images can be effectively utilized, and compared with the identification mode through a single frame of fingerprint image, the fusion feature information used to determine the authenticity of the fingerprint in the technical scheme has more information increments beneficial to the fingerprint anti-counterfeiting task, thereby effectively improving the accuracy of fingerprint anti-counterfeiting. Therefore, the fingerprint anti-counterfeiting scheme can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improve the fingerprint anti-counterfeiting effect, and effectively improve the accuracy of the obtained fusion feature information and the processing efficiency of fingerprint anti-counterfeiting.

[0217] It should be further noted that some fingerprint anti-counterfeiting schemes in the related art need to be assisted by touch signals. Compared with the technical scheme in the related art, the technical scheme in the embodiments of the present application can have more information increments beneficial to the fingerprint anti-counterfeiting task, thereby effectively improving the accuracy of fingerprint anti-counterfeiting, more accurately determining the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improving the fingerprint anti-counterfeiting effect, and improving the processing efficiency of fingerprint anti-counterfeiting. On the basis of the technical effects, the technical scheme has the following benefits: on the one hand, the technical scheme does not need the hardware support of the touch screen, thereby reducing the use cost and increasing the application range; on the other hand, when the neural network model is used to realize fingerprint anti-counterfeiting, the technical scheme only needs to collect fingerprint images and does not need to synchronously collect touch signals when collecting the data set used by the neural network model, thereby reducing the time and labor cost of data set construction; on the other hand, when the technical scheme is realized through the neural network model, since only the encoder network needs to be constructed for the fingerprint image, the same network weight can be reused for multiple frames of fingerprint images, and the encoder network does not need to be constructed for the touch signal, so that the network inference cost is also reduced.

[0218] Based on the same inventive concept as the fingerprint anti-counterfeiting method provided in the first aspect, referring to FIG. 15, according to the second aspect of the embodiments of the present application, a fingerprint anti-counterfeiting device 100 is further provided, comprising:

[0219] The acquisition module 102 is configured to acquire n frames of fingerprint images collected by the ultrasonic fingerprint sensor, where n≥2 and n is an integer.

[0220] The determination module 104 is configured to determine the authenticity of the fingerprint corresponding to the n frames of fingerprint images by using the neural network model.

[0221] The fingerprint anti-counterfeiting scheme provided in the embodiments of the present application can effectively utilize the inter-frame differences and change information between multiple frames of fingerprint images during fingerprint anti-counterfeiting. Compared with the identification method using a single frame of fingerprint image, the neural network model can utilize more information increments beneficial to the fingerprint anti-counterfeiting task to perform fingerprint anti-counterfeiting, thereby effectively ensuring the accuracy. Furthermore, since the neural network model is used to perform fingerprint anti-counterfeiting on n frames of fingerprint images, the accuracy can be further ensured. Therefore, the fingerprint anti-counterfeiting scheme of the embodiments of the present application can more accurately determine the authenticity of the fingerprint collected by the ultrasonic fingerprint sensor, improve the fingerprint anti-counterfeiting effect, and improve the processing efficiency of fingerprint anti-counterfeiting.

[0222] In some optional embodiments, the n frames of fingerprint images are n frames of fingerprint images collected after the ultrasonic fingerprint sensor transmits a single ultrasonic signal.

[0223] In some optional embodiments, the n frames of fingerprint images are n frames of fingerprint images collected after the ultrasonic fingerprint sensor transmits multiple ultrasonic signals. Optionally, at least two ultrasonic signals in the multiple ultrasonic signals are ultrasonic signals of different frequencies.

[0224] In some optional embodiments, the determination module 104 is specifically configured to perform fingerprint feature fusion processing on the n frames of fingerprint images by using the neural network model to obtain fusion feature information, and determine the authenticity of the fingerprint corresponding to the n frames of fingerprint images based on the fusion feature information.

[0225] In some optional embodiments, the determination module 104 is specifically configured to perform fingerprint feature extraction on the n frames of fingerprint images by using the neural network model, and perform fusion processing on the extracted feature information to obtain fusion feature information.

[0226] In some optional embodiments, the fingerprint anti-counterfeiting device 100 is further configured to: divide the n frames of fingerprint images into g groups of fingerprint images, each group of fingerprint images including a frames of fingerprint images, where 2≤g≤n, n / g≤a<n, and g and a are integers; and the determination module 104 is specifically configured to: perform first fingerprint feature extraction processing on the g groups of fingerprint images respectively to obtain g first feature vectors; perform feature fusion processing on the g first feature vectors to obtain a first fused feature vector, and determine the first fused feature vector as the fused feature information.

[0227] In some optional embodiments, a≥2.

[0228] In some optional embodiments, the a frames of fingerprint images in each group of fingerprint images are a frames of continuously collected fingerprint images, and there is an intersection of fingerprint images between at least two groups of fingerprint images in the g groups of fingerprint images; or, the a frames of fingerprint images in each group of fingerprint images are a frames of continuously collected fingerprint images, and there is no intersection of fingerprint images between any two groups of fingerprint images in the g groups of fingerprint images.

[0229] In some optional embodiments, the determination module 104 is specifically configured to, for each group of fingerprint images in the g groups of fingerprint images: perform convolution processing on the group of fingerprint images to obtain a first processing result; perform processing on the first processing result through i processing layers to obtain a second processing result, where i≥2 and i is an integer, each of the i processing layers includes a convolution layer and a linear rectification layer; perform down-sampling processing on the second processing result to obtain a third processing result; perform processing on the third processing result through j processing layers to obtain a fourth processing result, where j≥2 and j is an integer, each of the j processing layers includes a convolution layer and a linear rectification layer; and perform down-sampling processing on the fourth processing result to obtain a first feature vector corresponding to the group of fingerprint images.

[0230] In some optional embodiments, the determining module 104 is specifically configured to: take the first processing result as input data of a first processing layer in the i processing layers, so that the first processing layer processes the first processing result to obtain an output result of the first processing layer; for a bth processing layer in the i processing layers, perform connection processing on input data of a (b-1)th processing layer and the output result of the (b-1)th processing layer, take a connection result as input data of the bth processing layer, so that the bth processing layer processes the input data of the bth processing layer to obtain an output result of the bth processing layer, where 2≤b≤i and b is an integer; and take an output result of an i th processing layer in the i processing layers as the second processing result. The determining module 104 is specifically configured to: perform addition processing on input data of the i th processing layer and the second processing result, and perform average pooling processing on the second processing result after the addition processing to obtain the third processing result.

[0231] In some optional embodiments, the determining module 104 is specifically configured to: take the third processing result as input data of a first processing layer in the j processing layers, so that the first processing layer processes the third processing result to obtain an output result of the first processing layer; for a cth processing layer in the j processing layers, perform connection processing on input data of a (c-1)th processing layer and the output result of the (c-1)th processing layer, take a connection result as input data of the cth processing layer, so that the cth processing layer processes the input data of the cth processing layer to obtain an output result of the cth processing layer, where 2≤c≤j and c is an integer; and take an output result of a j th processing layer in the j processing layers as the fourth processing result. The determining module 104 is specifically configured to: perform average pooling processing on the fourth processing result to obtain the first feature vector corresponding to the group of fingerprint image pairs.

[0232] In some optional embodiments, the determining module 104 is specifically configured to: perform connection processing on the g first feature vectors to obtain a second feature vector; perform second fingerprint feature extraction processing on the second feature vector to obtain a third feature vector; and perform deep convolution processing on the third feature vector to obtain the first fusion feature vector.

[0233] In some optional embodiments, the determining module 104 is specifically configured to: process the second feature vector through k processing layers to obtain a fifth processing result, where k is an integer greater than or equal to 2, each of the k processing layers comprises a convolutional layer and a linear rectifier layer; perform down-sampling processing on the fifth processing result to obtain a sixth processing result; process the sixth processing result through h processing layers to obtain a seventh processing result, where h is an integer greater than or equal to 2, each of the h processing layers comprises a convolutional layer and a linear rectifier layer; and perform down-sampling processing on the seventh processing result to obtain the third feature vector.

[0234] In some optional embodiments, the determining module 104 is specifically configured to: take the second feature vector as input data of a first processing layer in the k processing layers, so that the first processing layer processes the second feature vector to obtain output data of the first processing layer; for a dth processing layer in the k processing layers, where 2 is less than or equal to d and d is an integer, perform connection processing on the input data of a (d-1)th processing layer and the output data of the (d-1)th processing layer, take the connection result as input data of the dth processing layer, so that the dth processing layer processes the input data of the dth processing layer to obtain output data of the dth processing layer; take output data of a kth processing layer in the k processing layers as the fifth processing result; and the determining module 104 is specifically configured to: perform addition processing on the input data of the kth processing layer and the fifth processing result, and perform average pooling processing on the fifth processing result after the addition processing to obtain the sixth processing result.

[0235] In some optional embodiments, the determining module 104 is specifically configured to: take the sixth processing result as input data of a first processing layer in the h processing layers, so that the first processing layer processes the sixth processing result to obtain output data of the first processing layer; for an e th processing layer in the h processing layers, where 2 is less than or equal to e and e is an integer, perform connection processing on the input data of a (e-1)th processing layer and the output data of the (e-1)th processing layer, take the connection result as input data of the e th processing layer, so that the e th processing layer processes the input data of the e th processing layer to obtain output data of the e th processing layer; take output data of a h th processing layer in the h processing layers as the seventh processing result; and the determining module 104 is specifically configured to: perform average pooling processing on the seventh processing result to obtain the third feature vector.

[0236] In some optional embodiments, the determining module 104 is specifically configured to: perform third fingerprint feature extraction processing on the n frames of fingerprint images to obtain a fourth feature vector; perform deep convolution processing on the fourth feature vector to obtain a second fusion feature vector, and determine the second fusion feature vector as the fusion feature information.

[0237] In some optional embodiments, the determining module 104 is specifically configured to: take the n frames of fingerprint images as a group of fingerprint images, and perform convolution processing on the group of fingerprint images to obtain an eighth processing result; process the eighth processing result through m processing networks to obtain the fourth feature vector, wherein the processing network comprises: t processing layers, a first down-sampling processing layer, r processing layers, and a second down-sampling processing layer arranged in sequence, each of the processing layers comprises a convolution layer and a linear rectification layer, m≥2, t≥2, r≥2, and m, t, and r are all integers.

[0238] In some optional embodiments, the determining module 104 is specifically configured to: for the first processing network in the m processing networks: process the eighth processing result through the t processing layers of the first processing network to obtain a ninth processing result; perform down-sampling processing on the ninth processing result through the first down-sampling processing layer to obtain a first down-sampling processing result; process the first down-sampling processing result through the r processing layers of the first processing network to obtain a tenth processing result; perform down-sampling processing on the tenth processing result through the second down-sampling processing layer to obtain a second down-sampling processing result; take the second down-sampling processing result as the output result of the first processing network; for the yth processing network in the m processing networks: take the output result of the y-1th processing network as the input data of the first processing layer in the t processing layers of the yth processing network, and perform processing through the yth processing network until the yth processing network calculates a corresponding second down-sampling processing result as the output result of the yth processing network, wherein 2≤y≤m and y is an integer; and take the output result of the mth processing network in the m processing networks as the fourth feature vector.

[0239] In some optional embodiments, the determining module 104 is specifically configured to: determine the eighth processing result as input data of a first processing layer in the t processing layers of the first processing network, so that the first processing layer processes the eighth processing result to obtain output result of the first processing layer; for a f th processing layer in the t processing layers: perform connection processing on the input data of the f-1 th processing layer and the output result of the f-1 th processing layer, and take the connection result as input data of the f th processing layer, so that the f th processing layer processes the input data of the f th processing layer to obtain output result of the f th processing layer, where 2≤f≤t and f is an integer; and determine the output result of a t th processing layer in the t processing layers as the ninth processing result; and the determining module 104 is specifically configured to: perform addition processing on the input data of the t th processing layer and the ninth processing result, and perform average pooling processing on the ninth processing result after the addition processing by a first down-sampling processing layer to obtain a first down-sampling processing result.

[0240] In some optional embodiments, the determining module 104 is specifically configured to: determine the first down-sampling processing result as input data of a first processing layer in the r processing layers of the first processing network, so that the first processing layer processes the first down-sampling processing result to obtain output result of the first processing layer; for a q th processing layer in the r processing layers: perform connection processing on the input data of the q-1 th processing layer and the output result of the q-1 th processing layer, and take the connection result as input data of the q th processing layer, so that the q th processing layer processes the input data of the q th processing layer to obtain output result of the q th processing layer, where 2≤q≤r and q is an integer; and determine the output result of a r th processing layer in the r processing layers as the tenth processing result; and the determining module 104 is specifically configured to: perform average pooling processing on the tenth processing result by a second down-sampling processing layer to obtain a second down-sampling processing result.

[0241] In some optional embodiments, the determining module 104 is specifically configured to: determine score information for determining the authenticity of the fingerprint corresponding to the n frames of fingerprint images based on the fusion feature information; and determine a classification result for indicating the authenticity of the fingerprint corresponding to the n frames of fingerprint images according to the score information and a preset threshold condition.

[0242] The various optional embodiments of the fingerprint anti-counterfeiting device 100 in the embodiments of the present application have been described in detail in the foregoing fingerprint anti-counterfeiting method embodiments, and thus the related content and beneficial effects can be understood with reference to the content of the various optional embodiments of the fingerprint anti-counterfeiting method, which will not be described again here.

[0243] According to a third aspect of the embodiments of the present application, an electronic device is provided, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus; the memory is configured to store a computer program; and the processor is configured to execute the fingerprint anti-counterfeiting method according to the first aspect by running the computer program stored in the memory.

[0244] FIG. 16 shows a structural block diagram of an optional electronic device according to the embodiments of the present application. The embodiments of the present application do not limit the specific implementation of the electronic device 1600. For example, referring to FIG. 16, the electronic device 1600 provided by the embodiments of the present application comprises a processor 1602, a communication interface 1604, a memory 1606 and a communication bus 1608. Wherein:

[0245] The processor 1602, the communication interface 1604 and the memory 1606 complete communication with each other through the communication bus 1608.

[0246] The communication interface 1604 is configured to communicate with other electronic devices or servers.

[0247] The processor 1602 is configured to execute the computer program 1610, and specifically can execute the related steps in any of the fingerprint anti-counterfeiting methods.

[0248] Specifically, the computer program 1610 can include program codes, which include computer operation instructions.

[0249] The processor 1602 can be a CPU, or a GPU (Graphic Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the smart device can be processors of the same type, such as one or more CPUs; or can be processors of different types, such as one or more CPUs and one or more ASICs.

[0250] The memory 1606 is configured to store the computer program 1610. The memory 1606 can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0251] The computer program 1610 can be specifically used to enable the processor 1602 to perform the fingerprint anti-counterfeiting method in any of the preceding embodiments.

[0252] The specific implementation of each step in the computer program 1610 can refer to the corresponding description in the corresponding steps and units in any of the preceding fingerprint anti-counterfeiting method embodiments, and details are not described herein. It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, the specific working processes of the devices and modules described above can refer to the corresponding process descriptions in the preceding method embodiments, and details are not described herein.

[0253] The electronic device 1600 in the embodiments of the present application has been described in detail in the preceding fingerprint anti-counterfeiting method embodiments, and therefore its related content and beneficial effects can be understood with reference to the above method embodiments, and details are not described herein.

[0254] According to a fourth aspect of the embodiments of the present application, the embodiments of the present application further provide a computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the fingerprint anti-counterfeiting method described in any of the preceding method embodiments. The computer storage medium includes but is not limited to a compact disc read-only memory (CD-ROM), a random access memory (RAM), a floppy disk, a hard disk, or a magneto-optical disk, etc.

[0255] According to a fifth aspect of the embodiments of the present application, the embodiments of the present application further provide a computer program product including a computer program, the computer program being executed by a processor to implement the fingerprint anti-counterfeiting method described in any of the preceding method embodiments.

[0256] The fingerprint anti-counterfeiting device 100 / electronic device 1600 / computer storage medium / computer program product embodiments in the embodiments of the present application have been described in detail in the preceding fingerprint anti-counterfeiting method embodiments, and therefore their related content and beneficial effects can be understood with reference to the above method embodiments, and details are not described herein.

[0257] In addition, it should be noted that the information related to the user (including but not limited to user device information, user personal information, etc.) and data (including but not limited to sample data for training a model, data for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or authorized by all parties, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose authorization or refusal.

[0258] It should be noted that, according to the needs of implementation, each component / step described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or part operations of the components / steps can be combined into a new component / step, to achieve the purpose of the embodiments of the present application.

[0259] The above method according to the embodiments of the present application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium such as a CD-ROM, a RAM, a floppy disk, a hard disk, or a magneto-optical disk, or be downloaded through a network originally stored in a remote recording medium or a non-transitory machine readable medium and stored in a local recording medium, so that the method described herein can be stored in a recording medium on a general computer, a special purpose processor, or programmable or special hardware (such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA)) using such software processing. It can be understood that the computer, processor, microprocessor controller, or programmable hardware includes a storage component (for example, Random Access Memory (RAM), Read-Only Memory (ROM), flash memory, etc.) that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor, or hardware, the method described herein is implemented. In addition, when a general computer accesses the code for implementing the method shown herein, the execution of the code will convert the general computer into a special computer for executing the method shown herein.

[0260] Those skilled in the art can realize that the units and method steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for a specific application, but such implementation should not be considered beyond the scope of the embodiments of the present application.

[0261] The above embodiments are only used to illustrate the present application, and not to limit the present application, and those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application, therefore all equivalent technical solutions also belong to the scope of the present application, the patent protection scope of the present application should be defined by the claims.

Claims

1. A fingerprint anti-counterfeiting method, comprising: Acquire n frames of fingerprint images captured by an ultrasonic fingerprint sensor, where n≥2 and n is an integer; The authenticity of the fingerprints corresponding to the n frames of fingerprint images is determined by a neural network model.

2. The method according to claim 1, wherein, The n-frame fingerprint images are the n-frame fingerprint images acquired after the ultrasonic fingerprint sensor emits a single ultrasonic signal.

3. The method according to claim 1, wherein, The n-frame fingerprint images are the n-frame fingerprint images acquired after the ultrasonic fingerprint sensor emits multiple ultrasonic signals. At least two of the multiple ultrasonic signals are ultrasonic signals of different frequencies.

4. The method according to any one of claims 1-3, wherein, The step of determining the authenticity of the fingerprints corresponding to the n frames of fingerprint images using a neural network model includes: The fingerprint feature fusion process is performed on the n frames of fingerprint images using a neural network model to obtain fused feature information; Based on the fused feature information, the authenticity of the fingerprint corresponding to the n frames of fingerprint images is determined.

5. The method according to claim 4, wherein, The step of performing fingerprint feature fusion processing on the n frames of fingerprint images using a neural network model to obtain fused feature information includes: Fingerprint features are extracted from the n frames of fingerprint images using a neural network model, and the extracted feature information is then fused to obtain fused feature information.

6. The method according to claim 5, wherein, Before extracting fingerprint features from the n frames of fingerprint images using a neural network model, the method further includes: dividing the n frames of fingerprint images into g groups of fingerprint images, each group of fingerprint images including a frames of fingerprint images, where 2≤g≤n, n / g≤a<n, and g and a are both integers; a≥2; The step of extracting fingerprint features from the n frames of fingerprint images and fusing the extracted feature information to obtain fused feature information includes: The first fingerprint feature extraction process is performed on the g groups of fingerprint images respectively to obtain g first feature vectors; The g first feature vectors are subjected to feature fusion processing to obtain a first fused feature vector, and the first fused feature vector is determined as the fused feature information.

7. The method according to claim 6, wherein, Each set of fingerprint images contains a-frame fingerprint images that are continuously acquired, and at least two sets of fingerprint images in the g sets of fingerprint images have overlapping fingerprint images. or, The a-frame fingerprint images in each group of fingerprint images are continuously acquired a-frame fingerprint images, and there is no overlap between any two groups of fingerprint images in the g groups.

8. The method according to claim 6, wherein, The first fingerprint feature extraction process is performed on the g groups of fingerprint images respectively to obtain g first feature vectors, including: For each of the g groups of fingerprint images: The fingerprint images are convolved to obtain the first processing result; The first processing result is processed by i processing layers to obtain the second processing result, where i ≥ 2 and i is an integer, and each of the i processing layers includes a convolutional layer and a linear rectifier layer; The second processing result is downsampled to obtain the third processing result; The third processing result is processed through j processing layers to obtain the fourth processing result, where j≥2 and j is an integer, and each of the j processing layers includes a convolutional layer and a linear rectifier layer. The fourth processing result is downsampled to obtain the first feature vector corresponding to the fingerprint image.

9. The method according to claim 8, wherein, The process of processing the first processing result through i processing layers to obtain the second processing result includes: The first processing result is used as the input data of the first processing layer among the i processing layers, so that the first processing layer processes the first processing result and obtains the output result of the first processing layer; For the b-th processing layer among the i processing layers: the input data of the (b-1)-th processing layer and the output result of the (b-1)-th processing layer are concatenated, and the concatenation result is used as the input data of the b-th processing layer, so that the b-th processing layer processes the input data of the b-th processing layer to obtain the output result of the b-th processing layer, where 2≤b≤i and b is an integer; The output result of the i-th processing layer among the i processing layers is determined as the second processing result; The step of downsampling the second processing result to obtain the third processing result includes: The input data of the i-th processing layer and the second processing result are added together, and the second processing result after the addition is subjected to average pooling to obtain the third processing result.

10. The method according to claim 8, wherein, The process of processing the third processing result through j processing layers to obtain the fourth processing result includes: The third processing result is used as the input data of the first processing layer among the j processing layers, so that the first processing layer processes the third processing result to obtain the output result of the first processing layer; For the c-th processing layer among the j processing layers: the input data and output result of the (c-1)-th processing layer are concatenated, and the concatenation result is used as the input data of the c-th processing layer, so that the c-th processing layer processes the input data of the c-th processing layer to obtain the output result of the c-th processing layer, where 2≤c≤j and c is an integer; The output of the j-th processing layer among the j processing layers is determined as the fourth processing result; The downsampling process performed on the fourth processing result to obtain the first feature vector corresponding to the set of fingerprint images includes: The fourth processing result is subjected to average pooling to obtain the first feature vector corresponding to the fingerprint image.

11. The method according to claim 6, wherein, The step of performing feature fusion processing on the g first feature vectors to obtain a first fused feature vector includes: The g first feature vectors are concatenated to obtain the second feature vector; The second feature vector is subjected to second fingerprint feature extraction processing to obtain the third feature vector; The third feature vector is subjected to depthwise convolution to obtain the first fused feature vector.

12. The method according to claim 11, wherein, The step of performing a second fingerprint feature extraction process on the second feature vector to obtain a third feature vector includes: The second feature vector is processed through k processing layers to obtain the fifth processing result, where k≥2 and k is an integer, and each of the k processing layers includes a convolutional layer and a linear rectified layer. The fifth processing result is downsampled to obtain the sixth processing result; The sixth processing result is processed by h processing layers to obtain the seventh processing result, where h ≥ 2 and h is an integer, and each of the h processing layers includes a convolutional layer and a linear rectifier layer. The seventh processing result is downsampled to obtain the third feature vector.

13. The method according to claim 12, wherein, The process of processing the second feature vector through k processing layers to obtain the fifth processing result includes: The second feature vector is used as the input data of the first processing layer among the k processing layers, so that the first processing layer processes the second feature vector and obtains the output result of the first processing layer; For the d-th processing layer among the k processing layers: the input data and output result of the (d-1)-th processing layer are concatenated, and the concatenation result is used as the input data of the d-th processing layer, so that the d-th processing layer processes the input data of the d-th processing layer to obtain the output result of the d-th processing layer, where 2≤d≤k and d is an integer; The output of the kth processing layer among the k processing layers is determined as the fifth processing result; The step of downsampling the fifth processing result to obtain the sixth processing result includes: The input data of the k-th processing layer and the fifth processing result are added together, and the fifth processing result after the addition is averaged to obtain the sixth processing result.

14. The method according to claim 12, wherein, The process of processing the sixth processing result through h processing layers to obtain the seventh processing result includes: The sixth processing result is used as the input data of the first processing layer among the h processing layers, so that the first processing layer processes the sixth processing result to obtain the output result of the first processing layer; For the e-th processing layer among the h processing layers: the input data of the (e-1)-th processing layer and the output result of the (e-1)-th processing layer are concatenated, and the concatenation result is used as the input data of the e-th processing layer, so that the e-th processing layer processes the input data of the e-th processing layer to obtain the output result of the e-th processing layer, where 2≤e≤h and e is an integer; The output of the h-th processing layer out of the h processing layers is determined as the seventh processing result; The downsampling process performed on the seventh processing result to obtain the third feature vector includes: The seventh processing result is subjected to average pooling to obtain the third feature vector.

15. The method according to claim 5, wherein, The step of extracting fingerprint features from the n frames of fingerprint images and fusing the extracted feature information to obtain fused feature information includes: The n frames of fingerprint images are subjected to a third fingerprint feature extraction process to obtain a fourth feature vector; The fourth feature vector is subjected to depthwise convolution to obtain the second fused feature vector, and the second fused feature vector is determined as the fused feature information.

16. The method according to claim 15, wherein, The step of performing a third fingerprint feature extraction process on the n frames of fingerprint images to obtain a fourth feature vector includes: The n frames of fingerprint images are treated as a group of fingerprint images, and convolution processing is performed on the group of fingerprint images to obtain the eighth processing result; The eighth processing result is processed by m processing networks to obtain the fourth feature vector. The processing network includes t processing layers, a first downsampling processing layer, r processing layers, and a second downsampling processing layer arranged in sequence. Each processing layer includes a convolutional layer and a linear rectified layer. m≥2, t≥2, r≥2, and m, t, and r are all integers.

17. The method according to claim 16, wherein, The process of processing the eighth processing result through m processing networks to obtain the fourth feature vector includes: For the first processing network among the m processing networks: the eighth processing result is processed through t processing layers of the first processing network to obtain the ninth processing result; the ninth processing result is downsampled through the first downsampling processing layer to obtain the first downsampling processing result; the first downsampling processing result is processed through r processing layers of the first processing network to obtain the tenth processing result; the tenth processing result is downsampled through the second downsampling processing layer to obtain the second downsampling processing result; the second downsampling processing result is used as the output result of the first processing network. For the y-th processing network among the m processing networks: the output result of the (y-1)-th processing network is determined as the input data of the first processing layer in the t processing layers of the y-th processing network, and processed by the y-th processing network until the y-th processing network calculates the corresponding second downsampling processing result as the output result of the y-th processing network, where 2≤y≤m and y is an integer; The output of the m-th processing network among the m processing networks is determined as the fourth feature vector.

18. The method according to claim 17, wherein, The step of processing the eighth processing result through t processing layers of the first processing network to obtain the ninth processing result includes: determining the eighth processing result as the input data of the first processing layer in the t processing layers of the first processing network, so that the first processing layer processes the eighth processing result to obtain the output result of the first processing layer; for the f-th processing layer in the t processing layers: concatenating the input data of the (f-1)-th processing layer and the output result of the (f-1)-th processing layer, and using the concatenation result as the input data of the f-th processing layer, so that the f-th processing layer processes the input data of the f-th processing layer to obtain the output result of the f-th processing layer, where 2≤f≤t and f is an integer; and determining the output result of the t-th processing layer in the t processing layers as the ninth processing result. The step of downsampling the ninth processing result through the first downsampling processing layer to obtain the first downsampling processing result includes: adding the input data of the t-th processing layer and the ninth processing result, and then performing average pooling on the ninth processing result after the addition through the first downsampling processing layer to obtain the first downsampling processing result.

19. The method of claim 17, wherein, The step of processing the first downsampling result through r processing layers of the first processing network to obtain the tenth processing result includes: determining the first downsampling result as the input data of the first processing layer among the r processing layers of the first processing network, so that the first processing layer processes the first downsampling result to obtain the output result of the first processing layer; for the q-th processing layer among the r processing layers: concatenating the input data of the (q-1)-th processing layer and the output result of the (q-1)-th processing layer, and using the concatenation result as the input data of the q-th processing layer, so that the q-th processing layer processes the input data of the q-th processing layer to obtain the output result of the q-th processing layer, where 2≤q≤r and q is an integer; and determining the output result of the r-th processing layer among the r processing layers as the tenth processing result. The step of downsampling the tenth processing result through the second downsampling processing layer to obtain the second downsampling processing result includes: performing average pooling on the tenth processing result through the second downsampling processing layer to obtain the second downsampling processing result.

20. The method according to claim 4, wherein determining the authenticity of the fingerprint corresponding to the n frames of fingerprint images based on the fused feature information comprises: Based on the fused feature information, scoring information is determined to determine the authenticity of the fingerprints corresponding to the n frames of fingerprint images; Based on the scoring information and preset threshold conditions, a classification result is determined to indicate the authenticity of the fingerprints corresponding to the n frames of fingerprint images.

21. A fingerprint anti-counterfeiting device, comprising: The acquisition module is used to acquire n frames of fingerprint images collected by the ultrasonic fingerprint sensor, where n≥2 and n is an integer; The determination module is used to determine the authenticity of the fingerprints corresponding to the n frames of fingerprint images through a neural network model.

Images