Anti-spoofing method and device

Abstract

To provide a spoofing prevention method and apparatus.SOLUTION: A spoofing prevention method according to an embodiment comprises: extracting an input embedding vector from input biometric information; and determining a confidence value of the input embedding vector based on at least two of a real embedding vector of a real model, an enrollment embedding vector of an enrollment model, and a fake embedding vector of a fake model so as to determine whether the input biometric information is forged or not.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The following embodiments relate to an anti-spoofing method and device, and more particularly to a method for detecting whether biometric information is forged. [Background technology]

[0002] Biometric information security authentication technologies are used in smartphones and various mobile / wearable devices. For example, fingerprint recognition technology, which is one of the security authentication technologies, is used due to its advantages such as convenience, security, and economy. In fingerprint recognition, a user's fingerprint image is generally acquired through a sensor, and the acquired fingerprint image is compared with a pre-registered fingerprint image to authenticate the user. Here, if a skillfully manipulated fake fingerprint pattern is input to the sensor, the fingerprint recognition device may not be able to distinguish the fake fingerprint pattern and may inappropriately recognize the fake fingerprint pattern as a biometric fingerprint. For example, if a material such as rubber, silicon, gelatin, epoxy, or latex on which a fingerprint is engraved comes into contact with the sensor, the fingerprint engraved on the material may be recognized as a human fingerprint. In terms of the security of biometric information, it is important to distinguish fake biometric information that is not real biometric information. Summary of the Invention [Problem to be solved by the invention]

[0003] It is an object of the present invention to provide an anti-spoofing method and apparatus. [Means for solving the problem]

[0004] An anti-spoofing method according to one aspect includes the steps of: extracting an input embedding vector from input biometric information; acquiring a forged embedding vector of a forged model that is pre-prepared based on forged biometric information; acquiring at least one of an actual embedding vector of an actual model that is pre-prepared based on actual biometric information and an enrolled embedding vector of an enrolled model that is generated based on biometric information of a user that has been pre-registered; determining a reliability value of the input embedding vector based on the actual embedding vector, at least one of the enrolled embedding vector and the forged embedding vector; and determining whether the input biometric information is forged based on the reliability value.

[0005] The confidence value may be determined based on at least one of the actual embedding vector, the registered embedding vector, and a posterior probability between the forged embedding vector and the input embedding vector.

[0006] The posterior probability may be calculated based on the distance or similarity between the actual embedding vector, the registered embedding vector, and the forged embedding vector and the input embedding vector.

[0007] The step of determining the reliability value may include the steps of: determining a value of a first category corresponding to a case where the input biometric information is not forged based on a first distance between the actual embedding vector and the input embedding vector and a third distance between the enrollment embedding vector and the input embedding vector; determining a value of a second category corresponding to a case where the input biometric information is forged based on a second distance between the forged embedding vector and the input embedding vector; and determining a reliability value of the input embedding vector based on the value of the first category and the value of the second category.

[0008] Determining the value of the first category may include calculating a weighted sum of the smaller of the first distance and the third distance and the third distance.

[0009] Determining the confidence value based on the first category values ​​and the second category values ​​may include determining a confidence value for the input embedding vector by applying a softmax function to the first category values ​​and the second category values.

[0010] The step of determining the confidence value may include the steps of estimating at least one of a first posterior probability between the actual embedding vector and the input embedding vector and a third posterior probability between the registered embedding vector and the input embedding vector, estimating a second posterior probability between the forged embedding vector and the input embedding vector, and determining a confidence value of the input embedding vector based on at least one of the first posterior probability, the third posterior probability, and the second posterior probability.

[0011] determining the confidence value includes calculating a probability ratio between at least one of the first posterior probability and the third posterior probability and the second posterior probability; The step of determining whether the input biometric information is forged can include the step of determining whether the input biometric information is forged based on the probability ratio.

[0012] The step of calculating the probability ratio includes a step of calculating a first probability ratio between the first posterior probability and the second posterior probability, and a step of calculating a second probability ratio between the third posterior probability and the second posterior probability, and the step of determining whether the input biometric information is forged based on the probability ratio includes a step of calculating a weighted sum of the first probability ratio and the second probability ratio.

[0013] According to one aspect, the spoofing prevention method may further include a step of determining, as a representative embedding vector, at least one embedding vector that is closest to the input embedding vector among a plurality of embedding vectors included in the corresponding model, for at least one of the actual embedding vector, the forged embedding vector, and the registered embedding vector.

[0014] The forgery embedding vector may comprise an average of a predetermined number of forgery embedding vectors of the forgery model that are closest to the input embedding vector.

[0015] The step of extracting the input embedding vector includes a step of extracting the input embedding vector from the input biometric information using a neural network trained to extract appropriate features for detecting whether or not the biometric information is counterfeit; The neural network may be trained using the real biometric information and the fake biometric information.

[0016] The forgery model may include a set of forged embedding vectors extracted from forged biometric information, and the real model may include a set of real embedding vectors extracted from forged, unspecified real biometric information.

[0017] According to one aspect, the spoofing prevention device includes a processor that extracts an input embedding vector from input biometric information, obtains a forged embedding vector of a forged model that is pre-prepared based on forged biometric information, obtains at least one of an actual embedding vector of an actual model that is pre-prepared based on actual biometric information and a registered embedding vector of a registered model generated based on biometric information of a user that has been pre-registered, determines a reliability value of the input embedding vector based on the actual embedding vector, at least one of the registered embedding vector and the forged embedding vector, and determines whether the input biometric information is forged based on the reliability value.

[0018] The confidence value may be determined based on the actual embedding vector, the registered embedding vector, and a posteriori probabilities between the forged embedding vector and the input embedding vector.

[0019] The processor may determine a value of a first category corresponding to a case where the input biometric information is not forged based on a first distance between the actual embedding vector and the input embedding vector and a third distance between the enrollment embedding vector and the input embedding vector, determine a value of a second category corresponding to a case where the input biometric information is forged based on a second distance between the forged embedding vector and the input embedding vector, and determine a reliability value of the input embedding vector based on the value of the first category and the value of the second category.

[0020] The processor may estimate at least one of a first posterior probability between the actual embedding vector and the input embedding vector and a third posterior probability between the registered embedding vector and the input embedding vector, estimate a second posterior probability between the forged embedding vector and the input embedding vector, and determine a confidence value of the input embedding vector based on at least one of the first posterior probability, the third posterior probability, and the second posterior probability.

[0021] The processor can calculate a first probability ratio between the first posterior probability and the second posterior probability, calculate a second probability ratio between the second posterior probability and the third posterior probability, and determine whether the input biometric information is forged based on the first probability ratio and the second probability ratio.

[0022] The processor may weightedly sum the first probability ratio and the second probability ratio.

[0023] The processor may determine, as a representative embedding vector, at least one embedding vector that is closest to the input embedding vector among a plurality of embedding vectors included in the corresponding model, for at least one of the actual embedding vector, the forged embedding vector, and the registered embedding vector.

[0024] The processor extracts the input embedding vector from the input biometric information using a neural network trained to extract features suitable for detecting whether or not the biometric information is counterfeit, and the neural network may have been trained using the real biometric information and the counterfeit biometric information.

[0025] The biometric information may include a sensor configured to detect the biometric information.

[0026] The sensor may include a fingerprint sensor, and the biometric information may include a fingerprint.

[0027] According to one aspect, the spoofing prevention method includes the steps of: extracting an input embedding vector from input biometric information; determining a first posterior probability between the input embedding vector and an actual embedding vector of a pre-prepared actual model, a second posterior probability between the input embedding vector and a forged embedding vector of a pre-prepared forgery model, and a third posterior probability between the input embedding vector and an enrolled embedding vector of an enrolled model of an enrolled user; determining a reliability value based on a first probability ratio between the first posterior probability and the second posterior probability, and a second probability ratio between the third posterior probability and the second posterior probability; and determining whether the input biometric information is forged based on the reliability value. Effect of the Invention

[0028] According to the present invention, an anti-spoofing method and apparatus can be provided. [Brief description of the drawings]

[0029] [Figure 1] FIG. 1 is a diagram for explaining a biometric information recognition method according to an embodiment. [Diagram 2] 1 is a flowchart illustrating a method for preventing spoofing according to an embodiment. [Diagram 3] FIG. 13 is a diagram for explaining a method for estimating a posteriori probability according to an embodiment. [Figure 4A] FIG. 13 is a diagram for explaining a method for determining a reliability value of an input embedding vector based on an estimated posterior probability. [Figure 4B] FIG. 13 is a diagram for explaining a method for determining a reliability value of an input embedding vector based on an estimated posterior probability. [Figure 4C] FIG. 13 is a diagram for explaining a method for determining a reliability value of an input embedding vector based on an estimated posterior probability. [Figure 5A] 1 is a diagram illustrating a method for registering biometric information through anti-spoofing according to an embodiment; [Figure 5B] 1 is a diagram illustrating a method for registering biometric information through anti-spoofing according to an embodiment; [Figure 6] FIG. 2 is a block diagram of an anti-spoofing device according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0030] The specific structural or functional description of the embodiment is disclosed for the purpose of illustration only, and may be modified in various forms. Therefore, the embodiment is not limited to the specific disclosed form, and the scope of the present specification includes modifications, equivalents, or alternatives within the technical spirit.

[0031] Although terms such as first or second may be used to describe multiple components, such terms should be construed only for the purpose of distinguishing one component from the other components. For example, a first component may be named a second component, and similarly, a second component may be named a first component.

[0032] The singular expression includes the plural expression unless the context clearly indicates otherwise. In this specification, the terms "comprise" or "have" and the like indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, and should be understood as not precluding the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0033] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. Commonly used predefined terms should be interpreted as having a meaning consistent with the meaning they have in the context of the relevant art, and should not be interpreted as ideal or overly formal unless expressly defined in this specification.

[0034] The embodiments may be realized as various types of products such as personal computers, laptop computers, tablet computers, smartphones, televisions, smart home appliances, intelligent vehicles, kiosks, wearable devices, etc. Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. The same reference numerals in the various drawings denote the same components.

[0035] Fig. 1 is a diagram for explaining a biometric information recognition method according to an embodiment. For convenience of explanation, it is assumed that biometric information is a fingerprint in Fig. 1, but the embodiment described below can be substantially equally applied to various biometric information that can be recognized in the form of an image, such as veins and irises.

[0036] 1, a fingerprint recognition device 100 according to an embodiment includes a fingerprint sensor 110 for detecting a user's fingerprint. The fingerprint recognition device 100 can obtain an input fingerprint image 115 showing the user's fingerprint via the fingerprint sensor 110. The fingerprint recognition device 100 can be one of a personal computer (PC), a laptop computer, a tablet computer, a smartphone, a television (TV), a smart appliance, an intelligent vehicle, a kiosk, and a wearable device.

[0037] Fingerprint registration may be performed as a prerequisite for fingerprint recognition. The registered fingerprint images 121, 122, and 123 may be stored in advance in the registered fingerprint database 120 through a fingerprint registration process. In order to protect personal information, the registered fingerprint database 120 may store features extracted from the registered fingerprint images 121, 122, and 123 instead of storing the registered fingerprint images 121, 122, and 123 as they are. The registered fingerprint database 120 may be stored in a memory included in the fingerprint recognition device 100, or in an external device such as a server capable of communicating with the fingerprint recognition device 100.

[0038] At a later point in time, when the fingerprint recognition device 100 receives the input fingerprint image 115 for authentication, the fingerprint recognition device 100 can recognize the fingerprint of the user by comparing the fingerprint shown in the input fingerprint image 115 (hereinafter referred to as the "input fingerprint") with the registered fingerprints shown in the registered fingerprint images 121 to 123. The fingerprint recognition device 100 can compare the characteristics of the input fingerprint with the characteristics of the registered fingerprints.

[0039] When such authentication is performed by a general fingerprint recognition device, the input fingerprint image 115 detects a fake fingerprint, and if the fingerprint pattern of the input fingerprint image 115 resembles any one of the fingerprint patterns of the registered fingerprint images 121, 122, and 123, authentication of the fake fingerprint may be successful. In order to prevent such a spoofing attack, the fingerprint recognition device 100 according to an embodiment is required to determine whether the input fingerprint shown in the input fingerprint image 115 is a fake fingerprint or a real fingerprint of a person. According to the embodiment, the fingerprint recognition device 100 may include an anti-spoofing device, and may detect whether the input fingerprint is a fake fingerprint through the anti-spoofing device. Hereinafter, the anti-spoofing device may be called a fake fingerprint detection device.

[0040] Conventional counterfeit fingerprint detection systems can distinguish between real fingerprints and counterfeit fingerprints using a neural network trained using a database prepared in advance. Therefore, conventional counterfeit fingerprint detection systems may have a problem of degraded performance in specific situations where the neural network has not been trained. For example, conventional counterfeit fingerprint detection systems use only the dichotomous results of a generically trained network without considering the characteristics of each user, so performance may be degraded due to differences in biometric characteristics of users (e.g., cracked fingerprints, relatively shallow fingerprints, etc.), and since biometric information used for network training is not used during authentication, performance may be degraded in detection situations where it is difficult for the network to train. Furthermore, when a new vulnerability in counterfeit fingerprint detection is discovered, the neural network must be retrained, making it difficult to respond urgently.

[0041] As will be described in detail below, the fake fingerprint detection device according to an embodiment can determine whether an input fingerprint is a fake fingerprint by using a large number of pre-installed real fingerprint features, a large number of pre-installed fake fingerprint features, and / or a registered fingerprint feature of a device user. The fake fingerprint detection device can determine whether a fingerprint is fake by reflecting the biometric features of the user and characteristics such as the assembly tolerance of the device, thereby improving spoofing prevention performance.

[0042] FIG. 2 is a flowchart illustrating a method for preventing spoofing according to an embodiment.

[0043] Referring to Figure 2, steps S210 to S250 may be performed by the anti-spoofing device described above with reference to Figure 1. The anti-spoofing device may be realized by hardware modules, software modules, or various combinations thereof.

[0044] In step S210, the anti-spoofing device extracts an input embedding vector from the input biometric information. The anti-spoofing device can receive input biometric information detected from a user and obtain an input embedding vector corresponding to the input biometric information. The input biometric information may be the input fingerprint image 115 described above with reference to FIG. 1.

[0045] The anti-spoofing device generates an input embedding vector from the input biometric information using a trained neural network to extract features suitable for detecting whether the biometric information is forged. The neural network is trained using an unspecified number of real biometric information and an unspecified number of forged biometric information. The embedding vector generated by the neural network embeds feature information suitable for detecting whether the biometric information is forged, and in this case, the embedding vector may be referred to as a feature vector.

[0046] In step S220, the anti-spoofing device acquires a fake embedding vector of a fake model prepared in advance based on the fake biometric information. The fake model corresponds to a set of fake embedding vectors extracted from an unspecified number of fake biometric information. The unspecified number of fake biometric information may be the same as that used for learning the neural network described above. The fake model may include all of the unspecified number of fake embedding vectors, or may include only the embedding vector that is representative through clustering.

[0047] In step S230, the anti-spoofing device obtains real embedding vectors of a real model that is pre-prepared based on real biometric information and / or enrollment embedding vectors of an enrollment model that is generated based on biometric information of a pre-enrolled user. The real model corresponds to a set of real embedding vectors extracted from an unspecified number of real biometric information. The unspecified number of real biometric information may be the same as that used for learning the neural network described above. The real model may include all of the unspecified number of real embedding vectors, or may include only the embedding vectors represented through clustering.

[0048] The enrollment model refers to a set of enrollment embedding vectors extracted from pre-enrolled user biometric information (e.g., enrollment fingerprint images 121, 122, 123). The anti-spoofing device may store the enrollment model in the enrollment fingerprint database 120 during the fingerprint enrollment process.

[0049] In step S240, the anti-spoofing device determines a reliability value of the input embedding vector based on the real embedding vector, the forged embedding vector, and / or the registered embedding vector. For example, the anti-spoofing device may determine a reliability value of the input embedding vector by using i) the real embedding vector and / or the registered embedding vector together with ii) the forged embedding vector. The reliability value of the input embedding vector is an index of how close the input fingerprint image is to a real fingerprint, and the higher the reliability value, the higher the probability that it is a real fingerprint. A specific method of determining the reliability value will be described in detail below with reference to Figures 3 to 4C.

[0050] In step S250, the spoofing prevention device judges whether the input biometric information is forged or not based on the reliability value. For example, if the reliability value is equal to or greater than a predetermined threshold, the spoofing prevention device judges that the input biometric information is not forged, and if the reliability value is less than the threshold, the spoofing prevention device judges that the input biometric information is forged.

[0051] 3 is a diagram for explaining a method for estimating a posteriori probability according to an embodiment. The explanations of FIG. 1 and FIG. 2 are also applicable to FIG. 3, so that the explanations of the overlapping contents will be omitted.

[0052] 3, the anti-spoofing device according to an embodiment may determine a reliability value of the input embedding vector 310 based on the forged embedding vector of the forgery model 320, the actual embedding vector of the actual model 330, and / or the registered embedding vector of the registered model 340. The anti-spoofing device determines the reliability value of the input embedding vector 310 by considering the forged embedding vector as well as the registered embedding vector and the actual embedding vector of the user, and therefore may determine whether biometric information is forged even in an environment in which biometric characteristics of each user are different and / or in various detection environments.

[0053] The reliability value of the input embedding vector 310 may be determined based on the posterior probability between the forged embedding vector, the real embedding vector, and / or the registered embedding vector and the input embedding vector 310. The posterior probability means the probability that the corresponding event is estimated to have occurred due to a certain cause after the occurrence of the event, for example, the probability of estimating whether the observed input biometric information is real biometric information, registered biometric information, or forged biometric information. The anti-spoofing device may calculate the distance (e.g., Euclidean distance) or similarity (e.g., cosine similarity) between the embedding vectors to estimate the posterior probability.

[0054] The anti-spoofing device may determine a representative vector for estimating the posterior probability with respect to the input embedding vector 310. According to an embodiment, the anti-spoofing device may select a vector in the counterfeit model 320, the actual model 330, and / or the enrollment model 340 that is closest to the input embedding vector 310 as the representative vector. For example, the anti-spoofing device may determine the actual embedding vector 331 that is closest to the input embedding vector 310 as the representative embedding vector of the actual model 330 in the actual model 330. Similarly, the anti-spoofing device may determine the enrollment embedding vector 341 that is closest to the input embedding vector 310 as the representative embedding vector of the enrollment model 340 in the enrollment model 340.

[0055] Alternatively, the spoofing prevention device may determine a predetermined number of embedding vectors that are close to the input embedding vector 310 in the forgery model 320, the actual model 330, and / or the registered model 340 as a representative embedding vector set. For example, the spoofing prevention device may determine a predetermined number of forged embedding vectors 321 that are close to the input embedding vector 310 in the forgery model 320 as a representative embedding vector set of the forgery model 320. Alternatively, the spoofing prevention device may determine an average value of a predetermined number of embedding vectors that are close to the input embedding vector 310 in the forgery model 320, the actual model 330, and / or the registered model 340 as a representative embedding vector. For example, the spoofing prevention device may determine an average value of a predetermined number of forged embedding vectors 321 that are close to the input embedding vector 310 in the forgery model 320 as a representative embedding vector of the forgery model 320. The spoofing prevention device may estimate a posteriori probability between the input embedding vector 310 and the representative embedding vector. For example, in the forgery model 320, the predetermined number of forgery embedding vectors 321 close to the input embedding vector 310 are the predetermined number of forgery embedding vectors closest to the input embedding vector 310 among the forgery embedding vectors of the forgery model 320.

[0056] According to one embodiment, the posterior probability between the input embedding vector 310 and the embedding vector 331, which is the representative embedding vector of the actual model 330, may be expressed as in Equation (1).

number

[0057] The posterior probability between the input embedding vector 310 and an embedding vector 341 which is a representative embedding vector of the enrollment model 340 may be expressed as in Equation (2).

number

[0058] Equation (2) represents the largest value among the posterior probabilities between the registered embedding vector belonging to the registered model 340 and the input embedding vector 310.

[0059] Equation (2) represents the largest value of the posterior probability between the enrollment embedding vector belonging to the enrollment model 340 and the input embedding vector 310.

number

[0060] Equation (3) means the average of the posterior probability between the input embedding vector 310 and N forged embedding vectors (Nbest) 321 that are most similar to the input embedding vector among the forged embedding vectors belonging to the forgery model 320 (where N is a predetermined number). As described above, the posterior probability may be calculated based on the distance or similarity between the embedding vectors. In this case, according to Equation (3), the anti-spoofing device can estimate the posterior probability based on the average distance value between the input embedding vector 310 and the N vectors 321 included in the representative embedding vector set of the forgery model 320.

[0061] According to the embodiment, the method of determining the representative embedding vector is not limited to those expressed in Equations (1) to (3). For example, in the real model 330 and the registered model 340, a predetermined number of embedding vectors having a close distance to the input embedding vector 310 may be determined as a representative embedding vector set, and in the forgery model 320, an embedding vector having a close distance to the input embedding vector 310 may be determined as a representative embedding vector.

[0062] The anti-spoofing device can determine the reliability value of the input embedding vector based on the estimated posterior probability. A method for determining the reliability value of the input embedding vector based on the estimated posterior probability will be described below with reference to Figures 4A to 4C.

[0063] 4A to 4C are diagrams for explaining a method for determining a reliability value of an input embedding vector based on an estimated posterior probability. The explanations shown in Figs. 1 to 3 are also applicable to Figs. 4A to 4C, so the explanation of the overlapping contents will be omitted.

[0064] 4A, the anti-spoofing device can determine a confidence value of the input embedding vector based on the real embedding vector of the real model 420 and the forged embedding vector of the forged model 430. The anti-spoofing device can extract the input embedding vector corresponding to the input biometric information using a neural network 410 (e.g., DNN).

[0065] The anti-spoofing device can estimate a first posterior probability between the representative embedding vector of the real model 420 and the input embedding vector, and can estimate a second posterior probability between the set of representative embedding vectors of the forged model 430 and the input embedding vector.

[0066] The anti-spoofing device may calculate a first probability ratio, which is a probability ratio between the first posterior probability and the second posterior probability, and determine the first probability ratio as the reliability value of the input embedding vector. The first probability ratio may be calculated, for example, as shown in Equation (4).

number

[0067] The denominator of formula (4) corresponds to formula (3), and the numerator of formula (4) corresponds to formula (1).

[0068] Alternatively, the anti-spoofing device can estimate a first posterior probability between the representative embedding vector set of the real model 420 and the input embedding vector, and estimate a second posterior probability between the representative embedding vector set of the counterfeit model 430 and the input embedding vector. In this case, the first probability ratio may be calculated, for example, as shown in Equation (5).

number

[0069] The denominator of Equation (5) corresponds to Equation (3). The numerator of Equation (5) represents the average of the posterior probability between the input embedding vector and the N most similar actual embedding vectors (Nbest) among the actual embedding vectors belonging to the actual model 420.

[0070] The anti-spoofing device may determine that the input biometric information is not forged if a first probability ratio, which is a reliability value of the input embedding vector, is equal to or greater than a predetermined threshold, and may determine that the input biometric information is forged if the first probability ratio is less than the threshold.

[0071] Referring to FIG. 4B, the anti-spoofing device can determine a confidence value for the input embedding vector based on the enrollment embedding vector of the enrollment model 440 and the forgery embedding vector of the forgery model 430.

[0072] The anti-spoofing device can estimate a third posterior probability between the representative embedding vector of the enrollment model 440 and the input embedding vector, and can estimate a second posterior probability between the set of representative embedding vectors of the forgery model 430 and the input embedding vector.

[0073] The anti-spoofing device may calculate a second probability ratio, which is a probability ratio between the second posterior probability and the third posterior probability, and determine the second probability ratio as the reliability value of the input embedding vector. The second probability ratio may be calculated, for example, as shown in Equation (6).

number

[0074] The denominator of formula (4) corresponds to formula (3), and the numerator of formula (4) corresponds to formula (2).

[0075] Alternatively, the anti-spoofing device can estimate a third posterior probability between the representative embedding vector set of the registered model 440 and the input embedding vector, and estimate a second posterior probability between the representative embedding vector set of the counterfeit model 430 and the input embedding vector. In this case, the second probability ratio may be calculated, for example, as shown in Equation (7).

number

[0076] The denominator of Equation (7) corresponds to Equation (3). The numerator of Equation (7) means the average of the posterior probability between the input embedding vector and the N most similar registered embedding vectors (Nbest) among the registered embedding vectors belonging to the registered model 440.

[0077] The anti-spoofing device can determine that the input biometric information is not forged if the second probability ratio, which is a reliability value of the input embedding vector, is greater than or equal to a predetermined threshold, and can determine that the input biometric information is forged if the second probability ratio is less than the threshold.

[0078] The anti-spoofing device can normalize the probability ratio using a logarithm or sigmoid function to adjust the variability of the value or to limit the value range to a certain value. For example, the result of normalizing the probability ratio is shown in Equation (8).

number

[0079] The anti-spoofing device determines that the input biometric information is not forged if the normalized probability ratio, which is the reliability value of the input embedding vector, is equal to or greater than a predetermined threshold, and determines that the input biometric information is forged if the normalized probability ratio is less than the threshold.

[0080] Referring to FIG. 4C, the anti-spoofing device may determine a reliability value of the input embedding vector based on the actual embedding vector of the actual model 420, the forged embedding vector of the forged model 430, and the registered embedding vector of the registered model 440. As described above with reference to FIG. 4A and FIG. 4B, the anti-spoofing device may calculate the first probability ratio and the second probability ratio, and then determine the reliability value of the input embedding vector based on the first probability ratio and the second probability ratio. For example, the anti-spoofing device may determine a result of weighted sum of the first probability ratio and the second probability ratio as the reliability value of the input embedding vector. For example, the result of weighted sum of the first probability ratio and the second probability ratio is as shown in Equation (9).

number

[0081] The spoofing prevention device determines that the input biometric information is not forged if the reliability value of the input embedding vector is equal to or greater than a predetermined threshold, and determines that the input biometric information is forged if the reliability value is less than the threshold. For example, the spoofing prevention device can determine whether the input biometric information is forged by Equation (10).

number

[0082] Although not shown, the anti-spoofing device may determine a value of a first category corresponding to a case where the input biometric information is not forged based on a first distance between the real embedding vector and the input embedding vector and a third distance between the enrollment embedding vector and the input embedding vector, and may determine a value of a second category corresponding to a case where the input biometric information is forged based on a second distance between the forged embedding vector and the input embedding vector.

[0083] The anti-spoofing device may determine a confidence value for the input embedding vector based on the values ​​in the first category and the values ​​in the second category. The anti-spoofing device may calculate the confidence value for the input embedding vector by applying a softmax function to the values ​​in the first category and the values ​​in the second category.

[0084] More specifically, the anti-spoofing device may determine a reliability value of the input embedding vector based on equation (11) and equation (12) (e.g., the anti-spoofing device may determine the result of equation (11) as the reliability value of the input embedding vector).

number

number

[0085] 5A to 5B are diagrams for explaining a method for registering biometric information through spoofing prevention according to an embodiment. The explanations shown in Figs. 1 to 4C are also applicable to Figs. 5A to 5B, so that the explanations of the overlapping contents will be omitted.

[0086] The anti-spoofing device may generate an enrollment model based on enrollment biometric information enrolled for biometric authentication. Here, the enrollment model may be generated on the assumption that all enrollment biometric information is not forged, or the enrollment model may be generated using only enrollment biometric information determined to be not forged using a real model and a forged model.

[0087] 5A, an anti-spoofing device according to an embodiment may generate an enrollment model by assuming that all enrollment biometric information to be enrolled is not forged. The anti-spoofing device may extract an enrollment embedding vector corresponding to the enrollment biometric information using a neural network 510, and may store the corresponding enrollment embedding vector in an enrollment model 520.

[0088] 5B, an anti-spoofing device according to an embodiment can generate an enrollment model using only non-forged and enrolled biometric information using a real model and a forgery model. The anti-spoofing device can extract an enrollment embedding vector corresponding to the enrollment biometric information using a neural network 510, estimate a posterior probability between the real embedding vector and the enrollment embedding vector of the real model 530 and a posterior probability between the forgery embedding vector and the enrollment embedding vector of the forgery model 540, and determine a reliability value of the enrollment embedding vector based on a probability ratio of the estimated posterior probabilities. The anti-spoofing device can store an enrollment embedding vector whose reliability value is equal to or greater than a predetermined threshold in the enrollment model 520.

[0089] FIG. 6 is a block diagram of an anti-spoofing device according to one embodiment.

[0090] 6, an anti-spoofing device 600 according to one embodiment includes a processor 610. The anti-spoofing device 600 further includes a memory 630, a communication interface 650, and a sensor 670. The processor 610, the memory 630, the communication interface 650, and the sensor 670 can communicate via a communication bus 605.

[0091] The processor 610 extracts an input embedding vector corresponding to the input biometric information, obtains a forged embedding vector of a pre-provided forged model, obtains an actual embedding vector of an actual model and / or a registered embedding vector of a registered model, and determines a reliability value of the input embedding vector based on the forged embedding vector and the actual embedding vector and / or the registered embedding vector, thereby determining whether the input biometric information is forged.

[0092] The processor 610 can extract an input embedding vector from the input biometric information using a trained neural network to extract suitable features for detecting whether the biometric information is counterfeit.

[0093] The memory 630 may include a database that stores the enrollment model, the real model, and / or the counterfeit model. The memory 630 may store pre-trained parameters of the neural network. The sensor 670 is a device that detects biometric information of a user, and may include, for example, a fingerprint sensor that detects a fingerprint of a user.

[0094] The processor 610 may estimate posterior probabilities between the real embedding vector, the forged embedding vector, and / or the registered embedding vector and the input embedding vector, and determine a confidence value of the input embedding vector based on the posterior probabilities.

[0095] In addition, the processor 610 may perform the methods or algorithms described above with reference to FIGS. 2 to 5B. The processor 610 may execute a program to control the anti-spoofing device 600. The program code executed by the processor 610 may be stored in the memory 630. The anti-spoofing device 600 may be connected to an external device (e.g., a personal computer or a network) via an input / output device (not shown) to exchange data. The anti-spoofing device 600 may be installed in various computing devices and / or systems, such as a smartphone, a tablet computer, a laptop computer, a desktop computer, a television, a wearable device, a security system, a smart home system, etc.

[0096] The above-described embodiments may be implemented using hardware components, software components, or a combination of hardware and software components. For example, the devices and components described herein may be implemented using one or more general-purpose or special-purpose computers, such as a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), a programmable logic unit (PLU), a microprocessor, or other devices that execute and respond to instructions. The processing device executes an operating system (OS) and one or more software applications that run on the operating system. The processing device also accesses, stores, manipulates, processes, and generates data in response to the execution of the software. For ease of understanding, the processing device may be described as being one in which only one processing device is used, but one skilled in the art will appreciate that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, the processing device may include multiple processors or one processor and one controller. Other processing configurations, such as parallel processors, are also possible.

[0097] The software may include computer programs, codes, instructions, or any combination of one or more thereof, capable of configuring or instructing a processing device to operate as desired, either independently or in combination. The software and / or data may be embodied in any type of machine, component, physical device, virtual device, computer storage medium or device, or transmitted signal wave, either permanently or temporarily, to be interpreted by the processing device or to provide instructions or data to the processing device. The software may be distributed across computer systems coupled to a network, and may be stored and executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

[0098] The method according to the present invention may be embodied in the form of program instructions to be executed by various computer means and recorded on a computer-readable recording medium. The recording medium may include program instructions, data files, data structures, and the like, alone or in combination. The recording medium and the program instructions may be specially designed and constructed for the purposes of the present invention, or may be well known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code, such as generated by a compiler, but also high-level language code executed by a computer using an interpreter, and the like. The hardware devices may be configured to operate as one or more software modules to perform the operations shown in the present invention, and vice versa.

[0099] Although the embodiments have been described above with reference to limited drawings, those skilled in the art may apply various technical modifications and variations based on the above description. For example, the described techniques may be performed in a different order than described, and / or the components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different manner than described, or may be replaced or substituted with other components or equivalents to achieve appropriate results.

[0100] Therefore, the scope of the present invention should not be limited to the disclosed embodiment, but should be determined by the appended claims and their equivalents. [Explanation of symbols]

[0101] 100 fingerprint recognition device 110 Fingerprint sensor 115 Input fingerprint image 120 registered fingerprint database 121 Registered fingerprint image 122 Registered fingerprint image 123 Registered fingerprint image

Claims

1. Extracting an input embedding vector from the input biometric information; Obtaining a forged embedding vector of a forged model based on the forged biometric information; obtaining at least one of an actual embedding vector of an actual model prepared in advance based on actual biometric information and an enrollment embedding vector of an enrollment model generated based on biometric information of a user who has been enrolled in advance; determining a confidence value for the input embedding vector based on at least one of the actual embedding vector and the registered embedding vector, and the forged embedding vector; determining whether the input biometric information is forged based on the reliability value; 2. An anti-spoofing method comprising:

2. The method of claim 1 , wherein the confidence value is determined based on at least one of the actual embedding vector, the registered embedding vector, and a posteriori probability between the forged embedding vector and the input embedding vector.

3. The method of claim 2 , wherein the posterior probability is calculated based on the actual embedding vector, the registered embedding vector, and a distance or similarity between the forged embedding vector and the input embedding vector.

4. The step of determining the confidence value comprises: determining a value of a first category corresponding to a case where the input biometric information is not forged based on a first distance between the actual embedding vector and the input embedding vector and a third distance between the enrollment embedding vector and the input embedding vector; determining a value of a second category corresponding to a case where the input biometric information is forged based on a second distance between the forged embedding vector and the input embedding vector; determining a confidence value for the input embedding vector based on the first category value and the second category value; 2. The method of claim 1, comprising:

5. 5. The method of claim 4, wherein the step of determining the value of the first category includes a step of performing a weighted sum of the smaller of the first distance and the third distance and the third distance.

6. 5. The method of claim 4, wherein determining the confidence value based on the first category value and the second category value comprises determining the confidence value of the input embedding vector by applying the first category value and the second category value to a softmax function.

7. The step of determining the confidence value comprises: estimating at least one of a first posterior probability between the actual embedding vector and the input embedding vector and a third posterior probability between the registered embedding vector and the input embedding vector; estimating a second posterior probability between the forged embedding vector and the input embedding vector; determining a confidence value for the input embedding vector based on at least one of the first posterior probability and the third posterior probability and the second posterior probability; 2. The method of claim 1, comprising:

8. determining the confidence value includes calculating a probability ratio between at least one of the first posterior probability and the third posterior probability and the second posterior probability; The spoofing prevention method according to claim 7 , wherein the step of determining whether the input biometric information is forged includes the step of determining whether the input biometric information is forged based on the probability ratio.

9. The step of calculating the probability ratio comprises: calculating a first probability ratio between the first posterior probability and the second posterior probability; calculating a second probability ratio between the third posterior probability and the second posterior probability; Including, The spoofing prevention method according to claim 8 , wherein the step of determining whether the input biometric information is forged based on the probability ratio includes a step of calculating a weighted sum of the first probability ratio and the second probability ratio.

10. The method of claim 1 , further comprising: determining, as a representative embedding vector, at least one embedding vector that is closest to the input embedding vector among a plurality of embedding vectors included in a corresponding model, for at least one of the actual embedding vector, the forged embedding vector, and the registered embedding vector.

11. The anti-spoofing method of claim 1 , wherein the forged embedding vector comprises an average value of a predetermined number of forged embedding vectors of the forged model that are closest to the input embedding vector.

12. The step of extracting the input embedding vector includes a step of extracting the input embedding vector from the input biometric information using a neural network trained to extract appropriate features for detecting whether or not the biometric information is counterfeit; The method of claim 1 , wherein the neural network is trained using the real biometric information and the forged biometric information.

13. A computer program stored on a medium for performing, in combination with hardware, the method according to any one of claims 1 to 12.

14. the forgery model includes a set of forgery embedding vectors extracted from forged biometric information; The method of claim 1 , wherein the real model includes a set of real embedding vectors extracted from real, non-specified, forged biometric information.

15. An anti-spoofing device comprising a processor that extracts an input embedding vector from input biometric information, obtains a forged embedding vector of a forged model that is pre-prepared based on forged biometric information, obtains at least one of an actual embedding vector of an actual model that is pre-prepared based on actual biometric information and an enrolled embedding vector of a registered model generated based on biometric information of a user that has been pre-registered, determines a reliability value of the input embedding vector based on the actual embedding vector, at least one of the enrolled embedding vector and the forged embedding vector, and determines whether the input biometric information is forged based on the reliability value.

16. The anti-spoofing device of claim 15 , wherein the confidence value is determined based on the actual embedding vector, the enrolled embedding vector, and a posteriori probabilities between the forged embedding vector and the input embedding vector.

17. The processor, determining a value of a first category corresponding to a case where the input biometric information is not forged based on a first distance between the actual embedding vector and the input embedding vector and a third distance between the enrollment embedding vector and the input embedding vector; determining a value of a second category corresponding to a case where the input biometric information is forged based on a second distance between the forged embedding vector and the input embedding vector; 16. The anti-spoofing apparatus of claim 15, further comprising: determining a confidence value for the input embedding vector based on the first category value and the second category value.

18. The processor, 16. The anti-spoofing device of claim 15, further comprising: estimating at least one of a first posterior probability between the actual embedding vector and the input embedding vector and a third posterior probability between the registered embedding vector and the input embedding vector; estimating a second posterior probability between the forged embedding vector and the input embedding vector; and determining a confidence value of the input embedding vector based on at least one of the first posterior probability, the third posterior probability, and the second posterior probability.

19. The processor, 20. The spoofing prevention device of claim 18, further comprising: a first probability ratio between the first posterior probability and the second posterior probability; a second probability ratio between the second posterior probability and the third posterior probability; and a determination as to whether the input biometric information is forged based on the first probability ratio and the second probability ratio.

20. 20. The anti-spoofing apparatus of claim 19, wherein the processor performs a weighted sum of the first probability ratio and the second probability ratio.

21. The spoofing prevention device of claim 15, wherein the processor determines, as a representative embedding vector, at least one embedding vector that is closest to the input embedding vector among a plurality of embedding vectors included in the corresponding model for at least one of the actual embedding vector, the forged embedding vector, and the registered embedding vector.

22. The processor, extracting the input embedding vector from the input biometric information using a neural network trained to extract features suitable for detecting whether the biometric information is counterfeit; 16. The anti-spoofing device of claim 15, wherein the neural network is trained using the real biometric information and the forged biometric information.

23. The anti-spoofing device of claim 15, comprising a sensor configured to detect the biometric information.

24. the sensor includes a fingerprint sensor; 24. The anti-spoofing device of claim 23, wherein the biometric information includes a fingerprint.

25. Extracting an input embedding vector from the input biometric information; determining a first posterior probability between the input embedding vector and an actual embedding vector of a pre-prepared actual model, a second posterior probability between the input embedding vector and a forged embedding vector of a pre-prepared forgery model, and a third posterior probability between the input embedding vector and an enrollment embedding vector of an enrollment model of an enrolled user; determining a confidence value based on a first probability ratio between the first posterior probability and the second posterior probability and a second probability ratio between the third posterior probability and the second posterior probability; determining whether the input biometric information is forged based on the reliability value; 2. An anti-spoofing method comprising:

Images