User authentication method and apparatus using generalized user model

Abstract

To provide a user authentication method and an apparatus using a generalized user model.SOLUTION: A user authentication method comprises: a step for generating a test feature vector corresponding to a test user based on input data input by the test user; a step for determining a first parameter indicating a similarity between a test feature vector and an enrollment feature vector that is pre-enrolled for user authentication; a step for determining a second parameter indicating a similarity between the test feature vector and a user model corresponding to a generalized user; and a step for generating an authentication result for the user based on a first parameter and a second parameter.SELECTED DRAWING: Figure 10

Description

[Technical field]

[0001] The following embodiments relate to a method and apparatus for user authentication using a generalized user model. [Background technology]

[0002] In recent years, there has been much research being conducted on applying efficient pattern recognition methods possessed by humans to actual computers as a solution to the problem of classifying input patterns into specific groups. One such research is on artificial neural networks that model the characteristics of human biological nerve cells using mathematical expressions. To solve the problem of classifying input patterns into specific groups, artificial neural networks use algorithms that mimic the learning ability possessed by humans. This algorithm enables the artificial neural network to generate a mapping between input patterns and output patterns, and the artificial neural network has a learning ability. In addition, the artificial neural network has a generalization ability that can generate a relatively correct output for input patterns that have not been used in learning based on the learned results. Summary of the Invention [Problem to be solved by the invention]

[0003] An object of the embodiments is to provide a method and apparatus for performing user authentication using a generalized user model. [Means for solving the problem]

[0004] According to one embodiment, a user authentication method includes the steps of generating a feature vector corresponding to a user based on input data corresponding to the user, determining a first parameter indicating a similarity between the feature vector and a registered feature vector pre-registered for user authentication, determining a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, and generating an authentication result for the user based on the first parameter and the second parameter.

[0005] The user model may include feature clusters in which generalized feature vectors corresponding to the generalized users are clustered.

[0006] The step of determining the second parameter may include a step of selecting at least a portion of the feature clusters as representative feature clusters based on the feature vector and the similarity between each of the feature clusters, and a step of determining the second parameter based on the similarity between the feature vector and the representative feature cluster.

[0007] Determining the second parameter may include determining the second parameter based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution, the Gaussian distribution may correspond to a distribution of a generalized feature vector indicative of a generalized user. Determining the second parameter may include determining the second parameter using a neural network pre-trained to output a similarity between a network input and the user model.

[0008] Generating the feature vector may include extracting features in the input data using a neural network pre-trained to extract features from input to generate the feature vector.

[0009] Generating an authentication result for the user may include determining a confidence score based on the first parameter and the second parameter, and comparing the confidence score and a threshold to obtain the authentication result for the user. The confidence score may increase as the similarity between the feature vector and the enrollment feature vector increases, and may decrease as the similarity between the feature vector and the user model increases.

[0010] Generating an authentication result for the user may include determining a threshold based on the second parameter, and comparing a confidence score corresponding to the first parameter and the determined threshold to generate the authentication result for the user, the threshold being increased as a similarity between the feature vector and the user model increases.

[0011] The first parameter may increase as the distance between the feature vector and the enrollment feature vector decreases, and the second parameter may increase as the distance between the feature vector and the user model decreases.

[0012] A speaker authentication method according to one embodiment includes the steps of extracting features from speech data corresponding to a user and generating a feature vector corresponding to the user, determining a first parameter indicating a similarity between the feature vector and an enrollment feature vector pre-enrolled for user authentication, determining a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, and generating an authentication result for the user based on the first parameter and the second parameter.

[0013] The enrollment feature vector may be generated based on speech data input by an enrolled user, the feature vector may include information for identifying the user, and the enrollment feature vector may include information for identifying the enrolled user.

[0014] A user authentication device according to one embodiment includes a processor and a memory for storing computer-readable instructions, and when the instructions are executed by the processor, the processor generates a feature vector corresponding to the user based on input data entered by the user, determines a first parameter indicating a similarity between the feature vector and a registered feature vector pre-registered for user authentication, determines a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, and generates an authentication result for the user based on the first parameter and the second parameter.

[0015] A speaker authentication device according to one embodiment includes a processor and a memory for storing computer-readable instructions, and when the instructions are executed by the processor, the processor extracts features from speech data corresponding to a user, generates a feature vector corresponding to the user, determines a first parameter indicating a similarity between the feature vector and an enrollment feature vector pre-enrolled for user authentication, determines a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, and generates an authentication result for the user based on the first parameter and the second parameter.

[0016] An electronic device according to one embodiment includes a sensor for receiving an input from a user, a memory for storing the input, a registered feature vector for user authentication, a user model corresponding to a generalized user, and an instruction, and a processor for executing the instruction, wherein when the instruction is executed by the processor, the processor implements a feature extractor for extracting a feature vector corresponding to the user from the input, implements a first comparator for determining a first parameter indicating a similarity between the feature vector and the registered feature vector, implements a second comparator for determining a second parameter indicating a similarity between the feature vector and the user model, and authenticates the user based on the first parameter and the second parameter. Effect of the Invention

[0017] According to the present invention, a method and apparatus for performing user authentication using a generalized user model can be provided. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 illustrates a user authentication device using a generalized user model according to an embodiment. [Diagram 2] FIG. 13 illustrates adjustment of a threshold according to an embodiment. [Diagram 3] FIG. 13 illustrates an increase in threshold value according to an embodiment. [Figure 4] FIG. 13 illustrates a threshold decrease according to an embodiment. [Diagram 5] FIG. 2 illustrates a clustering-based discrete user model according to one embodiment. [Figure 6] FIG. 2 illustrates a GMM-based continuous user model according to one embodiment. [Figure 7] FIG. 1 illustrates a user model-based neural network according to an embodiment. [Figure 8] 1 is a block diagram showing a user authentication device according to an embodiment; [Figure 9] FIG. 1 is a block diagram illustrating an electronic device according to an embodiment. [Figure 10] 1 is an operational flowchart illustrating a user authentication method according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] 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.

[0020] 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.

[0021] When a component is referred to as being "coupled" or "connected" to another component, it should be understood that although it is directly coupled or connected to the other component, there may be other components in between.

[0022] The terms used in this specification are merely used to describe certain embodiments and are not intended to limit the present invention. A singular expression includes a plural expression unless the context clearly indicates otherwise. In this specification, the terms "include" or "have" 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 presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0023] 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.

[0024] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments will be described in detail with reference to the accompanying drawings, in which the same reference numerals in the various drawings denote the same elements.

[0025] FIG. 1 is a diagram illustrating a user authentication device using a generalized user model according to an embodiment. Referring to FIG. 1, the user authentication device 100 receives input data and processes the received input data to generate output data. The input data corresponds to an input voice or an input image. For example, in the case of speaker authentication, the input data includes voice or audio and is referred to as voice data or speech data. In the case of face authentication, the input data may include a face image, in the case of fingerprint authentication, the input data may include a fingerprint image, and in the case of iris authentication, the input data may include an iris image.

[0026] The input data is input by a test user attempting user authentication. The user authentication device 100 processes the input data to generate an authentication result for the test user. The output data includes the authentication result for the test user generated by the user authentication device 100. For example, the authentication result indicates authentication success or authentication failure.

[0027] The user authentication device 100 performs user authentication using a feature extractor 110 and comparators 120, 130. The feature extractor 110 and the comparators 120, 130 are each realized through at least one hardware module, at least one software module, or a combination of at least one hardware module and at least one software module. As an example, the feature extractor 110 and the comparators 120, 130 are each realized as a neural network. In this case, at least a part of the neural network may be embodied in software, in hardware including a neural processor, or in a combination of software and hardware.

[0028] For example, the neural network is a deep neural network (DNN), which includes a fully connected network, a deep convolutional network, and a recurrent neural network. The DNN includes multiple layers. The multiple layers include an input layer, at least one hidden layer, and an output layer.

[0029] A neural network is trained to perform a given operation by mapping input data and output data that are in a nonlinear relationship to each other based on deep learning. Deep learning is a machine learning method for solving problems given from big data sets. For example, a neural network may perform object classification, object recognition, audio or speech recognition, and video recognition. Deep learning is understood as an optimization problem that aims to reach a point where energy is minimized while training a neural network using prepared training data. Through supervised or unsupervised learning of deep learning, weights corresponding to the structure or model of a neural network are obtained, and input data and output data are mapped to each other through such weights.

[0030] The feature extractor 110 generates a test feature vector corresponding to the test user based on the input data input by the test user. The feature extractor 110 extracts at least one feature in the input data to generate the test feature vector. For example, in the case of speaker authentication, the feature extractor 110 extracts features from the speech data input by the test user to generate a test feature vector corresponding to the test user. The test feature vector includes information for identifying the test user. The feature extractor 110 includes a neural network pre-trained to extract features from the network input.

[0031] The comparator 120 determines a first parameter indicative of the similarity between the test feature vector and the enrollment feature vector, and outputs the determined first parameter. The first parameter increases as the distance between the test feature vector and the enrollment feature vector decreases. According to one embodiment, the comparator 120 includes a neural network that is pre-trained to output the similarity between the network input and the enrollment feature vector.

[0032] The enrollment feature vector is registered in advance for user authentication. For example, the enrollment feature vector is registered by an enrollment user and includes information for identifying the enrollment user. As an example, the enrollment feature vector includes at least one feature extracted from enrollment data input by the enrollment user to enroll the enrollment user. In the case of speaker authentication, features are extracted from the speech data input by the enrollment user, and an enrollment feature vector corresponding to the enrollment user is generated. If it is determined that the test user is the same as the enrollment user, the authentication of the test user is successful, and if it is determined that the test user is different from the enrollment user, the authentication of the test user is unsuccessful.

[0033] The comparator 130 determines a second parameter indicative of a similarity between the test feature vector and a user model corresponding to a generalized user, and outputs the determined second parameter. The second parameter increases as the distance between the test feature vector and the user model decreases. The user model includes data corresponding to a generalized user. The user model includes a generalized feature vector corresponding to the generalized user. The similarity between the test feature vector and the user model is determined based on the distance between the test feature vector and the generalized feature vector.

[0034] For example, the data corresponding to the generalized user may be constructed based at least in part on the data corresponding to the users of the population. The user model may be viewed as a stranger to the test user, i.e., as another user. According to one embodiment, the comparator 130 includes a neural network that is pre-trained to output a similarity between the network input and the user model. The data corresponding to the generalized user may correspond to training data used in the course of training the neural network included in the comparator 130.

[0035] Depending on the embodiment, the user model may be stored as a generalized feature vector, the generalized feature vector may be clustered, a Gaussian mixture model (GMM), or may be reflected in a neural network. Clustering, GMM, and neural networks based on the user model are described in more detail below.

[0036] The user authentication device 100 generates an authentication result for the test user based on the first parameter and the second parameter. The user authentication device 100 determines a confidence score based on the first parameter and the second parameter, and compares the confidence score with a threshold to generate an authentication result for the test user. For example, if the confidence score is higher than the threshold, an authentication result indicating authentication success is generated, if the confidence score is lower than the threshold, an authentication result indicating authentication failure is generated, and if the confidence score is equal to the threshold, it may be determined by policy whether to process the result as authentication success or authentication failure.

[0037] According to the embodiment, not only the registered feature vector but also the user model is taken into consideration during user authentication, and an authentication result suitable for the characteristics of the test user is generated. In other words, variable authentication conditions may be applied to each test user. For example, if the test user has general characteristics, the user model may be close to the test feature vector, and strict authentication conditions may be applied to the test user. Alternatively, if the test user has unique characteristics that are not general, the user model may be far from the test feature vector, and lenient authentication conditions may be applied to the test user.

[0038] User identification is a process of determining who a test user is, and user verification is a process of determining whether a test user is a registered user. In general, a neural network trained from the viewpoint of user identification is used during user authentication, and in this case, characteristics due to user authentication are not reflected in the authentication process. According to an embodiment, during user authentication, the reliability of user authentication is improved by considering the relationship between the test user and other users through a user model.

[0039] FIG. 2 is a diagram illustrating adjustment of authentication conditions according to an embodiment.

[0040] 2, there are illustrated a case 210 where a fixed authentication condition is applied and a case 220 where a variable authentication condition is applied according to a user model. In case 210, the distance between the test feature vector x and the enrollment feature vector y cannot satisfy the authentication condition, and authentication using the test feature vector x fails.

[0041] In case 220, a test feature vector x and a generalized feature vector c corresponding to the user model are 1 or c n It is assumed that the similarity between the test feature vector x and the enrollment feature vector y is low, and therefore the authentication condition is relaxed. Therefore, in case 220, the distance between the test feature vector x and the enrollment feature vector y satisfies the authentication condition, and authentication using the test feature vector x is successful. According to case 220, a legitimate test user that corresponds to an enrollment user can reduce unnecessary authentication attempts.

[0042] The authentication conditions include a confidence score higher than a threshold. A second parameter indicating the similarity between the test feature vector and the user model is reflected in the confidence score or in the threshold. As the second parameter increases, i.e., as the distance between the test feature vector and the user model becomes closer, the confidence score becomes lower and the threshold increases. As the confidence score decreases and the threshold increases, the probability of successful authentication decreases, and therefore the authentication conditions can be considered to be strengthened. Conversely, as the confidence score increases and the threshold decreases, the probability of successful authentication increases, and therefore the authentication conditions can be considered to be relaxed. The confidence score can be expressed as in Equation (1).

[0043]

number

[0044]

number

[0045]

number

[0046] The denominator of formula (3) is Σ n=1 k P(x|θ n ) is the user model θ n When exists, the test user x is a user model θ n The posterior probability corresponding to the user model θ n corresponds to other users other than the test user x. User model θ n is the generalized feature vector c shown in Figure 2. 1 or c n Corresponds to.

[0047] P(x|θ c ) corresponds to the similarity between the test feature vector and the registered feature vector, or corresponds to the distance between the test feature vector and the registered feature vector. For example, (x|θ c ) is determined based on the cosine distance between the test feature vector and the registered feature vector. Distance is a concept corresponding to difference, and similarity and distance are inversely proportional to each other. For example, the higher the similarity between the test feature vector and the registered feature vector, or the closer the distance between the test feature vector and the registered feature vector, the smaller P(x|θ c ) has a large value. Similarly, Σ n=1 k P(x|θn ) corresponds to the similarity or distance between the test feature vector and the generalized feature vector.

[0048] For convenience of explanation, P(x|θ c ) and Σ n=1 k P(x|θ n ) is explained by the concept of similarity. For example, P(x|θ c ) is explained as corresponding to the first parameter indicating the similarity between the test feature vector and the enrollment feature vector, and Σ n=1 k P(x|θ n ) is explained as corresponding to a second parameter indicating the similarity between the test feature vector and the user model.

[0049] Based on equation (3), the authentication condition can be shown in equation (4).

[0050]

number

[0051] According to formula (4), as the second parameter increases, the probability of successful authentication decreases, and therefore the authentication condition can be considered to be strengthened. Also, as the second parameter decreases, the probability of successful authentication increases, and therefore the authentication condition can be considered to be relaxed. According to formula (4), the user authentication device determines a confidence score based on the first parameter and the second parameter, and compares the confidence score with a threshold to generate an authentication result for the test user. Formula (4) can be expressed as formula (5).

[0052]

number

[0053] The right hand side of formula (5) is TH*Σ n=1 k P(x|θ n ) can be defined as a new threshold. According to Equation (5), as the second parameter increases, the probability of successful authentication decreases, and therefore the authentication condition can be considered to be strengthened. Also, as the second parameter decreases, the probability of successful authentication increases, and therefore the authentication condition can be considered to be relaxed. According to Equation (5), the user authentication device can determine a threshold based on the second parameter, and generate the authentication result for the test user by comparing the confidence score corresponding to the first parameter and the determined threshold.

[0054] FIG. 3 is a diagram illustrating the strengthening of authentication conditions according to an embodiment. Referring to FIG. 3, an enrollment feature vector a and generalized feature vectors c1 to c8 are illustrated. The generalized feature vectors c1 to c8 correspond to a user model. Since the similarity between the enrollment feature vector a and the generalized feature vectors c1 to c4 is high, the authentication conditions are strengthened.

[0055] For example, the original threshold TH is adjusted to a new threshold TH×A based on the correction coefficient A. The threshold TH shown in FIG. 3 corresponds to the threshold TH in the above-mentioned formulas (4) and (5), and the correction coefficient A shown in FIG. 3 corresponds to Σ n=1 k P(x|θ n ) For example, the correction coefficient A is a real number greater than 1 that can increase the original threshold value TH. If the correction coefficient A increases the original threshold value TH, the probability of successful authentication decreases, so that the authentication conditions can be considered to be strengthened.

[0056] 4 is a diagram illustrating the relaxation of authentication conditions according to an embodiment. Referring to FIG. 4, an enrollment feature vector b and generalized feature vectors c1 to c8 are illustrated. Since the similarity between the enrollment feature vector b and the adjacent generalized feature vectors c5 to c8 is low, the authentication conditions are relaxed.

[0057] For example, the original threshold TH is adjusted to a new threshold TH×B based on the correction coefficient B. The threshold TH shown in FIG. 4 corresponds to the threshold TH in the above-described formulas (4) and (5), and the correction coefficient B shown in FIG. 4 corresponds to the Σ n=1 k P(x|θ n ) For example, the correction coefficient B is a real number smaller than 1 that can reduce the original threshold value TH. When the correction coefficient B reduces the original threshold value TH, the probability of successful authentication increases, and therefore the authentication conditions can be considered to be relaxed.

[0058] 5 is a diagram illustrating a discrete user model based on clustering according to one embodiment. Referring to FIG 5, an original user model 510 and a clustered user model 520 are illustrated.

[0059] The original user model 510 includes generalized feature vectors c1 to c8. Although eight feature vectors c1 to c8 are illustrated in FIG. 5, the original user model 510 may actually include a greater number of feature vectors. In order to reduce the computational load consumed in processing operations related to such a plurality of feature vectors, the feature vectors c1 to c8 are grouped into a feature cluster θ 1 Or θ 4 are clustered into

[0060] In addition, the feature cluster θ 1 Or θ 4 A part of the representative feature cluster θ 1 Or θ 3 is chosen and used to determine the second parameter.1 Or θ 4 A predetermined number of feature clusters are selected from the representative feature clusters θ in order of their proximity to the test feature vector x. 1 Or θ 3 As an example, the number of feature vectors may be 55,000, the number of feature clusters may be 10, 50, or 100, and the number of representative feature clusters may be 5.

[0061] For example, the user authentication device may receive a test feature vector x and respective feature clusters θ 1 Or θ 4 Based on the similarity between feature clusters,θ 1 Or θ 4 At least a part of the representative feature cluster θ 1 Or θ 3 The user authentication device selects the test feature vector x and the representative feature cluster θ 1 Or θ 3 For example, the user authentication device determines the second parameter based on the similarity between the denominator Σ n=1 k P(x|θ n ) to the user model θ n As a representative feature cluster θ 1 Or θ 3 The user authentication device determines the second parameter by substituting the test feature vector x and the enrollment feature vector θ c Based on this, the numerator of formula (3), P(x|θ c ) can be calculated to determine the confidence score.

[0062] FIG. 6 illustrates a GMM-based continuous user model according to an embodiment. Referring to FIG. 6, an original user model 610 and a Gaussian mixture model (GMM) 620 are illustrated. The GMM 620 uses a Gaussian distribution N(μ n, σ n ), where μ denotes the mean and σ denotes the standard deviation. Figure 6 shows how GMM620 estimates the first Gaussian distribution N(μ 1, σ 1) or the fourth Gaussian distribution N(μ 4, σ 4 ) is illustrated. 1, σ 1 ) or the fourth Gaussian distribution N(μ 4, σ 4 ) corresponds to the distribution of the generalized feature vectors c1 to c8. The user authentication device determines the second parameter based on the test feature vector x and the GMM 620. For example, the user authentication device determines the second parameter Σ n=1 k P(x|θ n ) to determine the second parameter. The user authentication device calculates the test feature vector x and the enrollment feature vector θ c Based on this, the numerator of formula (3), P(x|θ c ) can be calculated to determine the confidence score.

[0063] 7 is a diagram illustrating a user model-based neural network according to an embodiment. Referring to FIG. 7, a user model 710 and a neural network 720 trained based on the user model 710 are illustrated. The neural network 720 is pre-trained to output a similarity between the network input and the user model 710. For example, the network input is pre-trained to output a similarity between the network input and generalized feature vectors c1 to c8 included in the user model 710. In this case, the neural network 720 corresponds to the comparator 130 shown in FIG. 1, and the output of the neural network 720 is the denominator Σ n=1 k P(x|θ n ) Accordingly, the user authentication device can input the test feature vector to the neural network 720 and obtain the second parameter corresponding to the output of the neural network 720.

[0064] FIG. 8 is a block diagram illustrating a user authentication device according to an embodiment. Referring to FIG. 8, the user authentication device 800 receives input data. In the case of speaker authentication, the user authentication device 800 may be referred to as a speaker authentication device. The input data corresponds to an input voice or an input video. The user authentication device 800 processes the input data using a neural network. For example, the user authentication device 800 processes an authentication operation on the input data using a neural network. The database 830 stores an enrollment feature vector and a user model, and the processor 810 can use the enrollment feature vector and the user model stored in the database 830 for user authentication. According to an aspect, the user model may be stored in a clustered state, stored in a GMM, or stored in the neural network 720 shown in FIG. 7.

[0065] The user authentication device 800 may perform one or more operations described or illustrated herein in connection with user authentication, and provide a user authentication result to the user. For example, the result of the user authentication may be provided as audio or video feedback via the input / output module. The user authentication device 800 includes one or more processors 810 and a memory 820. The memory 820 is coupled to the processor 810 and stores instructions executable by the processor 810, data operated by the processor 810, or data processed by the processor 810. The memory 820 may include a non-transitory computer-readable recording medium, such as a high-speed random access memory and / or a non-volatile computer-readable storage medium (e.g., one or more disk storage devices, flash memory devices, or other non-volatile solid-state memory devices).

[0066] The processor 810 executes instructions to perform one or more operations described with reference to Figures 1 to 7. For example, the processor 810 may generate a test feature vector corresponding to the test user based on input data input by the test user, determine a first parameter indicating a similarity between the test feature vector and a registered feature vector previously registered for user authentication, determine a second parameter indicating a similarity between the test feature vector and a user model corresponding to a generalized user, and generate an authentication result for the test user based on the first parameter and the second parameter.

[0067] FIG. 9 is a block diagram illustrating an electronic device according to an embodiment. Referring to FIG. 9, the electronic device 900 receives input data and processes a user authentication operation related to the input data. The electronic device 900 may use the user model described above in the process of processing the user authentication operation. The electronic device 900 may include the user authentication device described with reference to FIGS. 1 to 8 or may perform the function of the user authentication device described with reference to FIGS. 1 to 8.

[0068] The electronic device 900 includes a processor 910, a memory 920, a camera 930, a storage device 940, an input device 950, an output device 960, and a network interface 970. The processor 910, the memory 920, the camera 930, the storage device 940, the input device 950, the output device 960, and the network interface 970 communicate via a communication bus 980.

[0069] The processor 910 executes functions and instructions to be executed within the electronic device 900. For example, the processor 910 processes instructions stored in the memory 920 or the storage device 940. The processor 910 may perform one or more of the operations described with reference to FIGS.

[0070] The memory 920 stores information for performing user authentication. The memory 920 may include a computer-readable storage medium or a computer-readable storage device. The memory 920 stores instructions for execution by the processor 910 and stores relevant information during execution of software or applications by the electronic device 900.

[0071] The camera 930 captures still images, video images, or both. The camera 930 captures a face region that a user inputs to attempt face recognition. The camera 930 provides 3D images that include depth information about objects.

[0072] Storage device 940 includes a computer readable storage medium or a computer readable storage device. Storage device 940 stores a database containing information for processing user authentication, such as enrollment feature vectors and user models. According to one embodiment, storage device 940 stores a larger amount of information than memory 920 and stores information for a longer period of time. For example, storage device 940 includes a magnetic hard disk, an optical disk, a flash memory, a floppy disk, or different forms of non-volatile memory known in the art.

[0073] The input device 950 can receive input from a user through traditional input methods such as through a keyboard and mouse, and newer input methods such as touch input, voice input, and image input. For example, the input device 950 can include a keyboard, a mouse, a touch screen, a microphone, or any other device capable of detecting input from a user and communicating the detected input to the electronic device 900.

[0074] The output device(s) 960 provide output of the electronic device 900 to a user via visual, auditory, or tactile channels. The output device(s) 960 may include, for example, a display, a touch screen, a speaker, a vibration generator, or any other device capable of providing output to a user. The network interface 970 may communicate with external devices via a wired or wireless network.

[0075] FIG. 10 is an operation flowchart showing a user authentication method according to an embodiment. Referring to FIG. 10, the user authentication device generates a test feature vector corresponding to the test user based on input data input by the test user in step S1010, determines a first parameter indicating a similarity between the test feature vector and a registered feature vector registered in advance for user authentication in step S1020, determines a second parameter indicating a similarity between the test feature vector and a user model corresponding to a generalized user in step S1030, and generates an authentication result for the test user based on the first parameter and the second parameter in step S1040. In the case of speaker authentication, the user authentication method may be referred to as a speaker authentication method. In this case, steps S1010 to S1040 are performed by a speaker authentication device, and the speaker authentication device may extract features from the speech data input by the test user in step S1010 and generate a test feature vector corresponding to the test user. In addition, the matters described with reference to FIG. 1 to FIG. 9 may be applied to the user authentication method.

[0076] 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.

[0077] 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.

[0078] 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.

[0079] 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.

[0080] (Appendix 1) generating a feature vector corresponding to a user based on input data corresponding to the user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector that is registered in advance for user authentication; determining a second parameter indicative of a degree of similarity between the feature vector and a user model corresponding to a generalized user; generating an authentication result for the user based on the first parameter and the second parameter; A user authentication method including: (Appendix 2) The user authentication method of claim 1, wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized user are clustered. (Appendix 3) The step of determining the second parameter includes a step of selecting at least a part of the feature clusters as a representative feature cluster based on the feature vector and a similarity between each of the feature clusters; determining the second parameter based on a similarity between the feature vector and the representative feature cluster; 3. The user authentication method of claim 2, comprising: (Appendix 4) The step of determining the second parameter includes: determining the second parameter based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; 4. The user authentication method according to claim 1, wherein the Gaussian distribution corresponds to a distribution of a generalized feature vector representing a generalized user. (Appendix 5) The user authentication method according to any one of appendices 1-3, wherein the step of determining the second parameter includes a step of determining the second parameter using a neural network pre-trained to output a similarity between a network input and the user model. (Appendix 6) The user authentication method according to any one of appendices 1 to 5, wherein the step of generating the feature vector includes a step of extracting features from the input data using a neural network pre-trained to extract features from an input to generate the feature vector. (Appendix 7) The step of generating an authentication result for the user includes: determining a confidence score based on the first parameter and the second parameter; comparing the confidence score and a threshold to obtain the authentication result for the user; 7. A user authentication method according to any one of appendices 1-6, comprising: (Appendix 8) 8. The user authentication method of claim 7, wherein the confidence score increases as the similarity between the feature vector and the enrolled feature vector increases, and decreases as the similarity between the feature vector and the user model increases. (Appendix 9) The step of generating an authentication result for the user includes: determining a threshold value based on the second parameter; comparing a confidence score corresponding to the first parameter and the determined threshold to generate the authentication result for the user; 2. A user authentication method as set forth in any one of appendixes 1-6, including: (Appendix 10) 10. The user authentication method of claim 9, wherein the threshold value increases as the degree of similarity between the feature vector and the user model increases. (Appendix 11) A user authentication method described in any one of Appendices 1-10, wherein the first parameter increases as the distance between the feature vector and the registered feature vector becomes closer, and the second parameter increases as the distance between the feature vector and the user model becomes closer. (Appendix 12) extracting features from speech data corresponding to a user to generate a feature vector corresponding to the user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector that is registered in advance for user authentication; determining a second parameter indicative of a degree of similarity between the feature vector and a user model corresponding to a generalized user; generating an authentication result for the user based on the first parameter and the second parameter; A speaker authentication method comprising: (Appendix 13) The speaker authentication method of claim 12, wherein the enrollment feature vector is generated based on speech data input by an enrollment user. (Appendix 14) 14. The speaker authentication method of claim 13, wherein the feature vector includes information for identifying the user, and the enrollment feature vector includes information for identifying the enrollment user. (Appendix 15) 13. The speaker authentication method of claim 12, wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized user are clustered. (Appendix 16) The step of determining the second parameter includes: determining the second parameters based on the feature vector and a GMM including a Gaussian distribution; 16. The speaker authentication method according to any one of appendices 12-15, wherein the Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of a generalized user. (Appendix 17) 17. The speaker authentication method according to any one of claims 12-16, wherein the step of determining the second parameter includes a step of determining the second parameter using a neural network pre-trained to output a similarity between a network input and the user model. (Appendix 18) The step of generating an authentication result for the user includes: determining a confidence score based on the first parameter and the second parameter; comparing the confidence score and a threshold to obtain the authentication result for the user; 18. A speaker authentication method according to any one of appendices 12-17, comprising: (Appendix 19) The step of generating an authentication result for the user includes: determining a threshold value based on the second parameter; comparing a confidence score corresponding to the first parameter and the determined threshold to generate the authentication result for the user; 19. A speaker authentication method according to any one of appendices 12-18, comprising: (Appendix 20) 19. A program comprising instructions for causing a computer to execute the method according to any one of claims 1 to 19. (Appendix 21) A processor; a memory for storing computer readable instructions; Including, When the instruction is executed by the processor, the processor generating a feature vector corresponding to the user based on input data entered by the user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector that is registered in advance for user authentication; determining a second parameter indicative of a degree of similarity between the feature vector and a user model corresponding to a generalized user; A user authentication device that generates an authentication result for the user based on the first parameter and the second parameter. (Appendix 22) The user authentication device of claim 21, wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized user are clustered. (Appendix 23) The processor, selecting at least a portion of the feature clusters as a representative feature cluster based on the feature vector and a similarity between each of the feature clusters; 23. The user authentication device of claim 22, further comprising: determining the second parameter based on a similarity between the feature vector and the representative feature cluster. (Appendix 24) The processor determines the second parameters based on the feature vector and a GMM including a Gaussian distribution; 24. The user authentication device according to any one of appendices 21-23, wherein the Gaussian distribution corresponds to a distribution of a generalized feature vector representing a generalized user. (Appendix 25) 25. The user authentication device of claim 21, wherein the processor determines the second parameter using a neural network-based comparator pre-trained to output a similarity between a network input and the user model. (Appendix 26) 26. The user authentication device of claim 21, wherein the processor extracts features from the input data using a neural network-based feature extractor pre-trained to extract features from network input to generate the feature vector. (Appendix 27) The processor, determining a confidence score based on the first parameter and the second parameter; 22. The user authentication apparatus of claim 21, further comprising: a step of comparing the confidence score and a threshold to generate the authentication result for the user. (Appendix 28) The processor, determining a threshold value based on the second parameter; 22. The user authentication device of claim 21, further comprising: a confidence score corresponding to the first parameter and the determined threshold to generate the authentication result for the user. (Appendix 29) A processor; a memory for storing computer readable instructions; Including, When the instruction is executed by the processor, the processor Extracting features from speech data corresponding to a user to generate a feature vector corresponding to the user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector that is registered in advance for user authentication; determining a second parameter indicative of a degree of similarity between the feature vector and a user model corresponding to a generalized user; A speaker authentication device that generates an authentication result for the user based on the first parameter and the second parameter. (Appendix 30) 30. The speaker authentication device of claim 29, wherein the enrollment feature vector is generated based on speech data input by an enrollment user. (Appendix 31) A sensor for receiving input from a user; a memory for storing the input, an enrollment feature vector for user authentication, a user model corresponding to a generalized user, and an instruction; a processor for executing the instructions; Including, When the instruction is executed by the processor, the processor providing a feature extractor for extracting from said input a feature vector corresponding to said user; providing a first comparator for determining a first parameter indicative of a similarity between the feature vector and the enrollment feature vector; providing a second comparator for determining a second parameter indicative of a similarity between the feature vector and the user model; authenticating the user based on the first parameter and the second parameter; electronic equipment. (Appendix 32) The processor, determining a confidence score based on the first parameter and the second parameter; 32. The electronic device of claim 31, further comprising: comparing the trust score and a threshold to authenticate the user.

Claims

1. generating a feature vector corresponding to a test user based on input data corresponding to the test user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector previously registered for user authentication, the first parameter corresponding to a probability P(x|θc) that a test user x corresponds to a registered user θc under a condition that the registered user θc exists; determining a second parameter indicating a degree of similarity between the feature vector and a user model corresponding to a generalized user, the second parameter corresponding to a probability Σn=1kP(x|θn) that a test user x corresponds to a user model under a condition that a user model θn corresponding to another user other than the test user x exists; generating an authentication result for the test user depending on whether an authentication condition determined based on (the first parameter) / (the second parameter) and a threshold, or (the first parameter) and a threshold x (the second parameter) is satisfied; Including, The step of determining the second parameter comprises: determining the second parameters based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; The method of user authentication, wherein the Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of generalized users.

2. The method of claim 1 , wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized users are clustered.

3. The step of determining the second parameter includes a step of selecting at least a part of the feature clusters as a representative feature cluster based on the feature vector and a similarity between each of the feature clusters; determining the second parameter based on a similarity between the feature vector and the representative feature cluster; The method of claim 2 , further comprising:

4. The user authentication method of any one of claims 1 to 3, wherein the step of determining the first parameter includes a step of determining the second parameter using a neural network pre-trained to output a similarity between a network input and the user model.

5. The user authentication method of any one of claims 1 to 4, wherein the step of generating the feature vector includes a step of extracting features from the input data using a neural network pre-trained to extract features from an input to generate the feature vector.

6. The step of generating an authentication result for the test user includes: determining a confidence score based on the first parameter and the second parameter; comparing the confidence score and the threshold to obtain the authentication result for the test user; The method of user authentication according to any one of claims 1 to 5, comprising:

7. The user authentication method according to claim 6 , wherein the confidence score increases as the similarity between the feature vector and the enrolled feature vector increases, and decreases as the similarity between the feature vector and the user model increases.

8. The step of generating an authentication result for the test user includes: determining the threshold value based on the second parameter; comparing a confidence score corresponding to the first parameter and the threshold determined in the step of determining the threshold to generate the authentication result for the test user; A method for authenticating a user according to any one of claims 1 to 5, comprising:

9. The method of claim 8 , wherein the threshold value increases as the degree of similarity between the feature vector and the user model increases.

10. A user authentication method described in any one of claims 1 to 9, wherein the first parameter increases as the distance between the feature vector and the registered feature vector becomes closer, and the second parameter increases as the distance between the feature vector and the user model becomes closer.

11. extracting features from speech data corresponding to a test user to generate a feature vector corresponding to the test user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector previously registered for user authentication, the first parameter corresponding to a probability P(x|θc) that a test user x corresponds to a registered user θc under a condition that the registered user θc exists; determining a second parameter indicating a degree of similarity between the feature vector and a user model corresponding to a generalized user, the second parameter corresponding to a probability Σn=1kP(x|θn) that a test user x corresponds to a user model under a condition that a user model θn corresponding to another user other than the test user x exists; generating an authentication result for the test user depending on whether an authentication condition determined based on (the first parameter) / (the second parameter) and a threshold, or (the first parameter) and a threshold x (the second parameter) is satisfied; Including, The step of determining the second parameter includes a step of determining the second parameter based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; A method of speaker authentication, wherein the Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of a generalized user.

12. The speaker authentication method according to claim 11 , wherein the enrollment feature vector is generated based on speech data input by an enrollment user.

13. The method of claim 12 , wherein the feature vector includes information for identifying the test user, and the enrollment feature vector includes information for identifying the enrollment user.

14. The method of claim 11 , wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized users are clustered.

15. The step of determining the second parameter comprises: determining the second parameters based on the feature vector and a GMM including a Gaussian distribution; The method of claim 11, wherein the Gaussian distribution corresponds to a distribution of generalized feature vectors representative of generalized users.

16. The step of generating an authentication result for the test user includes: determining a confidence score based on the first parameter and the second parameter; comparing the confidence score and the threshold to obtain the authentication result for the test user; A method for speaker authentication according to any one of claims 11 to 15, comprising:

17. The step of generating an authentication result for the test user includes: determining the threshold value based on the second parameter; comparing a confidence score corresponding to the first parameter and the threshold determined in the step of determining the threshold to generate the authentication result for the test user; 17. The method of speaker authentication according to any one of claims 11 to 16, comprising:

18. A program comprising instructions for causing a computer to execute the method according to any one of claims 1 to 17.

19. A processor; a memory for storing computer readable instructions; Including, When the instructions are executed by the processor, the processor: generating a feature vector corresponding to a test user based on input data entered by the test user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector previously registered for user authentication, the first parameter corresponding to a probability P(x|θc) that a test user x corresponds to a registered user θc under a condition that the registered user θc exists; Determine a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, the second parameter corresponding to a probability Σn=1kP(x|θn) that the test user x corresponds to the user model under a condition that a user model θn corresponding to another user other than the test user x exists; generating an authentication result for the test user in response to whether an authentication condition determined based on (the first parameter) / (the second parameter) and a threshold, or (the first parameter) and a threshold x (the second parameter) is satisfied; In determining the second parameter, the second parameter is determined based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; The Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of generalized users.

20. 20. The user authentication device of claim 19, wherein the user model includes feature clusters in which generalized feature vectors corresponding to the generalized users are clustered.

21. The processor, selecting at least a portion of the feature clusters as a representative feature cluster based on the feature vector and a similarity between each of the feature clusters; The user authentication device according to claim 19 , further comprising: determining the second parameter based on a similarity between the feature vector and the representative feature cluster.

22. The user authentication device of any one of claims 19 to 21, wherein the processor determines the second parameter using a neural network-based comparator pre-trained to output a similarity between a network input and the user model.

23. The user authentication device of any one of claims 19 to 22, wherein the processor extracts features from the input data using a neural network-based feature extractor pre-trained to extract features from network inputs to generate the feature vector.

24. The processor, determining a confidence score based on the first parameter and the second parameter; The user authentication apparatus of claim 19 , further comprising: comparing the confidence score and the threshold to generate the authentication result for the test user.

25. The processor, determining the threshold value based on the second parameter; 20. The user authentication apparatus of claim 19, further comprising: comparing a confidence score corresponding to the first parameter and the determined threshold value to generate the authentication result for the test user.

26. A processor; a memory for storing computer readable instructions; Including, When the instructions are executed by the processor, the processor: Extracting features from speech data corresponding to a test user to generate a feature vector corresponding to the test user; determining a first parameter indicating a similarity between the feature vector and a registered feature vector previously registered for user authentication, the first parameter corresponding to a probability P(x|θc) that a test user x corresponds to a registered user θc under a condition that the registered user θc exists; Determine a second parameter indicating a similarity between the feature vector and a user model corresponding to a generalized user, the second parameter corresponding to a probability Σn=1kP(x|θn) that the test user x corresponds to the user model under a condition that a user model θn corresponding to another user other than the test user x exists; generating an authentication result for the test user in response to whether an authentication condition determined based on (the first parameter) / (the second parameter) and a threshold, or (the first parameter) and a threshold x (the second parameter) is satisfied; In determining the second parameter, the second parameter is determined based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; The Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of a generalized user.

27. 27. The speaker authentication device according to claim 26, wherein the enrollment feature vector is generated based on speech data input by an enrollment user.

28. a sensor for receiving input from a test user; a memory for storing the input, an enrollment feature vector for user authentication, a user model corresponding to a generalized user, and an instruction; a processor for executing the instructions; Including, When the instructions are executed by the processor, the processor: providing a feature extractor for extracting from said input a feature vector corresponding to said test user; A first comparator is provided for determining a first parameter indicating a similarity between the feature vector and the registered feature vector, the first parameter corresponding to a probability P(x|θc) that a test user x corresponds to a registered user θc under a condition that the registered user θc exists; A second comparator is implemented to determine a second parameter indicating a similarity between the feature vector and the user model, the second parameter corresponding to a probability Σn=1kP(x|θn) that a test user x corresponds to a user model under a condition that a user model θn corresponding to another user other than the test user x exists; authenticating the test user according to whether an authentication condition determined based on (the first parameter) / (the second parameter) and a threshold, or (the first parameter) and a threshold x (the second parameter) is satisfied; In determining the second parameter, the second parameter is determined based on a Gaussian mixture model (GMM) including the feature vector and a Gaussian distribution; The Gaussian distribution corresponds to a distribution of generalized feature vectors indicative of generalized users. electronic equipment.

29. The processor, determining a confidence score based on the first parameter and the second parameter; 30. The electronic device of claim 28, wherein the confidence score and the threshold are compared to authenticate the test user.

Images