KR20170055393A - Method and apparatus for adaptively updating registration database for user authentication - Google Patents

Abstract

A method for adaptively updating a registration database includes the steps of: extracting a first feature vector from an input image including a face of a user; determining whether to register the input image in the registration database based on the first feature vector, second feature vectors of the registration images registered in the registration database, and a representative vector representing the second feature vectors; and registering the input image in the registration database based on a determination result. Accordingly, the present invention can improve a verification rate and reduce a false acceptance rate.

Description

[0001] METHOD AND APPARATUS FOR ADAPTIVELY UPDATING REGISTRATION DATABASE FOR USER AUTHENTICATION [0002]

The following embodiments relate to an adaptive update method and apparatus for a registration database for user authentication.

Various mobile devices such as smart phones and wearable devices can use biometric information of the user such as fingerprint, iris, voice, face, and blood vessel for security authentication. In case of facial recognition, there are many change factors according to time such as user's make-up, hairstyle, beard and weight. Therefore, there is a problem in a method of performing authentication by comparing a registered image registered at the beginning with an inputted facial image Can be.

According to one aspect, an adaptive updating method of a registration database includes extracting a first feature vector from an input image including a user's face; Determining whether to register the input image in the registration database based on the first feature vector, second feature vectors of registered images registered in the registration database, and representative vectors representative of the second feature vectors ; And registering the input image in the registration database based on the determination result.

Wherein the step of determining whether the input image is to be registered includes determining whether the input image corresponds to an outlier based on the first feature vector, the second feature vectors, and the representative vector. And determining whether the feature range of the registration database is extended based on the first feature vector and the second feature vectors.

Wherein the step of determining whether the input image corresponds to an outlier comprises: calculating a minimum distance between the first feature vector and the second feature vector; Calculating a representative distance between the first feature vector and the representative vector; And determining whether the input image is an outlier based on the minimum distance and the representative distance.

Wherein the step of determining whether the input image is the outlier is based on whether the minimum distance is smaller than a first threshold preset for updating the registration database and whether the representative distance is smaller than a second predetermined threshold value, And determining whether the outlier is an outlier.

Wherein the step of determining whether the feature range of the registration database is extended comprises calculating, in a vector set including the first feature vector and the second feature vectors, a cumulative feature distance corresponding to each vector, Determined based on distances between one of the vectors and the remaining vectors; And determining whether the cumulative feature distance corresponding to the first feature vector is greater than at least one of cumulative feature distances corresponding to the second feature vectors.

Wherein the step of registering in the registration database further includes the step of, when a cumulative feature distance corresponding to the first feature vector is larger than at least one of cumulative feature distances corresponding to the second feature vectors, And replacing the input image with the input image.

The step of replacing any one of the registered images with the input image may include the step of replacing the registered image corresponding to the second feature vector having the minimum cumulative feature distance among the cumulative feature distances with the input image.

The step of determining whether or not to register may include comparing the number of registered images registered in the registration database with the maximum number of registered registered databases.

Wherein the step of registering the input image in the registration database comprises: adding the input image to the registration database when the number of registered images is smaller than the maximum number of registered images; And replacing any one of registered images registered in the registration database with the input image if the number of registered images is equal to the maximum registered number.

The adding to the registration database may include adding the input image to the registration database if the input image is not an outlier and the number of registered images is smaller than the maximum registered number.

The step of determining whether to register may include authenticating the user based on the first feature vector, the second feature vectors, and the representative vector.

Wherein authenticating the user comprises: calculating a minimum distance between the first feature vector and the second feature vector; Calculating a representative distance between the first feature vector and the representative vector; And authenticating the user based on the comparison between the minimum distance and a minimum value of the representative distance and a preset threshold value for user authentication.

The step of determining whether or not to register may include determining whether to register the input image in the registration database, based on whether the user authentication is successful.

The adaptive updating method of the registration database may further include updating the representative vector by the first feature vector when the input image is registered in the registration database.

According to one aspect, an adaptive updating device of a registration database comprises: a memory for storing a registration database; Extracting a first feature vector from an input image including a face of a user and extracting a first feature vector of the registered images registered in the registration database and a representative vector representing the second feature vectors, And determining whether to register the input image in the registration database, and registering the input image.

Wherein the processor is configured to determine whether the input image corresponds to an outlier based on the first feature vector, the second feature vectors, and the representative vector or based on the first feature vector and the second feature vectors And whether or not the feature range of the registration database is extended.

Wherein the processor calculates a minimum distance between the first feature vector and the second feature vectors and a representative distance between the first feature vector and the representative vector, and based on the minimum distance and the representative distance, It is possible to judge whether the image is an outlier.

Wherein the processor is configured to calculate, in a set of vectors comprising the first feature vector and the second feature vectors, a cumulative feature distance corresponding to each vector, the cumulative feature distance being a distance between one vector in the vector set and the remaining vectors - determining a feature range of the registration database based on whether the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors Can be expanded.

Wherein when the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors as a result of the determination, The registered image corresponding to the second feature vector can be replaced with the input image.

The processor may determine whether to add or replace the input image to the registration database based on a comparison result between the number of registered images registered in the registration database and the maximum number of registered databases.

Wherein the processor authenticates the user based on the first feature vector, the second feature vectors, and the representative vector, and registers the input image in the registration database based on whether the user authentication is successful Or not.

The processor may update the representative vector by the first feature vector when the input image is registered in the registration database.

An adaptive updating method of a registration database according to one side includes: Determining whether the input image is an outlier; Determining whether the feature range of the registration database is expanded due to the input image; And replacing the input image with any one of the registered images of the registration database.

Wherein the step of determining whether the input image is an outlier comprises: calculating a minimum distance between a first feature vector extracted from the input image and second feature vectors extracted from registered images of the registration database; Calculating a representative distance between the first feature vector and a representative vector representative of the second feature vectors; And determining a first condition for the minimum distance and a second condition for the representative distance are satisfied.

Wherein the step of determining whether the feature range of the registration database is expanded due to the input image includes a step of extracting a first feature vector extracted from the input image and a second feature vector extracted from the registered images of the registration database Calculating a cumulative feature distance corresponding to each vector; And determining whether the cumulative feature distance corresponding to the first feature vector is greater than at least one of cumulative feature distances corresponding to the second feature vectors.

The step of calculating a cumulative feature distance corresponding to each of the vectors may comprise summing the corresponding vectors and the distances between the remaining vectors in the vector set corresponding to the respective vectors.

The adaptive updating apparatus of the registration database according to one side authenticates an input image based on a registration database and determines whether the input image is an outlier based on a representative vector representing registered images of the registration database, And a processor for adaptively updating the registration database based on the authentication result and the determination result.

Wherein the processor calculates a minimum distance between a first feature vector extracted from the input image and second feature vectors extracted from registered images of the registration database to determine whether the input image is an outlier, 1 feature vector and the representative vector to determine whether a first condition for the minimum distance and a second condition for the representative distance are satisfied.

1 is a flow diagram illustrating an adaptive update method of a registration database according to an embodiment;
FIG. 2 is a diagram illustrating a principle of improving the identity verification rate by adaptive updating of a registration database according to an embodiment; FIG.
FIG. 3 illustrates a method for determining whether an input image corresponds to an outlier according to an exemplary embodiment. FIG.
4 is a flow diagram illustrating an adaptive update algorithm of a registration database according to an embodiment;
5 is a diagram illustrating a method for determining a threshold according to an embodiment;
6 is a diagram illustrating a method of determining whether a feature range of a registration database is expanded by an input image according to an exemplary embodiment.
FIG. 7 illustrates registered images adaptively updated in a registration database according to an embodiment; FIG.
8 is a diagram showing feature distances between registered images initially registered in a registration database and feature distances between finally registered registered images;
9 is a block diagram of an adaptive update apparatus of a registration database according to an embodiment;

The specific structural or functional descriptions disclosed herein are merely illustrative for purposes of illustrating embodiments in accordance with the technical concepts. The disclosed embodiments may be modified and implemented in various other forms, and the scope of the present disclosure is not limited to the disclosed embodiments.

The terms first or second may be used to describe various components, but such terms should be interpreted solely for the purpose of distinguishing one component from another. For example, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

It is to be understood that when an element is referred to as being "connected" to another element, it may be directly connected or connected to the other element, although other elements may be present in between.

The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises ", or" having ", and the like, are used to specify one or more of the described features, numbers, steps, operations, elements, But do not preclude the presence or addition of steps, operations, elements, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the meaning of the context in the relevant art and, unless explicitly defined herein, are to be interpreted as ideal or overly formal Do not.

Embodiments to be described below can be used to recognize a user's face. The act of recognizing the user's face may include an act of authenticating or identifying the user. The act of authenticating the user may include, for example, determining whether the user is an already registered user. In this case, the result of the operation of authenticating the user may be output as true or false. The operation of identifying a user may include, for example, an operation of determining which user of the plurality of users the user is registered with. In this case, the result of the operation of identifying the user may be output as the ID of any one of the pre-registered users. If the user does not correspond to any of the plurality of pre-registered users, a signal indicating that the user is not identified may be output.

Embodiments may be implemented in various forms of products, such as personal computers, laptop computers, tablet computers, smart phones, televisions, smart home appliances, intelligent cars, kiosks, wearable devices, For example, embodiments may be applied to authenticating users in smart phones, mobile devices, smart home systems, and the like. Embodiments can be applied to a payment service through user authentication. In addition, the embodiments can be applied to an intelligent automobile system that automatically authenticates a user and starts up. Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Like reference symbols in the drawings denote like elements.

1 is a flowchart illustrating an adaptive update method of a registration database according to an exemplary embodiment of the present invention. The update device that performs the adaptive update method may be included in the user authentication device, for example, or may be configured as a separate device. The updating device may be implemented as a software module, a hardware module, or various combinations thereof. The registration database may be referred to as an enrollment template.

Referring to FIG. 1, the updating apparatus extracts a first feature vector from an input image including a face of a user in step 110. The 'first feature vector' can be understood as a vector representing features for face recognition extracted from the input image. The updating device can extract a first feature vector from an input image using various techniques such as a Local Binary Pattern (LBP) technique, a Gabor technique, or a Deep Learning technique.

The update apparatus determines in step 120 whether to register the input image in the registration database. For example, the updating apparatus may perform authentication of an input image and determine whether to register the input image in the registration database by determining whether the input image corresponds to an outlier.

The updating device may use the first feature vector, the second feature vectors, and the representative vector. The 'second feature vector (s)' can be understood as vector (s) representing characteristics for face recognition extracted from the registered image (s) registered in the registration database. The second feature vector (s) may also be extracted by an LBP technique, a Gabor technique, or a deep learning technique.

)는 예를 들어, 아래의 <수학식 1>과 같이 표현될 수 있다. The 'representative vector' is a vector representing the second feature vectors, and can be calculated through a statistical calculation (for example, average, standard deviation, etc.) of the second feature vectors. Registered image with the second feature is extracted from _{(X 1, X 2, ...} , X n) of the vector (x _i) and the representative vector (

Can be expressed, for example, as Equation (1) below.

Here, n is the number of registered images registered in the registration database.

The updating device may perform authentication of the input image based on the first feature vector, the second feature vectors, and the representative vector. In one example, the updating device can perform authentication of the input image by comparing the first feature vector with the vectors in the set of {the second feature vectors, representative vector}.

The updating device can calculate the minimum distance between the first feature vector and the vectors in the set of {the second feature vectors, representative vector}. The distance between the first feature vector and the vectors in the set of {second feature vectors, representative vectors} can be understood as an Euclidean distance inversely proportional to, for example, the similarity between feature vectors. The updating apparatus can calculate the minimum distance d _n using Equation (2).

는 대표 벡터이고, d _n 은 {제2 특징 벡터들, 대표 벡터}의 집합 내 벡터들과 제1 특징 벡터 사이의 최소 거리일 수 있다.Here, y is a first feature vector, x ₁ to x _n are second feature vectors,

And d _n may be the minimum distance between the vectors in the set of {the second feature vectors, the representative vector} and the first feature vector.

When the minimum distance d _n is smaller than the first threshold distance, the updating device can determine that the authentication of the input image is successful. The first threshold distance may be predetermined to a distance corresponding to a false acceptance rate (FAR) of 1%. The false recognition rate (FAR) is the rate of wrong recognition of others.

The updating device can use the similarity instead of the distance. In this case, the updating apparatus can calculate the maximum similarity between the first feature vector and the vectors in the set of {the second feature vectors, the representative vector}. When the maximum similarity degree is larger than the first threshold similarity degree, the updating apparatus can determine that the input image has been successfully authenticated. The first threshold similarity may be predetermined by a degree of similarity corresponding to a false recognition rate (FAR) of 1%.

In another example, the updating device may determine whether the authentication of the input image is successful by comparing the first feature vector with the second feature vector.

Even if the authentication of the input image is successful, the input image may actually be false acceptance. For example, in the authentication process, it is determined that the input image corresponds to the user, but in reality, the input image may be an image corresponding to another person. The updating device can determine whether to register the input image in the registration database through a second criterion that is stricter than the first criterion for authentication in order to prevent the misleading input image from being registered in the registration database. The second criterion may be a criterion for determining whether the input image corresponds to an outlier. Here, 'outliers' can be understood as images corresponding to others who succeeded in authentication but actually are not themselves.

The updating device can determine whether the input image corresponds to an outlier based on the first feature vector, the second feature vectors, and the representative vector. For example, the updating device can check two conditions. First, the updating device can check whether the minimum distance among the distances between the first feature vector and the second feature vectors is smaller than a predetermined second threshold distance. Here, the second threshold distance may be determined more strictly than the first threshold distance for authentication of the input image. For example, the second threshold distance may be a distance corresponding to a false recognition rate (FAR) of 0.01%. Second, the updating device can check whether the distance between the first feature vector and the representative vector is smaller than a predetermined third threshold distance. The third threshold distance may be determined to be equal to or different from the second threshold distance.

The updating device can use the similarity instead of the distance. In this case, the updating apparatus can check whether the maximum similarity degree between the first feature vector and the second feature vector is larger than the predetermined second criticality similarity. The second threshold similarity may be, for example, a similarity corresponding to a false recognition rate (FAR) of 0.01%. Further, the updating apparatus can check whether or not the similarity degree between the first feature vector and the representative vector is larger than a predetermined third criticality similarity.

The update apparatus can determine that the input image does not correspond to an outlier when both of the above two conditions are satisfied. If the input image does not correspond to the outlier, the update device can determine that the input image is registered in the registration database. On the other hand, when the input image corresponds to an outlier, the update device can determine that the input image is not registered in the registration database.

In response to the determination that the input image is to be registered in the registration database, the updating apparatus registers the input image in the registration database in step 130. Here, 'registration' may be understood to include adding an input image to the registration database and replacing any of the registered images in the registration database with the input image.

For example, when the number of registered images included in the registration database is smaller than a predetermined threshold number (for example, 10), the updating apparatus can add the input image to the registration database.

As another example, if the number of registered images included in the registration database is equal to the threshold number, the updating apparatus can replace any one of the registered images with the input image depending on whether the feature range of the registration database is expanded due to the input image . Here, 'the feature range of the registration database is expanded' can be understood as meaning that the user's face changes that can be recognized by the registered images included in the registration database are varied. The user's facial changes can be attributed to various factors such as makeup, hairstyle, beard, and weight.

The updating device can determine whether the feature range of the registration database is expanded based on the first feature vector and the second feature vectors. A method for determining whether or not the feature range of the registration database is expanded will be described with reference to Fig.

The update apparatus according to the embodiments can adaptively update the registration database so that various images corresponding to the user are included in the registration database while preventing the image corresponding to the other person from being erroneously included in the registration database. Through the adaptively updated registration database, the identity verification rate can be improved.

2 is a diagram for explaining the principle of improving the identity verification rate by adaptive update of the registration database according to an embodiment. Referring to FIG. 2, as registration images of the registration database are adaptively added or replaced, the authentication range by the registered images of the registration database can be improved to effectively cover the range corresponding to the user. In FIG. 2, a circle drawn radially from each registered image represents an authentication range based on the registered image.

Referring to FIG. 2A, an authentication range based on the registered image X ₁ when one registered image X ₁ is registered in the registered database is shown. As described above, since the face has many changing elements such as illumination, makeup, hair style, beard, weight, and the like, it is difficult to effectively cover the range corresponding to the user with one registered image. In this case, the verification rate (VR) may be low and the false rejection rate (FRR) may be large. Recognition rate (VR) is the ratio of recognizing the user correctly, and false rejection rate (FRR) is the rate of rejecting the user wrongly. Although not illustrated in the figure, the registration image to cover the range corresponding to the user only by (X _1), to be extended to the scope of the registered image (X ₁₎ to the extent that misrecognized others for identity. In this case, the false recognition rate (FAR) can be seriously increased.

Referring to FIG. 2B _, an authentication range when a plurality of registered images X _1, X _2, X _3, X _{4, and} X ₅ are added to the registration database is shown. When a plurality of registered images X _2, X _3, X _{4, and} X ₅ are added to the registration database, the recognition rate VR can be improved as compared with the case where only one registered image X ₁ is registered.

In one embodiment, the recognition rate (VR) is increased by using the features (e.g., second feature vectors) of the plurality of registered images and the average feature (e.g., representative vector) of the registered images together, By setting the authentication range of the image narrower, the false recognition rate (FAR) can be lowered.

Referring to FIG. 2C _, the authentication range is shown when the registration database is replaced with a plurality of registered images (X ' _{1 ,} X' _{2 ,} X ' _3, X' _{4 ,} X ' ₅ ). The replaced multiple registered images of _{(X '1, X' 2} , X '3, X' 4, X '5) In this case, the possible range for the I widely herpes, the recognition rate of the face variously changing user (VR) can be improved.

In one embodiment, the registration images are continuously changed so that the distance between the registered images stored in the registration database is kept as wide as possible, so that the recognition rate (VR) can be increased to adapt to the user's face change. Here, 'maintaining the distance between the registered images as wide as possible' can be understood as meaning that the degree of similarity between the registered images is minimized so that the registered images exhibit various changes as much as possible. At this time, even if the distance between the registered images is kept as wide as possible, the outliers should not be included in the registered image. A method for preventing an outlier from being included in a registered image will be described with reference to FIG.

In one embodiment, the range corresponding to the user is effectively covered by adding and replacing the registered images as shown in (b) and (c) of FIG. 2, thereby increasing the recognition rate (VR) And the false rejection rate (FRR).

)를 포함한다고 하자. 대표 벡터는 등록 영상들의 대표 벡터(예를 들어, 평균 특징)에 해당할 수 있다.FIG. 3 illustrates a method for determining whether an input image corresponds to an outlier according to an exemplary embodiment of the present invention. Referring to FIG. 3 (a) shows a case where the input image is determined as an outlier, and FIG. 3 (b) shows a case where it is determined that the input image is not an outlier. At this time, the registration database stores the second feature vectors (x _1, x _2, x _3, x _4, x ₅ ) of the registered images and the representative vector

). The representative vector may correspond to a representative vector (e.g., average feature) of the registered images.

) 사이의 거리는 미리 정해진 제3 임계 거리보다 클 수 있다. 이 경우, 제1 입력 영상은 제4 등록 영상과 유사하지만, 등록 데이터베이스에 저장된 등록 영상들을 대표하는 영상과는 유사하지 않다고 판단될 수 있다. 3A, the distance between the first feature vector y ₁ of the first input image and the feature vector x ₄ of the fourth registered image stored in the registration database is smaller than a predetermined second threshold distance , A first feature vector (y ₁ ) and a representative vector of registered images (

May be greater than a predetermined third critical distance. In this case, it can be determined that the first input image is similar to the fourth registered image but is not similar to the representative image stored in the registration database.

Even if it is determined that the input image is similar to one or more of the registered images stored in the registration database, if the input image is judged not to be similar to the image representing the registered images, . As a result, the embodiments can prevent the outlier from being included in the registration database.

) 사이의 거리 또한 제3 임계 거리보다 작을 수 있다. 이는 제2 입력 영상이 등록 영상들 모두와 일정 수준 이상의 높은 유사도를 가짐을 나타낸다. 갱신 장치는 제1 특징 벡터(y₂)에 대응되는 제2 입력 영상을 본인에 해당하는 영상으로 판단하고, 제2 입력 영상을 등록 데이터베이스에 등록시킬 수 있다.3B, the distance between the first feature vector y ₂ of the second input image and the feature vector x 4 of the fourth registered image stored in the registration database is smaller than the second threshold distance, 1 feature vector (y ₂ ) and representative vector (

May also be less than the third critical distance. This indicates that the second input image has a high degree of similarity with all the registered images above a predetermined level. The updating device can determine the second input image corresponding to the first feature vector y _{2 as an} image corresponding to the user and register the second input image in the registration database.

4 is a flowchart illustrating an adaptive update algorithm of a registration database according to an embodiment. Referring to FIG. 4, the update device may receive the input image Y at step 410. The updating device can extract the first feature vector from the input image (Y). The update device may determine in step 420 whether the user authentication is successful. For example, the updating device may perform user authentication using the first feature vector and the registration database 405. [ If the user authentication fails, the update apparatus can end the operation without registering the input image Y in the registration database 405. [

If the user authentication is successful, the update device can determine in step 430 whether the input image Y is an outlier. The updating device can determine whether the input image Y is an outlier by checking the two conditions described above with reference to FIG.

The condition for determining whether the input image corresponds to an outlier can be variously modified. For example, the updating apparatus can determine whether the input image is an outlier according to Equation (3) below.

는 대표 벡터이다. d(x _i , y)는 제1 특징 벡터와 i번째 제2 특징 벡터 사이의 거리이고, d(

, y)는 제1 특징 벡터와 대표 벡터 사이의 거리이다.Here, y is a first feature vector, x _i is an i-th second feature vector, N is a number of second feature vectors,

Is a representative vector. d ( x _i , y ) is the distance between the first feature vector and the i th second feature vector, d (

, y ) is the distance between the first feature vector and the representative vector.

According to Equation ( ₃ ), the number of second feature vectors whose distance from the first feature vector among the N feature vectors is smaller than the threshold value (T ₃ ) is 2 or more, and the distance between the representative vector and the first feature vector Is smaller than the threshold value (T ₃ ), the first feature vector may be judged to be a registration update candidate rather than an outlier.

If it is determined that the input image Y is an outlier, the update apparatus can end the operation without registering the input image Y in the registration database 405. [ If it is determined that the input image Y is not an outlier, the update device can compare the number n of registered images registered in the registration database with the maximum number N of registered databases 405 in step 440 . The maximum number of registrations N may be a predetermined value (e.g., 10).

If the number of registered images is smaller than the maximum registered number, the updating device may add the input image Y to the registration database 405 in step 445. [ When the input image Y is added to the registration database 405, the representative vector representing the registration database 405 can be updated. For example, when the input image Y is added to the registration database 405, the representative vector can be updated as shown in Equation (4) below.

If the number of registered images is equal to or greater than the maximum number of registered images, the updating device can determine whether the feature range of the registration database 405 is expanded by the input image Y in step 450. A method for determining whether or not the feature range of the registration database 405 is expanded by the input image Y will be described with reference to FIG. If it is determined that the feature range of the registration database 405 is not expanded by the input image Y, the updating apparatus can end the operation without registering the input image Y in the registration database 405. [

If it is determined that the feature range of the registration database 405 is extended by the input image Y, the updating device updates the input image Y to any one of the registered images registered in the registration database 405 Can be replaced.

For example, the update apparatus can determine whether the input image Y corresponds to the registration update candidate of the registration database by Equation (5).

More specifically, the updating device may set the first feature vector y to the ( _{N + 1} ) th second feature vector x _{N + 1} . The updating device may calculate a cumulative feature distance S _i corresponding to each of the (N + 1) second feature vectors. When the index (i _min ) of the minimum cumulative feature distance among the cumulative feature distances is equal to or less than N, the updating apparatus replaces the i _min th image with the input image. In other words, if the input image has the minimum cumulative feature distance, the update device does not proceed to replace the update image, and if the existing registered image has the minimum cumulative feature distance, the update device can replace the registered image with the input image.

If the input image Y is replaced with any of the registered images in the registration database 405, the representative vector representing the registration database 405 can be updated.

If the input image Y is not an outlier and the input image Y can broaden the feature range (or the authentication range) of the registered images, the input image Y may be replaced with the existing registered image So that diversity of the registered images included in the registration database 405 can be ensured.

Step 410 corresponds to step 110 of Figure 1 and steps 420 and 430 correspond to step 120 of Figure 1 and steps 440, 445, 450, ), And step 455 may correspond to step 130 in FIG. 1, so that a more detailed description is omitted.

5 is a diagram illustrating a method for determining a threshold according to an embodiment. The threshold value may include a threshold distance or threshold similarity for user authentication, a critical distance or critical similarity for outlier determination, and the like. The threshold can be determined based on various performance indicators. For example, the threshold may be determined based on a recognition rate (VR), a false recognition rate (FAR), a false rejection rate (FRR), or various combinations thereof.

5, there is shown a normal distribution curve 510 relating to the feature distance between the user's own images and a normal distribution curve 530 relating to the feature distance between the tine images. Here, the feature distance can be understood as a distance inversely proportional to the similarity between images. In the graph of FIG. 5, the x-axis represents the characteristic distance between two images, and the y-axis represents the value of the probability density function corresponding to the characteristic distance.

For example, in order to set the first threshold distance for user authentication to a distance corresponding to a false recognition rate (FAR) of 1%, the updating device calculates the total area of the normal distribution curve 530 The characteristic distance of the boundary line 550 dividing the area of the lower 1% of the first threshold distance can be set as the first threshold distance.

FIG. 6 is a diagram for explaining a method of determining whether the feature range of the registration database is expanded by an input image according to an exemplary embodiment. Referring to FIG. 6A, when the input image is received, the first feature vector y of the input image and the second feature vectors x _1, x _2, x _3, x _4, x _{5 are} shown.

The updating device determines the cumulative feature distance S corresponding to each vector as shown in FIG. 6 (b) by the sum of the distances between any one of the vectors and the remaining vectors in the vector set of FIG. 6 (a) . The updating device calculates the distance between the second feature vector x ₁ and the first feature vector y and the distance between the second feature vector x ₁ and the remaining second feature vectors x _{2 ,} x _{4, and} x ₅ , respectively, to determine the cumulative feature distance S ₁ corresponding to the second feature vector x ₁ . Updating apparatus in this way, the other second feature vectors _{(x 2, x 3, x} 4, x 5) Cumulative characterized distances corresponding to the respective _{(S 2, S 3, S} 4, S 5) and the first feature The cumulative feature distance S _Y corresponding to the vector y can be determined.

The updating device is configured to determine whether the cumulative feature distance S _Y corresponding to the first feature vector is greater than at least one of the cumulative feature distances S ₁ , S ₂ , S ₃ , S ₄ , S ₅ corresponding to the second feature vectors It can be judged whether it is large or not. The updating device may replace the existing registered image with the input image when the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors. The update apparatus may replace the registered image corresponding to the second feature vector having the minimum cumulative feature distance among the cumulative feature distances with the input image.

For example, let S ₄ <S ₅ <S _Y <S ₃ <S ₁ <S ₂ be the size of cumulative feature distances corresponding to each feature vector. At this time, the cumulative feature distance (S _Y ) corresponding to the first feature vector is larger than the cumulative feature distances (S ₄ , S ₅ ) corresponding to the second feature vectors. The updating device may replace the input image with any of the registered images corresponding to the second feature vector (x _4, x ₅ ) having the cumulative feature distance (S ₄ , S ₅ ). The updating apparatus can replace the registered image corresponding to the second feature vector x ₄ having the minimum cumulative feature distance S ₄ with the input image. The result obtained by replacing the registered image corresponding to the second feature vector x ₄ with the input image is shown in (c) of FIG.

The updating device compares the feature distances of the input image and the registered images with each other to replace the feature images, thereby widening the feature distance between the registered images, thereby improving the authentication rate for the more variously changed input images.

FIG. 7 is a diagram illustrating registered images adaptively updated in a registration database according to an exemplary embodiment. FIG. 8 is a diagram illustrating a relationship between feature distances between registered images initially registered in the registration database in FIG. 7, Fig.

7, the first registered image A ₁ (710), the added nine registered images A _2, A _3, A _{4 ,} A _5, A _6, A _7, A _8, A _9, A ₁₀ , 730 and the final 10 registration image is replaced with _{(a 1 ''', a} 2, a 3', a 4 ', a 5''', a 6 '', a 7, a 8 '', A _9, A ₁₀ ''') 750 are shown. 7, A ₁ 'corresponds to the first registered image A ₁ A ₁ '' represents an image that has been replaced twice, and A ₁ '''represents an image that has been replaced three times.

8 shows the feature distance between the registered images initially registered in the registration database and the lower figure shows the feature distance between the registered images finally registered in the registration database. Referring to FIG. 8, it can be understood that the feature distance between the registered images finally registered with respect to the registered images registered for the first time has a larger value.

In one embodiment, the feature distance between the registered images is widened by adding and replacing the registered images, thereby improving the authentication performance of variously changed face images.

9 is a block diagram of an adaptive updating apparatus of a registration database according to an embodiment. 9, an updating apparatus 900 according to an embodiment includes a processor 910, a memory 920, and an image sensor 930. [ The processor 910, the memory 920, and the image sensor 930 may communicate with each other via the bus 940. [

The processor 910 may adaptively update the registration database using the input image including the face of the user and the registered images already registered.

The processor 910 extracts the first feature vector from the input image including the user's face. The processor 910 determines whether to register the input image in the registration database based on the first feature vector, the second feature vectors of the registered images registered in the registration database, and the representative vectors representing the second feature vectors . The processor 910 registers the input image in the registration database based on the determination result.

Based on the first feature vector, the second feature vectors, and the representative vector, the processor 910 determines whether the input image corresponds to an outlier or a first feature vector and a feature of the registration database based on the second feature vectors And whether or not the range is extended. If the input image does not correspond to an outlier and the number of registered images is smaller than the maximum registered number, the processor 910 may add the input image to the registration database. If the number of registered images is greater than or equal to the maximum registered number, the processor 910 can further determine whether the feature range of the registration database is expanded. When the feature range is expanded, the processor 910 may replace the input image with one of the registered images.

In addition, the processor 910 may perform at least one of the methods described above with respect to Figures 1-8.

The processor 910 may execute the program and control the update device. The program code executed by the processor 910 may be stored in the memory 920. [ The update device is connected to an external device (e.g., a personal computer or a network) through an input / output device (not shown) and can exchange data.

The memory 920 stores a registration database including pre-registered images. The memory 920 may include a representative vector representative of the first feature vector extracted from the input image, the second feature vectors of the registered images, and the second feature vectors. Also, the memory 920 may store a representative vector updated by the newly registered input image and the newly registered input image. The memory 920 may be a volatile memory or a non-volatile memory. The image sensor 930 can capture an input image including a user's face.

The updating device 900 may be implemented as a software module and may be operated by at least one processor. A software module may be written in a form of a program in a memory coupled to the processor. Alternatively, the updating device 900 may be implemented as a hardware module. Alternatively, the updating device 900 may be implemented as a combination of software modules and hardware modules. In this case, the functions implemented by the software module are performed by the processor, and the functions implemented by the hardware module may be performed by the corresponding hardware. Processors and hardware can exchange signals with each other through input / output buses.

The updating device 900 may be a computing device such as a mobile phone, a smart phone, a PDA, a tablet computer, a laptop computer, a personal computer, a tablet computer, a netbook, or an electronic device such as a television, a smart television, And may include various electronic systems.

The embodiments described above may be implemented in hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, such as an array, a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described with reference to the drawings, various technical modifications and variations may be applied to those skilled in the art. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI &gt; or equivalents, even if it is replaced or replaced.

Claims (29)

Extracting a first feature vector from an input image including a user's face;
Determining whether to register the input image in the registration database based on the first feature vector, second feature vectors of registered images registered in the registration database, and representative vectors representative of the second feature vectors ; And
Registering the input image in the registration database based on the determination result
And updating the registration database. The method according to claim 1,
The step of determining whether to register
Determining whether the input image corresponds to an outlier based on the first feature vector, the second feature vectors, and the representative vector; And
Determining whether a feature range of the registration database is expanded based on the first feature vector and the second feature vectors
The method comprising the steps of: 3. The method of claim 2,
Wherein the step of determining whether the input image corresponds to an outlier
Calculating a minimum distance between the first feature vector and the second feature vectors;
Calculating a representative distance between the first feature vector and the representative vector; And
Determining whether the input image is an outlier based on the minimum distance and the representative distance,
And updating the registration database. The method of claim 3,
The step of determining whether the outlier is
Determining whether the input image is an outlier based on whether the minimum distance is smaller than a predetermined first threshold for updating the registration database and whether the representative distance is smaller than a preset second threshold
And updating the registration database. 3. The method of claim 2,
Wherein the step of determining whether the feature range of the registration database is extended
In a set of vectors comprising the first feature vector and the second feature vectors, a cumulative feature distance corresponding to each vector, the cumulative feature distance being based on distances between one vector and the remaining vectors in the vector set - determining whether the user has been selected; And
Determining whether a cumulative feature distance corresponding to the first feature vector is greater than at least one of cumulative feature distances corresponding to the second feature vectors
And updating the registration database. 6. The method of claim 5,
The step of registering in the registration database
Replacing any one of the registered images with the input image when the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors
And updating the registration database. The method according to claim 6,
The step of replacing any one of the registered images with the input image
Replacing the registered image corresponding to the second feature vector having the minimum cumulative feature distance among the cumulative feature distances with the input image;
And updating the registration database. The method according to claim 1,
The step of determining whether to register
Comparing the number of registered images registered in the registration database with the maximum registered number of the registration database
And updating the registration database. 9. The method of claim 8,
The step of registering the input image in the registration database
Adding the input image to the registration database when the number of registered images is smaller than the maximum registered number; And
Replacing one of the registered images registered in the registration database with the input image when the number of registered images is equal to the maximum registered number
And updating the registration database. 10. The method of claim 9,
The step of adding to the registration database
Adding the input image to the registration database if the input image is not an outlier and the number of registered images is smaller than the maximum registered number
And updating the registration database. The method according to claim 1,
The step of determining whether to register
Authenticating the user based on the first feature vector, the second feature vectors, and the representative vector
And updating the registration database. 12. The method of claim 11,
The step of authenticating the user
Calculating a minimum distance between the first feature vector and the second feature vectors;
Calculating a representative distance between the first feature vector and the representative vector; And
Authenticating the user based on the minimum distance and a result of comparison between a minimum value of the distance and a preset threshold value for user authentication
And updating the registration database. 12. The method of claim 11,
The step of determining whether to register
Determining whether to register the input image in the registration database based on whether the user authentication is successful or not
And updating the registration database. The method according to claim 1,
Updating the representative vector by the first feature vector when the input image is registered in the registration database
Further comprising the steps of: 15. A computer program stored on a medium for executing a method according to any one of claims 1 to 14 in combination with hardware. A memory for storing a registration database; And
Extracting a first feature vector from an input image including a face of the user and generating a second feature vector corresponding to the first feature vector, second feature vectors of registered images registered in the registration database, and representative vectors representative of the second feature vectors A processor for determining whether to register the input image in the registration database and registering the input image in the registration database,
And an adaptive updating device in the registration database. 17. The method of claim 16,
The processor
Based on the first feature vector, the second feature vectors, and the representative vector, whether the input image corresponds to an outlier or based on the first feature vector and the second feature vectors And whether or not the feature range of the registration database is extended. 18. The method of claim 17,
The processor
Calculating a minimum distance between the first feature vector and the second feature vectors, and a representative distance between the first feature vector and the representative vector, and based on the minimum distance and the representative distance, An adaptive update device of the registration database for determining if the license is a licensee. 18. The method of claim 17,
The processor
In a set of vectors comprising the first feature vector and the second feature vectors, a cumulative feature distance corresponding to each vector, the cumulative feature distance being based on distances between one vector and the remaining vectors in the vector set Determines whether the feature range of the registration database is expanded based on whether the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors Wherein the determination unit determines whether the registration database is updated. 20. The method of claim 19,
The processor
Wherein when the cumulative feature distance corresponding to the first feature vector is greater than at least one of the cumulative feature distances corresponding to the second feature vectors as a result of the determination, And replaces the registered image corresponding to the vector with the input image. 17. The method of claim 16,
The processor
And determines whether to add or replace the input image to the registration database based on a result of comparison between the number of registered images registered in the registration database and the maximum number of registered databases. 17. The method of claim 16,
The processor
The method comprising: authenticating the user based on the first feature vector, the second feature vectors, and the representative vector; and determining whether to register the input image in the registration database, based on whether the user authentication is successful An adaptive update device of the registration database. 17. The method of claim 16,
The processor
And updates the representative vector by the first feature vector when the input image is registered in the registration database. Authenticating an input image;
Determining whether the input image is an outlier;
Determining whether the feature range of the registration database is expanded due to the input image; And
Replacing one of the registered images of the registration database with the input image
And updating the registration database. 25. The method of claim 24,
The step of determining whether the input image is an outlier
Calculating a minimum distance between a first feature vector extracted from the input image and second feature vectors extracted from registered images of the registration database;
Calculating a representative distance between the first feature vector and a representative vector representative of the second feature vectors; And
Determining a first condition for the minimum distance and a second condition for the representative distance are satisfied,
And updating the registration database. 25. The method of claim 24,
Wherein the step of determining whether the feature range of the registration database is expanded due to the input image
Calculating a cumulative feature distance corresponding to each vector in a vector set including a first feature vector extracted from the input image and second feature vectors extracted from registered images of the registration database; And
Determining whether a cumulative feature distance corresponding to the first feature vector is greater than at least one of cumulative feature distances corresponding to the second feature vectors
And updating the registration database. 27. The method of claim 26,
The step of calculating a cumulative feature distance corresponding to each vector
Corresponding to each of the vectors, summing the corresponding vector and the distances between the remaining vectors in the vector set
And updating the registration database. Based on a representative vector representative of registered images of the registration database, determines whether the input image is an outlier, and based on the authentication result and the determination result, Lt; RTI ID = 0.0 &gt;
And an adaptive updating device in the registration database. 29. The method of claim 28,
The processor
In order to determine whether the input image is an outlier,
Calculating a minimum distance between a first feature vector extracted from the input image and second feature vectors extracted from registered images of the registration database, calculating a representative distance between the first feature vector and the representative vector, And determines whether a first condition for the minimum distance and a second condition for the representative distance are satisfied.

Images