KR20130049099A - Method and apparatus for estimating age or gender using face image - Google Patents

Abstract

PURPOSE: A method for estimating an age or a gender by using a face image and a device thereof are provided to use similarity between a face of a user and a registered face, thereby accurately estimating the age and the gender of the user without inputting the age or gender related information. CONSTITUTION: A face feature extraction unit(120) extracts face feature information from a face image of a user. A first similarity calculation unit(130) calculates similarity by comparing the extracted face feature information with face feature information of face images. A selection unit(133) selects the number of face images among the face images corresponding to the similarity. A second similarity calculation unit(137) calculates the similarity by comparing face feature information of the selected face images. An age/gender estimation unit(140) estimates an age or a gender of the user based on the similarity and ages or genders of the selected face images. [Reference numerals] (110) Face image input unit; (120) Face feature extraction unit; (130) First similarity calculation unit; (133) Selection unit; (137) Second similarity calculation unit; (140) Age/gender estimation unit; (150) Face information database

Description

Method and apparatus for estimating age or gender using face image {Method and apparatus for estimating age or gender using face image}

The present invention relates to a method and apparatus for estimating a gender or age of a user using face information.

A person's face contains a lot of information about that person. For example, a person can determine whether he or she is a man or a woman through the face of the other person, and can determine age, race, and the like. In addition to this, in addition to the traits that appear directly on the face, the character or personality can also be measured.

Currently, most automated devices use IDs such as ID cards and passports, or determine gender and age based on user-entered data. For example, in the case of the automatic cigarette vending machine, the resident registration card is input to confirm the age so that only an adult can purchase a cigarette. In the case of using the identification card as in the previous example, it is possible to accurately determine the age or gender, but there are many limitations in the application if the user does not want to actively use the service. For example, if you configure a system that sends personalized ads by estimating the age or gender of a passing user, requiring users to enter IDs or entering their age or gender directly is not appropriate for system operation and may affect user accessibility. There are many restrictions.

An object of the present invention is to provide a method and apparatus for estimating age or gender from a face of a user without directly inputting age or gender related information.

According to an aspect of the present invention, there is provided a method of estimating age or gender, comprising: facial feature extraction step of extracting facial feature information from a face image of a user; A first similarity calculation step of comparing the extracted facial feature information with facial feature information of each of the plurality of pre-stored face images to calculate a similarity between the face image of the user and each of the plurality of face images; A face image selecting step of selecting a predetermined number of face images from among the plurality of previously stored face images according to the similarity calculated in the first similarity calculating step; A second similarity calculation step of calculating a similarity between the selected face images by cross-cross comparing face feature information of each of the selected face images; And an age / gender estimation for estimating the age or gender of the user based on the age or gender of each of the selected face images, the similarity calculated in the first similarity calculating step, and the similarity calculated in the second similarity calculating step. Characterized in that it comprises a step.

The estimating of the age / gender may include correcting the age or gender of each of the selected face images using the similarity calculated in the second similarity calculating step, and calculating the corrected age or gender in the first similarity calculating step. Similarity may be used to calculate the age or gender of the user.

In the second similarity calculation step, when the predetermined number is N (N is a natural number of 2 or more), an N × N similarity matrix, wherein (m, n) (m = 1,…, N, n = 1, ..., N) component can be obtained that is the similarity between the m-th face image and the n-th face image.

The estimating of the age / gender may include correcting an age or gender of each of the selected face images using the similarity matrix, and using the similarity calculated in the corrected age or sex and the first similarity calculating step. The age or gender of the can be calculated.

The age or gender of each of the plurality of face images may correspond to the actual age or gender of the subject of the face image.

The age or gender of each of the plurality of face images may correspond to an age or gender corrected from the actual age or gender of the photographed person of the face image.

The facial feature information extraction step may use a Principal Component Analysis (PCA) technique or a Fisher Linear Discriminant (FLD) technique.

In the extracting of facial feature information, edge information of the facial image may be extracted as the facial feature information.

In the extracting of facial feature information, the face image is filtered using an edge extraction filter, an edge pattern of neighboring pixels is encoded for each pixel of the filtered image, the encoded image is divided into a plurality of partial regions, and each portion is Histograms can be generated for each area.

According to an aspect of the present invention, there is provided an apparatus for estimating age or gender, comprising: a face feature extractor extracting face feature information from a face image of a user; A face information database storing face feature information of each of a plurality of face images and information on an age or gender of each of the plurality of face images; A first similarity calculator which compares the extracted facial feature information with facial feature information of each of the plurality of face images to calculate a similarity between the user's face image and each of the plurality of face images; A selector which selects a predetermined number of face images from among the plurality of face images according to the similarity calculated by the first similarity calculator; A second similarity calculator configured to cross-compare face feature information of each of the selected face images to calculate similarity between the selected face images; And an age / gender estimation for estimating the age or gender of the user based on the age or gender of each of the selected face images, the similarity calculated by the first similarity calculator, and the similarity calculated by the second similarity calculator. It is characterized by including a wealth.

The age / gender estimator corrects the age or gender of each of the selected face images by using the similarity calculated by the second similarity calculator and calculates the corrected age or gender by the first similarity calculator. Similarity may be used to calculate the age or gender of the user.

The second similarity calculator, when the predetermined number is N (N is a natural number of 2 or more), is an N × N similarity matrix, where (m, n) (m = 1,…, N, n of the similarity matrix). = 1, ..., N) components can be obtained as the similarity between the m th face image and the n th face image.

The age / gender estimator corrects an age or gender of each of the selected face images using the similarity matrix, and uses the corrected age or gender and the similarity calculated by the first similarity calculator to determine the user's age. The age or gender can be calculated.

According to the present invention, the user may estimate the age or gender from the face of the user without directly inputting age or gender related information.

A person's age and gender are characteristic features that appear directly through the face. According to the present invention, since the user's age and gender are inferred from the age and gender information associated with the registered face, the user's age and sex can be estimated more accurately. have.

As an example of using the access control of the present invention, the age of the user may be estimated to prevent the minor from purchasing a cigarette using a vending machine. Examples of personalized service provision include product recommendation and advertisement according to age and gender. An example of entertainment may be comparing your face age with friends. Alternatively, in the face authentication system for door control, the user's age and gender may be determined in advance, and a more accurate user authentication may be performed by applying a model or a method suitable for the user's age and gender.

1 illustrates a configuration of an age / gender estimation apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating an age / gender estimation method according to an embodiment of the present invention.
3 illustrates an example of a face image of a user.
4 shows an example of the Sobel operator.
5 shows an example of a free operator.
6 illustrates a result of applying a Gaussian first derivative filter to a face image.
7 illustrates an example of an image filtered through an edge extraction filter.
8 illustrates an example of peripheral pixels of a center pixel of a filtered image.
9A and 9B show an example of a method of encoding a pattern of edges.
10 shows an example of an image in which an edge pattern is encoded.
11 shows an example of a method of generating a histogram from an encoded image.
12 illustrates an example of a method of generating a histogram by matching histograms of partial regions.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. In the following description and the accompanying drawings, substantially the same components are denoted by the same reference numerals, and redundant description will be omitted. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

1 shows a configuration of an age / gender estimation apparatus according to an embodiment of the present invention, and FIG. 2 shows a flowchart of an age / gender estimation method according to an embodiment of the present invention. As shown in FIG. 1, an age / gender estimation apparatus according to an exemplary embodiment of the present invention includes a face image input unit 110, a face feature extractor 120, a first similarity calculator 130, and a selector 133. And a second similarity calculator 137, an age / gender estimator 140, and a face information database 150. Hereinafter, the operation of the age / gender estimation apparatus will be described with reference to a flowchart of the age / gender estimation method illustrated in FIG. 2.

The face image input unit 110 receives a face image of a user who wants to estimate an age or a gender (step 210). The face image input unit 110 may receive a face image of a user from various image acquisition devices such as a CCTV camera, a VCR, a DVR, a web camera, and the like. In addition, a face image may be input through an image file of a computer. The face image may be various types of images such as an infrared image as well as a visible light image. In some cases, only specific image information may be input or various types of image information may be received together.

The facial feature extractor 120 extracts facial feature information from the face image of the user in operation 220. In order to extract face feature information, the face feature extractor 120 may determine whether a face is included in an input image and detect a face region from the input image. The face feature extractor 120 may normalize the input face image as shown in FIG. 3, and then extract the face feature with the normalized face image.

In one embodiment, the facial feature extractor 120 may extract facial feature information by applying a Principal Component Analysis (PCA) technique to the face image. The PCA technique is also known as the eigenface method. As another example, the facial feature extractor 120 may extract a feature by applying a Fisher Linear Discriminant (FLD) technique to the face image. The FLD technique is also known as the fisherface method.

As another embodiment, the face feature extractor 120 may extract edge information of the face image and use the face feature information. In the face image, an edge corresponds to an edge of a face, an outline, or a portion in which a color value or brightness changes relatively large in the image. The edge information may better describe or express the shape, appearance, or shape of the face in the face image, the edge, contour, or edge information of the face.

The facial feature extractor 120 may extract edge information from an image using an image filter. At this time, the filtered image includes edge information.

The facial feature extractor 120 applies an edge extraction filter, such as a Sobel operator as shown in FIG. 4 or a Prewitt operator as shown in FIG. 5, to the face image to apply edge information of the face. Can be extracted.

Also, the facial feature extractor 120 may extract edge information using a Gaussian differential filter or a Laplacian-Gaussian filter in which a Gaussian filter and an edge extraction filter are combined.

Basically, the Gaussian filter is defined as in Equation 1.

Gaussian first-order differential filters are defined as in Equation 2.

The Gaussian second-order differential filter is defined as in Equation 3.

In addition, the Laplacian-Gaussian filter is defined as in Equation 4.

The Laplacian-Gaussian filter is a filter that extracts edges regardless of the direction of the edges. On the other hand, the Gaussian first-order derivative filter and the Gaussian second-order derivative filter can extract edges in a specific direction. For example, the result of applying the Gaussian first derivative filter to the face image as shown in FIG. 3 is shown in FIG. 6. 6 illustrates a result of applying a Gaussian first order differential filter to a face image and a corresponding Gaussian first order differential filter.

When the edge information is extracted, the facial feature extractor 120 divides the face image into a plurality of partial regions, and describes characteristics (eg, distribution or pattern of edges) of the edge for each divided partial region. An example of such a process will be described with reference to FIGS. 7 to 12.

First, it is assumed that the image of FIG. 7 is an image filtered through an edge extraction filter. If a plurality of edge extraction filters are used, the following process is equally applied to each filtered image.

Referring to FIG. 7, each column represents a pixel or a cell of the filtered image, and horizontal and vertical numbers represent positions of rows and columns of pixels in the image. For example, the position of the center pixel indicated by "X" is represented by (2, 3). Here, it is assumed that the values of the center pixel (2, 3) and the pixels around it are as shown in FIG.

One way to express the pattern of the edge is to set it to 1 if the value of the surrounding pixel is greater than or equal to the value of the center pixel, and to 0 if it is smaller. 9A shows the result of applying this method to the center pixel (2,3) and the pixels around it. Referring to FIG. 9A, a pattern of edges can be expressed by encoding binary values starting from pixel (1,2) and rotating clockwise to encode a single value of 01001100 ₍₂₎ or decimal 64 + 8 + 4 = 76. Can be. In this case, a total of 256 patterns can be generated, and this method is known as local binary pattern (LBP) analysis in the field of image processing.

Another way of expressing the pattern of the edge is to calculate the average value of the value of the center pixel and the surrounding pixels, and then set it to 1 if the value is greater than or equal to the average value and 0 if the value is smaller than the mean value. 9B shows the result of applying this method to the center pixel (2, 3) and the pixels around it. Referring to FIG. 9B, a pattern of edges can be expressed by rotating in a clockwise direction starting from pixel (1,2), arranging binary values with the last as a center pixel, and encoding them into a single value of 010011000 ₍₂₎ or decimal 152. have. In this case, a total of 512 patterns can be generated, and this method corresponds to the Modified Census Transform (MCT) analysis in the field of image processing.

Now, it is assumed that the filtered image 13X13 of FIG. 7 is represented by the image 11X11 as shown in FIG. 10 through an encoding process. The values of the pixels (1,2) represented by X in FIG. 10 are the center pixel (2,3) and the surrounding pixels (1,2), (1,3), (1,4), (2) in FIG. Corresponds to a value obtained by encoding edge patterns of (2, 4), (2, 4), (3, 2), (3, 3), and (3, 4).

The face feature extractor 120 splits the encoded image into a plurality of partial regions as shown in FIG. 11 and generates a histogram corresponding to an edge pattern for each partial region. At this time, the partial regions may be divided to overlap each other or may not be overlapped. In addition, without having to construct bins of histograms for all possible patterns, similar patterns can be grouped into one pattern to indicate frequency or ignore statistically insignificant patterns.

Next, the facial feature extractor 120 extracts a feature, that is, a global feature, for a wider range than the partial region or the entire region of the face using the characteristics of the edge of each partial region. 12 illustrates this global feature. The facial feature extractor 120 matches a histogram corresponding to a characteristic of an edge of each partial region as illustrated in FIG. 12 to generate one histogram. One histogram generated here corresponds to feature information of the entire area of the face. In this case, all histograms for each partial region may be used, or only some partial regions may be selected and used in various combinations. Alternatively, data analysis methods such as Principal Component Analysis (PCA) or Fisher Linear Discriminant (FLD) can be additionally applied to individual histograms or various combinations to matched histograms or to one histogram, thereby reducing the dimension of data and improving classification performance. .

1 and 2, the first similarity calculator 130 compares facial feature information extracted from a user's face image with facial feature information of each of the plurality of face images stored in the face information database 150. In operation 230, the degree of similarity between the face image of the user and each of the plurality of face images is calculated.

The face information database 150 stores face feature information of each of the plurality of face images and information about an age or gender of each of the plurality of face images.

The facial feature information stored in the face information database 150 corresponds to facial feature information of the same type as the facial feature information extracted by the facial feature extractor 120.

The information about the age or gender stored in the face information database 150 may be information about the actual age or gender of the photographer of the face image. However, since an age or gender may be determined differently from person to person according to a face, information about an age or a gender stored in the face information database 150 may be different from an actual age or gender of the subject of the face image in some cases. Can be stored with a calibrated age or gender.

For example, if the face looks much younger than the actual age, the age information may be registered in the face information database 150 less than the actual age. Or, if you set gender information to +1 for men and women to -1, if a man looks feminine and the female image is strong, the gender information for that face is between +1 and -1, such as +0.5 or +0.3. Can also be set to

In the present specification, the 'similarity' calculated by the first similarity calculator 130 is a concept including 'similarity'. Similarity is a value that increases with similar features between images, and dissimilarity decreases with similar features between images. Similarity and dissimilarity are inseparably related to each other and can be easily converted to each other. For example, dissimilarity can be obtained by taking the inverse of the similarity.

The first similarity calculator 130 may calculate the similarity between two facial feature information by using the cosine similarity. Alternatively, the first similarity calculator 130 may calculate dissimilarity between two facial feature information using Euclidean distance. The similarity calculator 130 may normalize or convert the obtained similarity or dissimilarity into a value between 0 and 1 to have a meaning of probability or posterior probability. Hereinafter, for convenience of explanation, the description will be based on the similarity and the similarity is assumed to have a positive value.

The selector 133 may include a predetermined number N of face images (or similarities) among the face images of the face information database 150 according to the similarity of each face image calculated by the first similarity calculator 130. Top N names) (step 240). If the user's age or age group is previously determined or given prior to the user's gender estimation, the upper predetermined number N of the face images of the face information database 150 corresponding to the same age group as or similar to the user's age Facial images (or top N names) may be selected. Herein, the N value may be a predetermined fixed value or may be a value that is fluidly determined according to the similarity calculation result. Of course, the N value may be the total number of faces (or the total number of people) stored in the face information database 150.

The second similarity calculator 137 calculates the similarity between the selected face images by performing cross-cross comparison between face feature information of each of the face images selected by the selector 133 (step 250). In other words, for each face pair obtained from the selected face images, similarity between face images is calculated by comparing face feature information. Since the similarity calculated by the second similarity calculator 137 is the same concept as the similarity calculated by the first similarity calculator 130, repeated description thereof will be omitted.

를 구할 수 있다. 여기서, 상기 유사도 행렬의 (m, n) (m=1, …, N, n=1, …, N) 성분은 m번째 얼굴 영상과 n번째 얼굴 영상 간의 유사도가 된다. 물론, m과 n이 동일한 경우는 해당 성분은 동일한 얼굴 영상 간의 유사도이므로, 100% 유사함을 나타내는 값(예를 들어, 1)이 될 것이다. The second similarity calculator 137 assumes that N face images are selected, the N × N similarity matrix having these similarities as a component.

Can be obtained. Here, the (m, n) (m = 1, ..., N, n = 1, ..., N) component of the similarity matrix is the similarity between the m th face image and the n th face image. Of course, when m and n are the same, since the corresponding components have similarities between the same face images, they will be values indicating 100% similarity (eg, 1).

The age / gender estimator 140 includes the age or gender of each of the face images selected by the selector 133, the similarity calculated by the first similarity calculator 130, and the second similarity calculator 137. In step 260, the user's age or gender is estimated based on the similarity calculated in step (a).

In this step, the age / gender estimator 140 first corrects the age or gender of each of the selected face images using the similarity calculated by the second similarity calculator 137.

(

)라 하면, N개의 얼굴 영상의 나이 또는 성별은

를 성분으로 하는 N 차원의 벡터

로 표현할 수 있다. 여기서, 성별 값은 전술한 바와 같이 +1, -1의 이진 값으로, 표현하거나, +1 과 -1 사이의 값으로 표현할 수 있다. The gender value or age value of the N face images selected by the selector 133

(

), The age or gender of the N facial images

N-dimensional vector containing

. Here, as described above, the gender value may be represented by a binary value of +1, -1, or a value between +1 and -1.

와 상기 유사도 행렬

를 이용하여, 선택된 얼굴 영상들 각각의 보정된 나이 또는 성별을 구할 수 있다.Age / gender estimation unit 140 is the vector

And the similarity matrix

By using, the corrected age or gender of each of the selected face images may be obtained.

는 보정된 나이 또는 성별을 나타낸다.

의 t는 반복 회수로서, 상기 수학식은

가 재귀적으로 반복되어 계산됨을 나타낸다. 즉, 보정된 나이 또는 성별

는 이전 단계에서 계산된 보정된 나이 또는 성별

과 유사도 행렬

, 그리고 상기 벡터

로부터 구해진다.

는 조합 가중치로서, 예컨대 0과 1 사이의 값이 될 수 있으며, 임의로 설정될 수 있다.

의 초기값은 임의로 설정될 수 있으며,

는 반복 회수가 늘어남에 따라 어떤 값에 수렴하게 된다. 위와 같은 수학식을 이용하여 보정된 나이 또는 성별 값을 계산함에 있어서, 반복 회수는 미리 정해진 회수가 될 수도 있고, 반복에 따른

값이 변화가 소정의 임계값 미만일 때(즉, 수렴한 것으로 판단될 때)까지 반복할 수도 있다. here,

Represents the corrected age or gender.

T is the number of repetitions, and

Denotes recursively repeated. That is, the corrected age or gender

Is the calibrated age or gender calculated in the previous step

And similarity matrix

, And the vector

Obtained from

Is a combination weight, for example, may be a value between 0 and 1, and may be arbitrarily set.

The initial value of can be arbitrarily set.

Converges to a value as the number of iterations increases. In calculating the corrected age or gender value by using the above equation, the number of repetitions may be a predetermined number,

The value may be repeated until the change is below a predetermined threshold (ie, determined to converge).

Age value or gender information is revealed through the face, but there is a limit in accurately representing the relationship between the face image and age or gender. The process of correcting the age or gender of each of the selected face images is a process of adjusting the age or gender of each face image in consideration of the similarity between the face images in order to more accurately estimate the age or gender of the user. .

When the age or gender of each of the selected face images is corrected, the age / gender estimator 140 uses the corrected age or gender and the similarity calculated by the first similarity calculator 130 to determine the age or gender of the user. Calculate Here, the age / gender estimator 140 calculates a weighted average of the corrected age or gender of each of the selected face images by weighting the similarity with the face image of the user of each of the selected face images, and thus the age or gender of the user. Gender can be obtained. To this end, for example, the age or gender of the user may be calculated according to the following equation.

는 선택부(133)에서 선택된 i번째 얼굴 영상의 사용자 얼굴 영상과의 유사도이고,

는 상기 i번째 얼굴 영상의 보정된 나이 또는 성별을 나타내고,

이다.

가 나이 값에 해당하는 경우, 수학식 6의 값이 사용자의 추정된 나이가 될 수 있다.

가 성별 값에 해당하는 경우, 수학식 6의 값이 부호가 사용자의 추정된 성별이 될 수 있다. 예컨대, 부호가 (+)이면 남성으로, (-)이면 여성으로 추정된다. here,

Is a similarity degree to the user face image of the i-th face image selected by the selecting unit 133,

Represents the corrected age or gender of the i-th face image,

to be.

Is a age value, the value of Equation 6 may be the estimated age of the user.

Is a gender value, the value of Equation 6 may be the estimated gender of the user. For example, if the sign is positive, it is assumed to be male, and if it is negative, it is assumed to be female.

The above-described embodiments of the present invention can be embodied in a general-purpose digital computer that can be embodied as a program that can be executed by a computer and operates the program using a computer-readable recording medium. The computer-readable recording medium includes a storage medium such as a magnetic storage medium (e.g., ROM, floppy disk, hard disk, etc.), optical reading medium (e.g., CD ROM,

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

Claims (15)

A facial feature extraction step of extracting facial feature information from a face image of a user;
A first similarity calculation step of comparing the extracted facial feature information with facial feature information of each of the plurality of pre-stored face images to calculate a similarity between the face image of the user and each of the plurality of face images;
A face image selecting step of selecting a predetermined number of face images from among the plurality of previously stored face images according to the similarity calculated in the first similarity calculating step;
A second similarity calculation step of calculating a similarity between the selected face images by cross-cross comparing face feature information of each of the selected face images;
An age / gender estimation step of estimating the age or gender of the user based on the age or gender of each of the selected face images, the similarity calculated in the first similarity calculating step, and the similarity calculated in the second similarity calculating step Age or gender estimation method comprising a. The method of claim 1,
The estimating of the age / gender may include correcting the age or gender of each of the selected face images using the similarity calculated in the second similarity calculating step, and calculating the corrected age or gender in the first similarity calculating step. And calculating the age or gender of the user using the similarity. The method of claim 1,
In the second similarity calculation step, when the predetermined number is N (N is a natural number of 2 or more), an N × N similarity matrix, wherein (m, n) (m = 1,…, N, n = 1, ..., N) component is a similarity between the m-th face image and the n-th face image. The method of claim 3,
The estimating of the age / gender may include correcting an age or gender of each of the selected face images using the similarity matrix, and using the similarity calculated in the corrected age or sex and the first similarity calculating step. Calculating the age or sex of the age or sex estimation method. The method of claim 1,
Age or gender of each of the plurality of face images corresponding to the actual age or gender of the subject of the face image. The method of claim 1,
The age or gender of each of the plurality of face images corresponds to the age or gender corrected from the actual age or gender of the subject of the face image. The method of claim 1,
The facial feature extracting method may include age or sex estimation method using a Principal Component Analysis (PCA) technique or a Fisher Linear Discriminant (FLD) technique. The method of claim 1,
The facial feature extracting step may include extracting edge information of the face image as the facial feature information. The method of claim 1,
In the facial feature extraction step, the face image is filtered using an edge extraction filter, an edge pattern of neighboring pixels is encoded for each pixel of the filtered image, the encoded image is divided into a plurality of partial regions, and each partial region is extracted. Age or gender estimation method characterized in that for generating a histogram for each. A facial feature extractor extracting facial feature information from a face image of a user;
A face information database storing face feature information of each of a plurality of face images and information on an age or gender of each of the plurality of face images;
A first similarity calculator which compares the extracted facial feature information with facial feature information of each of the plurality of face images to calculate a similarity between the user's face image and each of the plurality of face images;
A selector which selects a predetermined number of face images from among the plurality of face images according to the similarity calculated by the first similarity calculator;
A second similarity calculator configured to cross-compare face feature information of each of the selected face images to calculate similarity between the selected face images; And
An age / gender estimator for estimating the age or gender of the user based on the age or gender of each of the selected face images, the similarity calculated by the first similarity calculator, and the similarity calculated by the second similarity calculator; Age or gender estimation device comprising a. The method of claim 10,
The age / gender estimator corrects the age or gender of each of the selected face images by using the similarity calculated by the second similarity calculator and calculates the corrected age or gender by the first similarity calculator. Age or gender estimation device for calculating the age or gender of the user using the degree of similarity. The method of claim 10,
The second similarity calculator, when the predetermined number is N (N is a natural number of 2 or more), is an N × N similarity matrix, where (m, n) (m = 1,…, N, n of the similarity matrix). = 1, ..., N) components are similarities between the m th face image and the n th face image. The method of claim 12,
The age / gender estimator corrects an age or gender of each of the selected face images using the similarity matrix, and uses the corrected age or gender and the similarity calculated by the first similarity calculator to determine the user's age. Age or gender estimation device, characterized in that for calculating the age or gender. The method of claim 10,
Age or gender of each of the plurality of face images includes the actual age or gender of the subject of the face image. The method of claim 10,
The age or gender of each of the plurality of face images includes an age or gender corrected from the actual age or gender of the subject of the face image.

Images