KR101310886B1 - Method for extracting face characteristic and apparatus thereof - Google Patents

Abstract

PURPOSE: A face characteristic extracting method and a device thereof are provided to extract face characteristic while maintaining the maximum quantity of information demanded for face identification without linear conversion causing time and space complexity on mass source data. CONSTITUTION: A source characteristic obtaining unit (11) obtains a source characteristic group including source face characteristic elements which are filter response data obtained by applying a face recognition filter to a face image. A characteristic candidate group generating unit (12) extracts characteristic elements satisfying a first condition from the source characteristic group to create a characteristic candidate group. An intermediate characteristic generating unit (13) generates an intermediate characteristic group including the first set number of characteristic elements from the characteristic candidate group. A final characteristic obtaining unit (14) obtains a final characteristic group including the second set number of characteristic elements used for face recognition from the intermediate characteristic group. [Reference numerals] (1) Face characteristic extracting device; (11) Source characteristic obtaining unit; (12) Characteristic candidate group generating unit; (121) Dispersion calculating module; (122) Comparison module; (123) Generating module; (13) Intermediate characteristic generating unit; (14) Final characteristic obtaining unit; (15) Characteristic candidate group excluding unit; (16) Characteristic candidate group renewing unit

Description

Method for extracting face characteristic and apparatus thereof

The present invention relates to face recognition, and more particularly, to a method and apparatus for extracting a facial feature amount for face recognition from data acquired from a face image.

Face recognition among biometrics is applied to various fields, and various studies are being conducted to increase the accuracy of face recognition. Face identity recognition converts the shape information of the face portion into a feature amount in a given input image, and compares the feature amount obtained from the input image with the feature amounts of the registered image to determine the identity of the input face image. This is done throughout the process.

The human face is composed of parts of the forehead, eyebrows, eyes, nose, and mouth, and is a three-dimensional object. Therefore, a large change occurs in the shape projected onto the imaging plane depending on the change factors such as illumination, viewing angle (or camera angle), and the person who is the subject, that is, the presence or absence of a facial expression of the subject. The degree of change depends on the face. For example, since the nose is more three-dimensional than the eye, the shape projected onto the imaging plane has a larger change with the shooting angle. On the other hand, if the subject makes a big smile on the condition of the same shooting angle, the shape of the image of the nose is almost unchanged compared to the case of no expression, but the shape of the image of the eyes and the mouth is greatly changed.

Many existing face recognition technologies generally quantify and process the entire face area as a single object, and thus have a poor ability to respond to changing factors such as facial expressions, lighting, and angles.

Filtering that enables quantification in the frequency domain is widely used for computerized identification of acquired face images. The face recognition filters used for filtering include Gabor, Haar, and LBP (local). binary pattern)-including discriminative LBP (DLBP), uniform LBP (ULBP), number LBP (NLBP), and so on. The Gabor filter has frequency and direction characteristics, and raw feature variable vectors obtained by applying a combination of a plurality of frequencies and directions to an image are generally very large in order to secure an amount of information required for face recognition. For example, the raw feature quantity vectors resulting from applying a filter bank of 24 filters to a 64x64 image may consist of 98304 elements.

When face recognition is performed by processing such very large feature vectors, principal component analysis (PCA) and linear discriminant analysis (LDA) are used to obtain the final feature vector by reducing the dimension of the primitive vectors while minimizing the loss of information. ) May be used.

However, these methods are based on linear transformations and require very large matrices and long computation times for their operation. For example, to convert a vector of 98304 elements into a final feature vector of 256 elements, you need a matrix of about 25 million (n to 98304 × 256) elements, with computationally at least n multiplications and equal numbers. Addition is required. Therefore, this feature is an obstacle in applying the above methods in combination (for example, using all Gabor filter + PCA + LDA methods) or in real-time face recognition processing.

The technical problem to be solved by the present invention is to provide a method and apparatus for extracting the feature amount of the face more simply while maintaining the maximum amount of information required for face identification.

For the above technical problem, the facial feature extraction method according to an aspect of the present invention is a raw feature set including raw face feature feature elements that are filter response data obtained by applying at least one face recognition filter to a face image Obtaining a; Generating a feature variable candidate group by extracting feature variable elements satisfying a set condition from the original feature variable set; Generating an intermediate set of feature quantities comprising a first set number of feature quantities elements from the feature quantity candidate group; And obtaining a final feature set comprising a second set number of feature quantities elements to be used for face recognition from the intermediate feature set, wherein generating the intermediate feature set includes generating the feature variable candidate group If it is determined from that that one feature variable is included in the intermediate step feature set, some feature variables associated with the feature variable are deleted from the feature variable candidate group.

In the generating of the intermediate step feature set, the feature variable corresponding to the pixel of the face image positioned within a set radius with respect to the one feature variable may be deleted from the feature variable candidate group.

The generating of the feature variable candidate group may further include calculating a variance for each feature variable included in the raw feature variable set; Comparing the calculated variance with a first predetermined threshold; And extracting feature variable elements whose calculated variance is greater than a first threshold to generate the feature variable candidate group.

The generating of the intermediate step feature set may include selecting one of feature set elements included in the feature set candidate group; Calculating a correlation with the feature amount element already included in the intermediate step feature set for the selected feature variable; Comparing the calculated correlation with a second preset threshold; Including the corresponding feature variable in the intermediate step feature set when the calculated correlation is less than a second threshold; And deleting some feature variables related to the feature variable from the intermediate feature set from the feature variable candidate group. The selecting, the calculating, the comparing, the including, and the deleting are repeatedly performed until the first set number of feature elements is included.

The obtaining of the final feature set may include calculating a variance for each feature set element included in the intermediate feature set; Arranging feature quantity elements based on the value of the variance; And extracting the second set number of feature variable elements based on the feature variable having the largest variance from the sorted feature variable elements to obtain a final feature set.

In addition, the method may further include updating the feature variable candidate group based on an in-person variance value for each feature variable of the feature variable candidate group, and generating the intermediate feature set includes: A set of intermediate feature quantities may be generated using the selected feature variable candidate group.

In this case, the updating may include calculating an intra-person variance value of the feature variables included in the feature variable candidate group; Comparing the calculated in-person dispersion value with a third predetermined threshold value; And if the calculated intrapersonal variance value is smaller than the third threshold, including the corresponding feature variable in the new feature variable candidate group.

The generating of the feature variable candidate group may further include excluding feature variable elements corresponding to pixels at preset positions with respect to the face area included in the face image from the feature variable candidate group.

An apparatus for extracting facial features according to another aspect of the present invention is a primitive feature for obtaining a set of primitive feature quantities including primitive face feature variables that are filter response data obtained by applying at least one face recognition filter to a face image Quantity obtaining unit; A feature variable candidate group generation unit for generating a feature variable candidate group by extracting feature variable elements satisfying a first condition from the original feature variable set; An intermediate step feature variable generating unit for generating an intermediate step feature variable set including a first set number of feature variable elements from the feature variable candidate group; If it is determined from the feature variable candidate group that one feature variable is included in the intermediate step feature set, the feature variable corresponding to the pixel of the face image located within a set radius with respect to the one feature variable is determined. A feature amount candidate group exclusion unit to delete from the feature variable candidate group; And a final feature variable obtaining unit for obtaining a final feature variable set including a second set number of feature variable elements to be used for face recognition from the intermediate step feature set. Here, the intermediate step feature variable generating unit generates the intermediate step feature set based on the feature variable candidate group processed by the feature variable candidate group exclusion unit.

The feature variable candidate group generation unit includes: a variance calculation module configured to calculate a variance for each feature variable included in the source feature set; A comparison module for comparing the calculated variance with a first preset threshold; And a generation module for generating the feature variable candidate group by extracting feature variable elements whose calculated variance is greater than a first threshold.

The intermediate step feature variable generating unit may further include: a selecting module configured to select one of the feature variable elements included in the feature variable candidate group; A correlation calculation module for calculating a correlation with the feature amount element already included in the intermediate step feature set for the selected feature variable; A comparison module for comparing the calculated correlation with a preset second threshold value; And a set generation module for including the corresponding feature variables in the intermediate step feature set when the calculated correlation is smaller than the second threshold.

On the other hand, the apparatus calculates the in-person variance value for the feature amount elements included in the feature variable candidate group, and based on only the feature amount elements whose calculated in-person variance value is smaller than a third predetermined threshold, The apparatus may further include a feature amount candidate group updater.

According to an embodiment of the present invention, the maximum amount of information required for face identification is maximized without using a linear transformation that causes time and space complexity from a vast amount of raw data obtained for face recognition. A feature amount indicating a facial feature can be extracted while maintaining it.

1 is an exemplary view showing filter response data according to an exemplary embodiment of the present invention.
2 is a flowchart of a facial feature extraction method according to an exemplary embodiment of the present invention.
3 is a flowchart illustrating a process of including a feature variable in an intermediate stage feature set according to an embodiment of the present invention.
4 and 5 are flowcharts of a facial feature extraction method according to another embodiment of the present invention.
6 is a structural diagram of a facial feature extraction apparatus according to an embodiment of the present invention.
FIG. 7 is a structural diagram of an intermediate stage feature variable generating unit shown in FIG. 6.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention.

In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification and claims, when a section is referred to as including an element, it is understood that it may include other elements as well, without departing from the other elements unless specifically stated otherwise.

Now, a facial feature extraction method and a facial feature extraction apparatus using the same will be described.

In an embodiment of the present invention, a face recognition filter is used to extract a facial feature amount. The face recognition filter includes a plurality of factors, and the factors may include frequency, orientation, and the like. The characteristics of the face recognition filter change according to factors. The filter response data may be obtained by applying a face recognition filter to the face image.

1 is an exemplary view showing filter response data according to an exemplary embodiment of the present invention. In FIG. 1, filter response data obtained when a plurality of face recognition filters having different factors are applied to a face image, not only one face recognition filter is applied. Here, the face recognition filter is composed of Gabor filters, and a set of filters including face recognition filters having different characteristics based on 24 directions and three frequencies is applied to the face image. Here, as a face image, images included in a face population set including images of a face having statistical representativeness may be used. The images included in the face population set may be referred to as training images.

In this way, the filter response data obtained by applying the face recognition filters to the face image may be referred to as a facial feature amount. Since the face feature amounts are enormous, it takes a lot of processing time and computation to perform face recognition based on the face feature amount acquired by the filter action, that is, the raw feature amount. Therefore, in the embodiment of the present invention, the predetermined feature amounts are extracted from the original feature amounts and used as the final feature amounts that are substantially used for face recognition.

2 is a flowchart of a facial feature extraction method according to an exemplary embodiment of the present invention.

The facial feature extracting method according to an embodiment of the present invention may be performed by the facial feature extracting apparatus, which will be described in more detail later.

According to an embodiment of the present invention, as shown in FIG. 2, the raw facial feature amount element, which is filter response data obtained by applying at least one face recognition filter to a face image (hereinafter, referred to as a primitive feature) The quantity elements are called (S100). A set of planes including the primitive feature variables f may be referred to as a primitive feature set F0, and the primitive feature set F0 may have the characteristics shown in FIG.

A feature variable candidate group C is generated by extracting feature variable elements satisfying the first condition from the original feature set F0 (S110). In more detail, a variance is calculated for each face population set of feature variables included in the primitive feature set F0, and the first condition, that is, the calculated variance satisfies a condition larger than the preset first threshold value θv. The feature variables are extracted from the raw feature set F0. A variance var (f) is calculated for each of the feature quantities elements f constituting the set of raw feature quantities, and the feature quantity candidate groups are extracted by extracting feature quantities elements whose calculated variance is larger than the first threshold value θv. Generate C The feature variable candidate group C can be expressed as follows.

The feature variable candidate group C represents candidate feature variables for generating F1, which is an intermediate step feature set for calculating the final feature set. An intermediate step characteristic quantity set F1 is generated based on the characteristic quantity elements included in the characteristic quantity candidate group C. FIG.

Herein, in order to generate the intermediate step feature set F1 including the first set number of feature quantities, one of the feature quantities included in the feature group candidate group C may be selected (S120).

The selected feature variable may be included in the intermediate feature set F1 (S130).

3 is a flowchart illustrating a process of including a feature variable in an intermediate stage feature set according to an embodiment of the present invention.

As shown in FIG. 3, a random number vector r is generated and a feature amount element f corresponding to the random number vector r is selected from the feature variable candidate group C based on the random number vector r (S131 and S132).

With respect to the selected feature variable f, all correlations with each feature variable f 'already included in the intermediate feature set F1 are calculated (S133), and it is determined whether the calculated correlations satisfy the second condition. do. Specifically, the calculated correlation is compared with the preset second threshold value θc (S134), and when the value is smaller than the second threshold value θc, the feature amount element f is included in the intermediate step feature set F1. That is, when the second condition of the correlation cor (f, f) &lt; theta < c &gt; It is included in the step feature set F1 (S135) ( F1 ← F1 + f). However, if the above condition is not satisfied, the selected feature variable f is not included in the intermediate feature set F1 (S136).

On the other hand, the selected selected feature variable f may be included in the intermediate feature set F1 without performing the subsequent steps S133 and 134 in step S132.

Through the above steps, if one feature amount element f to be included in the intermediate step feature set F1 is selected from the feature group candidate group C, as shown in FIG. Then, in order to simplify the subsequent processing, some feature amount elements may be excluded from the feature amount candidate group C based on the selected feature amount element f (S140).

Due to the two-dimensional nature of the facial image itself, feature quantities corresponding to pixels that are adjacent or close to each other have a strong correlation with each other. Therefore, if it is determined that one feature amount element f is included in the intermediate step feature set F1, the feature amount elements n_w (f corresponding to neighboring pixels located within the set radius w with respect to the determined feature amount element f )) Is excluded from feature quantity candidate group C. Accordingly, the feature variable candidate group may be updated as C ← C -f-n_w (f).

This process is repeated until the intermediate stage feature set F1 including the feature number elements of the set number K1 (| F1 | = k1).

When the intermediate feature set F1 including the feature number elements of the set number K1 is generated (S150), a final feature set F2 to be used for face recognition is obtained from the intermediate feature set F1 (S160).

In order to obtain the final feature set F2, an existing algorithm, for example, linear discriminant analysis (LDA) can be used. Specifically, since the intermediate feature set is generated from the original feature set that includes the feature variables that can clearly represent the feature of the face, the final feature variable is applied to the intermediate feature set by applying the existing algorithm, LDA. The set F2 can be obtained ( F2 <-LDA ( F1 , k2)). Already, the number of feature elements of the intermediate feature set is significantly reduced compared to the original feature set, and thus, even when the existing algorithm (LDA) is used, computation time and complexity can be significantly reduced.

Also, in order to obtain the final feature set F2, the variances for each feature set included in the intermediate feature set F1 are obtained, and the feature quantities are sorted based on the value of the variance. Then, the feature number elements of the set number K2 are extracted based on the feature amount element having the largest variance from the aligned feature amount elements, thereby generating a final feature amount set F2.

On the other hand, unlike the embodiment described above, another method may be used when generating the intermediate feature set.

4 and 5 are flowcharts of a facial feature extraction method according to another embodiment of the present invention.

Here, unlike the embodiment described above, a new feature variable candidate group is generated by generating a feature variable candidate group and then performing additional screening on the generated feature variable candidate group.

Specifically, as shown in FIG. 4, in the same manner as in the above-described embodiment, the characteristic quantity elements satisfying the first condition C = {f | var (f)> θv} are extracted from the original characteristic quantity set After the quantity candidate group C is generated (S300 and S310), the intrapersonal variance value for the feature quantity element is calculated for the feature quantity candidate group C (S320), and the predetermined amount is determined from the feature quantity candidate group C based on the calculated intrapersonal variance value. Characteristic quantity elements are extracted to generate a new characteristic quantity candidate group C '.

The variance value within an individual represents a difference for each of a plurality of images of the same person, and in detail, represents a difference between feature elements corresponding to a pixel at a predetermined position in the plurality of images acquired for the same person. The intrapersonal variance value var_w (f) for the feature amount element f corresponding to the pixel at the predetermined position and included in the feature quantity candidate group C is calculated, and the calculated intrapersonal variance value var_w (f) is set in advance. It is determined whether it is smaller than the three threshold value θ_wv (S330).

When the in-person variance value var_w (f) for the feature amount element f is smaller than the third threshold value θ_wv, the feature amount element f (that is, the feature amount elements corresponding to the pixels at the predetermined position) is new. The amount is included in the candidate group C '(S340).

This process (S320 to S340) is applied to each feature variable in the feature variable candidate group C, thereby creating a new feature variable candidate group C 'including only feature variable elements satisfying the above condition (S330).

Subsequently, when the generation of the new feature variable candidate group C 'is completed, as shown in FIG. 5, the intermediate feature set F1 is generated from the feature variable candidate group C' in the same manner as in the above embodiment (S360 to S390). A final feature set F2 to be used for face recognition is obtained from the feature set F1 (S400).

Meanwhile, in the above-described embodiments, the feature amount candidate group C or C 'corresponds to a portion that is more affected by shooting conditions (eg, angles of the camera and the subject, lighting, etc.) than other portions of the face area. The feature amount elements corresponding to the position may be additionally excluded. For example, the part around the nose in the face area is more affected by the shooting conditions than the other parts. Therefore, after the position corresponding to this portion is determined in advance as the setting position, the characteristic quantity elements corresponding to the pixels of the setting position can be excluded from the characteristic quantity candidate group C or C '.

Next, a facial feature extraction apparatus using this facial feature extraction method will be described.

6 and 7 are views showing the structure of the facial feature extraction apparatus according to an embodiment of the present invention.

As shown in FIG. 6, the apparatus for extracting facial features according to an embodiment of the present invention 1 includes a raw feature variable obtaining unit 11, a feature group candidate group generating unit 12, and an intermediate stage feature variable generating unit 13. And a final feature variable obtaining unit 14, and may further include a feature variable candidate group exclusion unit 15 and / or a feature variable candidate group updating unit 16.

The raw feature variable obtaining unit 11 obtains a raw feature variable set including raw face feature variable elements which are filter response data obtained by applying at least one face recognition filter to a face image.

The feature variable candidate group generation unit 12 generates a feature variable candidate group by extracting feature variable elements satisfying the first condition from the original feature variable set.

To this end, the feature amount candidate group generation unit 12 calculates a variance for each of the feature amount elements included in the set of raw feature quantities, and calculates a variance for the variance calculation module 121 and compares the calculated variance with a preset first threshold value. The comparison module 122 includes a generation module 123 for generating the feature amount candidate group C including the feature amount elements whose calculated variance is less than the first threshold.

The intermediate step feature variable generating section 13 generates an intermediate step feature set which includes a first set number of feature variable elements from the feature variable candidate group. To this end, as shown in FIG. 7, the intermediate step feature variable generating unit 13 selects one of the feature variable elements included in the feature variable candidate group 131 and the intermediate step feature variable for the selected feature variable. A correlation coefficient calculating module 132 that calculates a correlation with the feature amount elements already included in the set, a comparison module 133 that compares the calculated correlation with a preset second threshold value, and the calculated correlation is second And a set generation module 134 for including the corresponding feature variables in the intermediate step feature set when the threshold is smaller than the threshold. The selection module 131, the correlation calculation module 132, the comparison module 133, and the set generation module 134 repeatedly apply the intermediate step feature set including the set number of feature elements. You can perform the operation.

The final feature variable obtaining unit 14 obtains the final feature variable set to be used for face recognition from the intermediate step feature variable set including the second set number of feature variable elements. The final feature set can be obtained by applying the existing algorithm LDA to the intermediate feature set. Or obtain variances for each of the feature elements included in the intermediate set of feature quantities, arrange the feature variables based on the value of the variance, and then select the feature variable with the largest Based on the feature number elements of the set number K2, the final feature amount set may be obtained.

On the other hand, if the candidate group exclusion unit 15 determines that one feature variable is included in the intermediate feature set by the intermediate feature set generation unit 13, the candidate group exclusion unit 15 is located within a set radius based on the determined feature variable. The feature amount elements corresponding to the pixels are excluded from the feature variable candidate group.

In addition, the feature amount candidate group updating unit 16 calculates an in-person variance value of the feature amount elements included in the feature amount candidate group, extracts predetermined feature variable elements from the feature amount candidate group based on the calculated in-person dispersion value, A feature variable candidate group is generated. Thereafter, the intermediate step feature variable generating unit 13 may generate an intermediate step feature set including the first set number of feature variable elements from the feature variable candidate group newly generated by the feature variable candidate group updater 16.

Although the present invention has been described and illustrated in detail, it is clearly understood that the same is by way of illustration and example only and is not to be construed as limited to the embodiments set forth herein. And the present invention can be easily implemented by those skilled in the art from the description of the embodiments described above.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, It belongs to the scope of right.

Claims (12)

Obtaining a raw feature variable set including raw face feature variable elements which are filter response data obtained by applying at least one face recognition filter to a face image;
Generating a feature variable candidate group by extracting feature variable elements satisfying a set condition from the original feature variable set;
Generating an intermediate set of feature quantities comprising a first set number of feature quantities elements from the feature quantity candidate group; And
Obtaining a final feature set comprising a second set number of feature quantities elements to be used for face recognition from the intermediate feature set
Lt; / RTI &gt;
Generating the intermediate feature set includes calculating a correlation with a feature variable already included in the intermediate feature set for one feature variable selected from the feature variable candidate group, and calculating the correlation And if it is determined based on the figure that the one feature variable is included in the intermediate step feature set, deleting some feature variable related to the one feature variable from the feature variable candidate group. The method of claim 1, wherein
The generating of the intermediate feature set may include deleting, from the feature variable candidate group, a feature variable corresponding to a pixel of the face image located within a set radius with respect to the one feature variable. . The method of claim 1, wherein
Generating the feature variable candidate group,
Calculating a variance for each of the feature quantities elements included in the set of raw feature quantities;
Comparing the calculated variance with a first predetermined threshold; And
Generating the feature variable candidate group by extracting feature variable elements whose calculated variance is greater than a first threshold
Including, facial feature extraction method. The method of claim 1, wherein
Generating the intermediate step feature set
Selecting one of the feature elements included in the feature variable candidate group;
Calculating a correlation with the feature amount element already included in the intermediate step feature set for the selected feature variable;
Comparing the calculated correlation with a preset second threshold value;
Including the corresponding feature variable in the intermediate step feature set when the calculated correlation is less than a second threshold; And
Deleting some feature variables related to the feature variables in the intermediate feature set from the feature variable candidate group
Lt; / RTI &gt;
Wherein the selecting, calculating, comparing, including, and deleting from the feature variable candidate group are repeatedly performed until the first set number of feature quantities elements is included. Feature extraction method. The method of claim 1, wherein
Acquiring the final feature set
Calculating a variance for each feature amount element included in the intermediate feature set;
Arranging feature quantity elements based on the value of the variance; And
Extracting the second set number of feature variable elements based on the feature variable having the largest variance from the sorted feature variable elements to obtain a final feature set
Including, facial feature extraction method. The method according to any one of claims 1 to 5.
Updating the feature variable candidate group based on the in-person dispersion value for each feature variable element of the feature variable candidate group
Further comprising:
The generating of the intermediate feature set may include generating an intermediate feature set using the updated feature variable candidate group. The method of claim 6, wherein
The step of updating
Calculating an in-person variance value for the feature variables included in the feature variable candidate group;
Comparing the calculated in-person dispersion value with a third predetermined threshold value; And
Incorporating the corresponding feature variable into a new feature variable candidate group when the calculated intrapersonal variance is smaller than the third threshold.
Including, facial feature extraction method. The method according to any one of claims 1 to 5.
Generating the feature variable candidate group
And excluding feature amount elements corresponding to pixels at preset positions with respect to the face area included in the face image from the feature variable candidate group. A raw feature variable obtaining unit obtaining a raw feature variable set including raw face feature variable elements which are filter response data obtained by applying at least one face recognition filter to a face image;
A feature variable candidate group generation unit for generating a feature variable candidate group by extracting feature variable elements satisfying a first condition from the original feature variable set;
An intermediate step feature variable generating unit for generating an intermediate step feature variable set including a first set number of feature variable elements from the feature variable candidate group;
For one feature variable selected from the feature variable candidate group, a correlation with the feature variable already included in the intermediate step feature set is calculated, and based on the calculated correlation, the feature with the feature A feature amount candidate group exclusion unit which deletes, from the feature variable candidate group, a feature variable corresponding to a pixel of the face image located within a set radius with respect to the one feature variable when it is determined to be included in the intermediate feature set; ; And
A final feature variable obtaining unit for obtaining a final feature variable set including a second set number of feature variable elements to be used for face recognition from the intermediate feature set;
Lt; / RTI &gt;
And the intermediate step feature variable generating unit generates the intermediate step feature variable set based on the feature variable candidate group processed by the feature variable candidate group exclusion unit. The method of claim 9, wherein
The feature variable candidate group generation unit
A variance calculation module for calculating a variance for each of the feature quantities elements included in the set of primitive features;
A comparison module for comparing the calculated variance with a first preset threshold; And
A generation module for extracting feature quantity elements having the calculated variance greater than a first threshold to generate the feature quantity candidate group
A facial feature extraction apparatus comprising a. The method of claim 9, wherein
The intermediate step feature variable generating unit,
A selection module for selecting one of the feature variables included in the feature variable candidate group;
A correlation calculation module for calculating a correlation with the feature amount element already included in the intermediate step feature set for the selected feature variable;
A comparison module for comparing the calculated correlation with a preset second threshold value; And
A set generation module for including the corresponding feature variables in the intermediate step feature set when the calculated correlation is less than a second threshold
A facial feature extraction apparatus comprising a. The method of claim 9, wherein
Calculating an in-person variance value for the feature amount elements included in the feature variable candidate group, and updating the feature amount candidate group based only on the feature amount elements whose calculated in-person variance value is smaller than a preset third threshold value, And a feature feature candidate group updater.

Images