KR101014506B1 - Method and apparatus for face features selection using fuzzy and boosting - Google Patents

Method and apparatus for face features selection using fuzzy and boosting Download PDF

Info

Publication number
KR101014506B1
KR101014506B1 KR1020090011812A KR20090011812A KR101014506B1 KR 101014506 B1 KR101014506 B1 KR 101014506B1 KR 1020090011812 A KR1020090011812 A KR 1020090011812A KR 20090011812 A KR20090011812 A KR 20090011812A KR 101014506 B1 KR101014506 B1 KR 101014506B1
Authority
KR
South Korea
Prior art keywords
feature
learning
vector
face image
partial
Prior art date
Application number
KR1020090011812A
Other languages
Korean (ko)
Other versions
KR20100092611A (en
Inventor
이종식
장원석
Original Assignee
인하대학교 산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 인하대학교 산학협력단 filed Critical 인하대학교 산학협력단
Priority to KR1020090011812A priority Critical patent/KR101014506B1/en
Publication of KR20100092611A publication Critical patent/KR20100092611A/en
Application granted granted Critical
Publication of KR101014506B1 publication Critical patent/KR101014506B1/en

Links

Images

Abstract

A method and apparatus for selecting facial features through a fuzzy and boosting technique are provided. Setting a number of partial vectors for feature selection of a face image, and learning feature vectors extracted from the face image through a boosting technique based on the number of partial vectors, )), Calculating an error of learning the feature based on the feature of the face image, and calculating the number of the partial vectors through a fuzzy technique based on the error of the learning of the feature. And further selecting features of the face image.
Figure R1020090011812
fuzzy, boosting, features vector

Description

METHOD AND APPARATUS FOR FACE FEATURES SELECTION USING FUZZY AND BOOSTING}

Embodiments of the present invention relate to a method and apparatus for selecting facial features.

The feature selection process in face recognition is a very important process for determining face recognition rate and learning speed. Face recognition is performed using face features extracted from a face image, and face features extracted from one face image may be expressed in a vector form called a feature vector.

Since the number of features extracted early is huge, the feature vector is very large and there are many unnecessary features that do not adequately express individual facial features. Of course, the problem of an increase in learning time occurs.

Therefore, a feature selection process that selects only certain features with discernment to distinguish individual faces is essential for face recognition. Recently, feature selection methods using boosting techniques have been recognized for their performance. It is actively developed.

In the facial feature selection method using the boosting technique, a feature vector is divided into individual feature components to be learned, and then a feature is selected by comparing the learning results.

In the feature selection method using the boosting technique and the feature vector segmentation technique, which is a more advanced method, the feature vector is divided into a predetermined number of partial vectors composed of a plurality of feature components, and the optimal vector is obtained by comparing the learning results. One is selected and one feature is selected by repeating the vector segmentation, learning, and selection process for the selected partial vector. Therefore, the feature selection method using the boosting technique and the feature vector segmentation technique selects more discriminating features than the feature selection method by further considering the complementary relations between the features as well as the discrimination of the individual features.

However, in the conventional feature selection method, the face recognition rate varies depending on the number of feature vectors to be divided, and the number of partial vectors to be selected after vector segmentation and learning is limited to one, so that more candidate features are considered for screening. This is a limiting factor in feature selection and face recognition performance.

An embodiment of the present invention uses fuzzy and boosting techniques to more effectively screen facial features having better facial discrimination from multiple facial features extracted from facial images, thereby improving face recognition rate and recognition performance. It is an object of the present invention to provide a method and apparatus for selecting facial features to ensure the stability of the.

In accordance with another aspect of the present invention, there is provided a method of selecting a facial feature, the method comprising: setting a number of partial vectors for selecting a feature of a face image; Selecting a feature of the face image by learning feature vectors extracted from the face image through a boosting technique based on the number of the partial vectors; Calculating an error on learning of the feature based on the feature of the face image; Calculating the number of the partial vectors through a fuzzy technique based on the error of the learning of the feature; And further selecting a feature of the face image.

In this case, the selecting of the feature of the face image by learning feature vectors extracted from the face image through the boosting technique may include setting weights for learning the feature vectors extracted from the plurality of face images. Making; Dividing the feature vector into a predetermined number of partial vectors; Learning features of the face image for each partial vector based on the weights for the learning; Selecting the partial vector based on the learned result; Merging the partial vectors; Selecting a feature of the face image; And storing the selected feature.

The facial feature selection device according to an embodiment of the present invention sets the number of partial vectors for feature selection of a face image, and is extracted from the face image through a boosting technique based on the number of the partial vectors. A feature selector configured to learn feature vectors to select a feature of the face image; A learning error calculator configured to calculate an error of learning the feature based on the feature of the face image; And a fuzzy inference unit that calculates the number of the partial vectors through a fuzzy technique based on an error of learning the feature.

The feature selector may include a weight adjuster configured to set a weight for learning the feature vector extracted from a plurality of face images; A vector segmentation and feature learning unit for dividing the feature vector into a predetermined number of partial vectors and learning a feature of the face image for each partial vector based on a weight for the learning; A partial vector selector which selects the partial vector and merges the partial vector or selects a feature of the face image based on the learned result; And a feature storage unit for storing the selected feature.

According to an embodiment of the present invention, by using a fuzzy and boosting technique to more effectively screen the facial features having better facial discrimination from a plurality of facial features extracted from the face image, improve the face recognition rate and recognize It is possible to provide a method and apparatus for selecting facial features to ensure stability of performance.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited or limited by the embodiments. Like reference numerals in the drawings denote like elements.

1 is a block diagram of a facial feature selection apparatus according to an embodiment of the present invention. An apparatus for selecting facial features according to an embodiment of the present invention will be described with reference to FIG. 1.

The facial feature selection apparatus according to the exemplary embodiment of the present invention includes a feature selector 120, a learning error calculator 130, and a fuzzy inference unit 140.

The feature selector 120 may include a weight adjuster 121, a vector segmentation and feature learner 122, a partial vector selector 123, and a feature storage 124.

The weight adjusting unit 121 assigns a learning weight to the feature vectors 110 extracted from the plurality of face images for learning. In this case, if there is no feature stored in the feature storage unit 124, the same learning weight is given to all the feature vectors, and if there is a stored feature, the feature vector learning error is large based on the feature vector learning error of each feature vector for the currently selected feature. Vectors can be given relatively large learning weights.

In this case, the facial feature selection apparatus may classify each feature vector using only the currently selected feature and calculate a feature vector learning error of each feature vector with respect to the currently selected feature.

The vector segmentation and feature learner 122 receives a feature vector from the weight adjuster 121 or receives a merged vector from the partial vector selector 123 and divides it into a predetermined number of partial vectors, and for each partial vector. Perform feature learning.

The partial vector selector 123 receives a learning result about the partial vector from the vector segmentation and feature learning unit 122, and calculates a partial vector learning error for each partial vector based on the learning result to reduce the partial vector learning error. Select a certain number of partial vectors in order. In this case, the partial vector learning error of each partial vector may be calculated by classifying the feature vectors having only features included in a specific partial vector, and the number of partial vector selections is obtained from the fuzzy inference unit 140.

In addition, when the dimension of the selected partial vector is greater than 1, the partial vector selector 123 merges the selected partial vectors into one vector, and then transfers the partial vector selector 123 to the vector segmentation and feature learning unit 122 to perform vector segmentation and feature learning. This rerun is performed, otherwise the partial vector having the minimum learning error among the selected partial vectors is selected and the features included in the vector are selected.

The feature storage unit 124 stores the selected feature with the feature weight when the feature is selected by the partial vector selector 123. In this case, as the partial vector learning error is smaller, larger feature weights may be assigned to selected features.

In addition, when a change occurs in the number of partial vector selections with reference to the number of partial vector selections inferred by the fuzzy inference unit 140, the feature storage unit 124 deletes all stored features and applies the changed number of partial vector selections. Choose features.

The learning error calculator 130 calculates a feature learning error using the features stored in the feature storage unit 124. In this case, the learning error calculator 130 may calculate the feature learning error by classifying the feature vectors 110 using the features stored in the feature storage unit 124.

The fuzzy inference unit 140 infers the optimal number of partial vector selections under a given condition by applying a fuzzy method based on the feature learning error calculated by the learning error calculator 130. In this case, the fuzzy inference unit 140 may receive a result of calculating the rate of change of the feature learning error according to the number of partial vector selections using the feature learning error, and display the minimum feature learning error by the fuzzy input. Fuzzy inference rules can be defined to infer the optimal number of partial vector selections.

2 is a flowchart illustrating a facial feature selection method according to an embodiment of the present invention. A facial feature selection method according to an embodiment of the present invention will be described with reference to FIG. 2.

The facial feature selection method according to an embodiment of the present invention may be performed by the facial feature selection device according to an embodiment of the present invention.

The facial feature selection apparatus sets the number of partial vectors to be selected at the time of feature selection (S210). In this case, the facial feature selection device may initially increase the number of partial vector selections from one at a time when a valid fuzzy output value is not obtained. Can be set.

Thereafter, the apparatus for selecting a facial feature selects a predetermined facial feature by learning feature vectors extracted from a plurality of face images for the learning through a boosting technique using the set partial vector selection number (S220). . In this case, the number of features selected by the face feature selection device may be limited to a minimum quantity for calculating the feature learning error and performing the fuzzy inference, and additional features required for face recognition may be additionally selected later. .

Thereafter, the facial feature selection apparatus calculates a feature learning error for the entire selected feature for fuzzy inference (S230). In this case, the feature learning error for the selected entire feature may be calculated by classifying the feature vectors used in the learning using the selected face features.

Thereafter, the facial feature selection apparatus infers the optimal number of partial vector selections under a given condition using a fuzzy technique based on the calculated feature learning error (S240). In this case, the facial feature selection device may calculate the degree of change in the feature learning error according to the number of partial vector selections using the feature learning error and use it as a fuzzy input, and show the minimum feature learning error by the fuzzy input. Fuzzy inference rules can be defined to infer the optimal number of partial vector selections.

Thereafter, the facial feature selection device compares the inferred partial vector selection number with the current partial vector selection number (S250).

When the face feature selection device compares the inferred partial vector selection number with the current partial vector selection number, the face feature selection device considers that the optimal number of partial vector selection is reached if the same is the same, and obtains the optimal values obtained through the above processes. The additional quantity of features required for face recognition is selected using the partial vector selection number of the at S260. In this case, the feature selection process of the facial feature selection device may be configured in the same manner as the feature selection process in step S220.

On the other hand, the facial feature selection device may compare the inferred partial vector selection number with the current partial vector selection number and, if different from each other, may perform steps S210 to S240 again.

3 is a flowchart illustrating a method of selecting a feature of a face using a boosting technique according to an embodiment of the present invention, and more particularly, a method of selecting a facial feature according to an embodiment of the present invention disclosed in FIG. 2. It is a flowchart for explaining the S220 step in more detail.

A method of selecting facial features using a boosting technique according to an embodiment of the present invention will be described with reference to FIG. 3.

The face feature selecting apparatus assigns a learning weight to each feature vector extracted from a plurality of face images for the learning (S310). In this case, when the feature is not selected, all the feature vectors are initially assigned the same learning weight, and when one or more features are selected, the feature vector having a large feature vector learning error is based on the feature vector learning error of each feature vector for the currently selected feature. More can be given a relatively large learning weight. In this case, the facial feature selection apparatus may classify each feature vector using the currently selected feature to calculate a feature vector learning error of each feature vector for the currently selected feature.

Thereafter, the facial feature selection apparatus divides the feature vector or the merged vector into a predetermined number of partial vectors (S320).

Thereafter, the facial feature selection apparatus learns the features included in the vector for each partial vector (S330). In this case, the facial feature selection apparatus may perform more learning as the features included in the feature vector having the large learning weight.

Subsequently, the face feature selection device calculates a partial vector learning error for each partial vector based on the learning result and selects a predetermined number of partial vectors in order of decreasing partial vector learning error (S340). In this case, the facial feature selection apparatus may classify the feature vectors using features included in a specific partial vector to calculate a partial vector learning error for each partial vector.

The facial feature selection apparatus determines whether the dimension of the selected partial vector is greater than 1 (S350), and if the dimension of the selected partial vector is not greater than 1, the facial feature selection device has a minimum partial vector learning error among the selected partial vectors. The partial vector is selected and the features included in the vector are selected (S370) and stored together with the feature weights (S380). In this case, the face feature selection apparatus may give a relatively large feature weight to the selected features as the partial vector learning error is smaller.

Meanwhile, if the dimension of the selected partial vector is greater than 1, the facial feature selection device merges the selected partial vectors into one vector (S360), and performs steps S310 to S340 again.

Thereafter, the facial feature selection apparatus compares the total number of stored features with a preset number (S390), and if the total number of stored features is less than the preset number, re-performs steps S310 to S380 to select additional features. Otherwise, the feature selection process ends.

That is, the face feature selection device selects a face feature by applying a fuzzy and boosting technique as described above, thereby more effectively selecting a useful face feature having a stronger discriminating power, improving the face recognition rate, It is possible to ensure the stability of the recognition performance by minimizing the effect.

As described above, the present invention has been described by specific embodiments such as specific components and the like. For those skilled in the art to which the present invention pertains, various modifications and variations are possible. Therefore, the spirit of the present invention should not be limited to the described embodiments, and all of the equivalents or equivalents of the claims as well as the claims to be described later will belong to the scope of the present invention. .

1 is a block diagram of a facial feature selection apparatus according to an embodiment of the present invention.

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

3 is a flowchart illustrating a method of selecting facial features using a boosting technique according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

110: feature vector 120: feature selector

121: weight adjustment unit 122: vector segmentation and feature learning unit

123: partial vector selection unit 124: feature storage unit

130: learning error calculation unit 140: fuzzy inference unit

Claims (4)

  1. Setting the number of partial vectors for feature selection of the face image;
    Selecting a feature of the face image by learning feature vectors extracted from the face image through a boosting technique based on the number of the partial vectors;
    Calculating a feature learning error based on the feature of the face image;
    Calculating the number of the partial vectors through a fuzzy technique based on the feature learning error; And
    Additionally selecting features of the face image
    Including,
    The feature learning error,
    It is calculated by classifying the feature vectors used in the learning using the features of the face image
    Facial feature selection method characterized in that.
  2. The method of claim 1,
    The step of selecting a feature of the face image by learning feature vectors extracted from the face image through the boosting technique,
    Setting a weight for learning the feature vector extracted from a plurality of face images;
    Dividing the feature vector into a predetermined number of partial vectors;
    Learning a feature of the face image for each partial vector based on the weight for learning;
    Selecting the partial vector based on the learned result;
    Merging the partial vectors;
    Selecting a feature of the face image; And
    Storing the selected feature
    Facial feature selection method comprising a.
  3. A number of partial vectors for selecting a feature of a face image is set, and based on the number of the partial vectors, the feature vector extracted from the face image is selected through a boosting technique to select a feature of the face image. Feature selection;
    A learning error calculator configured to calculate a feature learning error based on the feature of the face image; And
    Based on the feature learning error, fuzzy inference unit for calculating the number of the partial vector through a fuzzy technique
    Including,
    The feature learning error,
    It is calculated by classifying the feature vectors used in the learning using the features of the face image
    Facial feature selection device characterized in that.
  4. The method of claim 3,
    The feature selection unit,
    A weight adjusting unit for setting weights for learning the feature vectors extracted from the plurality of face images;
    A vector segmentation and feature learning unit for dividing the feature vector into a predetermined number of partial vectors and learning a feature of the face image for each partial vector based on a weight for the learning;
    A partial vector selector which selects the partial vector and merges the partial vector or selects a feature of the face image based on the learned result; And
    Feature storage unit for storing the selected feature
    Facial feature selection apparatus comprising a.
KR1020090011812A 2009-02-13 2009-02-13 Method and apparatus for face features selection using fuzzy and boosting KR101014506B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020090011812A KR101014506B1 (en) 2009-02-13 2009-02-13 Method and apparatus for face features selection using fuzzy and boosting

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020090011812A KR101014506B1 (en) 2009-02-13 2009-02-13 Method and apparatus for face features selection using fuzzy and boosting

Publications (2)

Publication Number Publication Date
KR20100092611A KR20100092611A (en) 2010-08-23
KR101014506B1 true KR101014506B1 (en) 2011-02-14

Family

ID=42757360

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020090011812A KR101014506B1 (en) 2009-02-13 2009-02-13 Method and apparatus for face features selection using fuzzy and boosting

Country Status (1)

Country Link
KR (1) KR101014506B1 (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20120056929A (en) 2010-09-20 2012-06-05 엘지전자 주식회사 Mobile terminal and operation control method thereof

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20040008792A (en) * 2002-07-19 2004-01-31 삼성전자주식회사 Method and system for face detecting using classifier learned decision boundary with face/near-face images
KR20070077973A (en) * 2006-01-25 2007-07-30 한국인식산업(주) Apparatus for recognizing a biological face and method therefor

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20040008792A (en) * 2002-07-19 2004-01-31 삼성전자주식회사 Method and system for face detecting using classifier learned decision boundary with face/near-face images
KR20070077973A (en) * 2006-01-25 2007-07-30 한국인식산업(주) Apparatus for recognizing a biological face and method therefor

Also Published As

Publication number Publication date
KR20100092611A (en) 2010-08-23

Similar Documents

Publication Publication Date Title
JP5025893B2 (en) Information processing apparatus and method, recording medium, and program
EP1583024A2 (en) Apparatus, method, computer program and computer readable storage medium for information prcessing
Yang et al. A constant-space belief propagation algorithm for stereo matching
JP4517633B2 (en) Object detection apparatus and method
CN103984959B (en) A kind of image classification method based on data and task-driven
US8401283B2 (en) Information processing apparatus, information processing method, and program
JP4744918B2 (en) Face detection method, apparatus, and program
CN101375608A (en) Intra-mode region-of-interest video object segmentation
JP4640825B2 (en) Specific orientation face determination method, apparatus, and program
Wu et al. Exploit the unknown gradually: One-shot video-based person re-identification by stepwise learning
JPWO2007020789A1 (en) Face image display device, face image display method, and face image display program
JP2007193740A (en) Face detection method and detection device, and program
JP2007327843A (en) Image processor and image processing program
JP4556891B2 (en) Information processing apparatus and method, recording medium, and program
JP2002056393A (en) Image processing method and device and storage medium
RU2648946C2 (en) Image object category recognition method and device
CN103544506B (en) A kind of image classification method and device based on convolutional neural networks
JP2000512046A (en) Method and apparatus for pattern identification based on statistical
Tsybakov et al. Square root penalty: adaptation to the margin in classification and in edge estimation
JP4708909B2 (en) Method, apparatus and program for detecting object of digital image
WO2016074247A1 (en) Face detection using machine learning
CN1828632A (en) Object detection apparatus, learning apparatus, object detection system, object detection method
US8693785B2 (en) Image matching devices and image matching methods thereof
WO2007055359A1 (en) Clustering system and image processing system having same
US8108324B2 (en) Forward feature selection for support vector machines

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant
FPAY Annual fee payment

Payment date: 20131217

Year of fee payment: 4

FPAY Annual fee payment

Payment date: 20141204

Year of fee payment: 5

FPAY Annual fee payment

Payment date: 20160113

Year of fee payment: 6

FPAY Annual fee payment

Payment date: 20161220

Year of fee payment: 7

LAPS Lapse due to unpaid annual fee