KR101449744B1 - Face detection device and method using region-based feature - Google Patents

Abstract

Disclosed are a face detection device and a method using a region-based feature. The present invention comprises: an input unit and a control unit which receives an image including a face, wherein the control unit may include: a window searching unit to detect face candidate regions based on the received image; a post-processing filter unit to filter by applying a post-processing filter on the detected face candidate regions; a score map unit to generate a score map based on the filtered candidate face regions; and a face region detection unit to detect face regions based on the generated score map.

Description

[0001] The present invention relates to a face detection apparatus and method using a region-based feature,

The present invention relates to a face detection apparatus and method, and more particularly, to a face detection apparatus and method using region-based features.

Face detection is an essential algorithm for various application fields such as face recognition, facial expression recognition, and drowsiness detection as well as importance of face detection algorithm itself by detecting the position and size of a face existing in the image. However, the accurate detection of a person's face in the image is problematic because it is sensitive to environmental factors such as various backgrounds of the image, pose of a person, facial expression and illumination change. Therefore, many researches on a method of accurately detecting human faces from various images have been continuing to date.

The methods of face detection, which have been studied so far, can be broadly classified into the following types.

First, it is a method based on prior knowledge information about feature components of face. This is a method of detecting a face by using the assumption that the human face has a certain distance and positional relationship between the face components and the body components such as eyes, nose, and mouth. However, such a method requires a different morphological model for the frontal and lateral images, which complicates the model of the face detector. For such a change in face angle, inclination, and obscuration, face detection is difficult.

Second, it uses template matching for face feature model. The template matching method is a method of generating a template for a face standard model by using face learning data and comparing the template with the template learned for a test image to detect an area having similar characteristics as a face. Such a method is not only susceptible to a change in the size of a face region existing in various kinds of images, distortion according to a direction of the eyes, but also has a problem that it is difficult to generate a template for a standard face model considering various races, facial expressions and postures. Also, since all size and position information of the face candidate region must be considered for template matching, a long time is required for detection.

Third, it is a method using feature based learning model. It extracts various kinds of features from training data, generates a discriminant model based on discriminant learning that can classify the features extracted from the face region and the features extracted from the non-face region, and detects the face region from the image Method. There are various machine learning methods such as SVM and LDA for distinguishing face and non-face areas from various features. This overcomes the constraints of the above-mentioned methods and shows good performance in face detection. However, many training data are needed to learn the distinction model, and it takes a lot of time to learn the model. Also, since the distinction model is influenced by the learning data, there is a problem that generalization performance deteriorates.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an apparatus and method for detecting a face using region-based features that reduce a face detector learning time using a region-based feature training method.

According to another aspect of the present invention, there is provided a face detection apparatus and method using a region-based feature in which a false detection rate is improved while a false detection rate is improved by combining a post-processing filter for a face candidate region.

It is another object of the present invention to provide a face detection apparatus and method which are not affected by environmental factors of high resolution, illumination, trinkets and facial expression change.

A face detection apparatus using an area-based feature includes an input unit and a control unit for receiving an image including a face, wherein the control unit includes a window search unit for detecting face candidate regions based on the received image, A score map unit for generating a score map based on the filtered face candidate region, a face region detection unit for detecting a face region based on the generated score map, . ≪ / RTI >

The window search unit may extract an area-based feature based on the image and detect the face candidate areas using a window search technique based on the extracted feature.

The window search unit may include a plurality of strong classifiers, and the strong classifier may include a plurality of weak classifiers.

The plurality of weak classifiers may be combined in a cascade structure.

The post-processing filter unit accumulates features of a local binary pattern (LBP) and histograms of orientated gradients (HOG) in a pyramid shape, and based on features piled up in the pyramid shape, a face vector can be determined using a support vector machine.

The face area detection unit can detect the face area when the score map of the face candidate area is equal to or larger than a preset threshold value.

A method for detecting a face using an area-based feature includes receiving an image including a face, detecting face candidate regions based on the received image, applying a post-processing filter to the detected face candidate regions, Generating a score map based on the filtered face candidate region, and detecting a face region based on the generated score map.

The apparatus and method for detecting a face using region-based features according to the present invention can reduce a face detector learning time using a region-based feature training method.

In addition, combining the post-processing filter with the face candidate region can improve the detection rate while lowering the false detection rate.

It can also be unaffected by high resolution, environmental factors of lighting, trinkets and facial changes.

1 is a block diagram illustrating components of a face detection apparatus according to an embodiment of the present invention.
FIG. 2 is an exemplary diagram illustrating an execution process of a face detection apparatus according to an embodiment of the present invention. Referring to FIG.
FIG. 3 is a diagram illustrating an example of a controller according to an embodiment of the present invention. Referring to FIG.
FIG. 4 is a diagram illustrating an exemplary method for an adaptive search step size scheme for a window search unit according to an embodiment of the present invention. Referring to FIG.
FIG. 5 is a diagram illustrating an example of a strong classifier according to an embodiment of the present invention. Referring to FIG.
FIG. 6 is a diagram illustrating an exemplary window searching unit using a strong classifier according to an embodiment of the present invention. Referring to FIG.
FIG. 7 is a diagram illustrating an example of a weak classifier according to an embodiment of the present invention. Referring to FIG.
FIG. 8 is a diagram illustrating an exemplary region-based feature used in a weak classifier according to an embodiment of the present invention.
9 is a graph illustrating an AUC score per stage according to an embodiment of the present invention.
10 is a graph illustrating a rejection rate per stage according to an embodiment of the present invention.
11 is a table illustrating a comparison of face detection performance of the Caltech Face DB according to an embodiment of the present invention.
FIG. 12 is a table illustrating a comparison of face detection performance of a CAS-PEAL Face DB according to an embodiment of the present invention.
13 is a flowchart illustrating a process of performing a face detection method according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to designate the same or similar components throughout the drawings. 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 obvious or understandable to those skilled in the art.

1 is a block diagram illustrating components of a face detection apparatus according to an embodiment of the present invention.

Referring to FIG. 1, the face detection apparatus 1 can detect a face of a received image using an area-based feature. The face detection apparatus 1 can detect a face candidate region existing in an image and detect a face region using region-based features. In addition, since the face detection apparatus 1 includes a post-processing filter, the accuracy of face detection can be enhanced. The face detection apparatus 1 may include an input unit 110, a control unit 120, an output unit 130, and a storage unit 140.

The input unit 110 can receive an image. The input unit 110 may receive an image including at least one face. The format of the image may be at least one of BMP, JPG, JPEG, JPE, JFIF, GIF, TIF, TIFF and PNG.

The control unit 120 may detect an area of a face included in the image received from the input unit 110. [ The control unit 120 may include a window search unit 122, a post-processing filter unit 124, a score map unit 126, and a face area detection unit 128.

The window searching unit 122 can detect face candidate regions based on the received image. The post-processing filter unit 124 may perform filtering by applying a post-processing filter to the detected face candidate regions. The score map section 126 may generate a score map based on the filtered face candidate regions. The face area detection unit 128 can detect the face area based on the generated score map.

The controller 120 may reduce the face detector learning time using a training method of region-based features. In addition, the control unit 120 can improve the positive detection rate while lowering the false detection rate by combining the post-processing filter for the face candidate region.

The output unit 130 may output the face region. The output unit 130 may display the face region of the face image included in the received image. The output unit 130 may include at least one of a liquid crystal, a monitor, a projector, a TV, and a printer.

The storage unit 140 may store an image received from the input unit 110. The storage unit 140 may store the face candidate region and the face region detected by the controller 120. Also, the storage unit 140 may store the generated score map.

FIG. 2 is an exemplary diagram illustrating an execution process of a face detection apparatus according to an embodiment of the present invention. Referring to FIG.

Referring to FIG. 2, the face detection apparatus 1 can detect a face region of an image including a plurality of faces. The face detection apparatus 1 can detect a face region using region-based features.

The face detection apparatus 1 can receive the image 210 through the input unit 110. [ The image 210 may include a plurality of faces. The face detection apparatus 1 can detect the face of the received image 210 through the control unit 120. [ The face detection apparatus 1 can output the image 220 for the detected face region through the output unit 130. [ The face detection apparatus 1 may use at least one of the face region as a circle, an ellipse, a triangle, a square, and a polygon.

FIG. 3 is a diagram illustrating an example of a controller according to an embodiment of the present invention. Referring to FIG.

Referring to FIG. 3, the controller 120 may detect a face region included in the received image using the region-based feature.

The window searching unit 122 may extract an area-based feature from the received image 310. FIG. The window searching unit 122 can detect the face candidate region 320 using the window searching method based on the extracted region-based feature. The face candidate region 320 may be a face region of the received image 310. The face candidate region 320 may also include at least one window.

)를 각 셀 영역마다 8차원으로 추출하여 32차원 특징 벡터를 생성할 수 있다. 여기서,

는 수평,

는 수직,

,

는 각각 대각 기울기 값을 의미한다. 얼굴 검출 장치(1)는 각 셀의 크기를 최소 12×12 픽셀로부터 가로, 세로 비율을 1:2, 1:3, 2:3 등으로 조절하여 총 358 종류의 영역 기반 특징 추출 윈도우를 이용할 수 있다.The window search unit 122 may use a region-based SURF (Speeded Up Robust Feature) feature. Conventionally, the SURF feature is generally used for feature point extraction and the like. However, the face detection apparatus 1 can detect the slope information of pixels existing in a 2x2 cell area within a 40x40 window area

) Can be extracted in eight dimensions for each cell region to generate a 32-dimensional feature vector. here,

The horizontal,

Vertical,

,

Respectively denote diagonal slope values. The face detection apparatus 1 can use a total of 358 area-based feature extraction windows by adjusting the size of each cell from at least 12 × 12 pixels to 1: 2, 1: 3, 2: have.

)가 주어지면, [수학식 1]을 이용하여 각 약분류기

를 얼굴 영역(

) 및 비얼굴 영역(

)을 구분할 수 있다.
The window search unit 122 may use logistic regression and gentle-boost in weak classifier learning. In addition, the window search unit 122 may use an Area Under the Curve (AUC) score for cascade structure learning for combining weak classifiers. The window search unit 122 includes N learning data (

), The following equation (1)

Face area (

) And the non-face area (

) Can be distinguished.

는 SURF 특징을 의미하고,

는 가중치 변수를 의미한다.here,

Means the SURF feature,

Means a weighting variable.

)를 이전 스테이지

의 분류기와 현재 스테이지

의

번째 라운드에 대한 분류기와의 결합을 통해 생성할 수 있다. 상기 강분류기는 캐스케이드 구조 학습이 이용될 수 있다.
The window search unit 122 uses the equation (2)

) To the previous stage

Classifier and current stage

of

Can be generated by combining with the classifier for the second round. The strong classifier can use cascade structure learning.

는 캐스케이드의 구조 학습을 위한 목적함수를 의미하고,

는 이전 스테이지를 의미한다. 또한

는 현재 스테이지를 의미하고,

는 현재 스테이지에서의 약분류기를 의미한다.here,

Is an objective function for structural learning of cascade,

Means the previous stage. Also

Means the current stage,

Means a weak classifier in the current stage.

번째 스테이지에서

의 값이 k를 증가시키며, 약분류기를 결합하더라도 더 이상 변화하지 않고 수렴하면 현재 스테이지에서의 분류기 결합을 종료하고, 다음 스테이지로 넘어갈 수 있다. 또한 윈도우 탐색부(122)는 스테이지의 수가 증가 및 스테이지 내의 라운드 수가 증가할수록 [수학식 2]의 값이 목표 학습 수렴치 값이 도달하며, 최종 목표값에 도달할 때까지 [수학식 2]를 반복할 수 있다.The window search unit 122

In the second stage

The value of k increases, and even if the weak classifier is combined, the classifier combination at the current stage is terminated and the process can be advanced to the next stage. Further, the window searching unit 122 calculates the following equation (2) until the target learning convergence value reaches the value of the formula (2) and the final target value is reached as the number of stages increases and the number of rounds in the stage increases It can be repeated.

The post-processing filter unit 124 may apply the post-processing filter 330 to the face candidate region 320 detected in the above manner

The post-processing filter unit 124 may detect the face candidate region from the image in the above-described manner and perform a post-processing filter 330 to reduce false detection before obtaining the final face detection result. The post-processing filter unit 124 can learn by reusing the false detection areas estimated from the face in the non-facial image and the learning facial image in order to properly process the patterns that are erroneously detected by the post-processing filter 330.

The post-processing filter unit 124 accumulates the features of the local binary pattern (LBP) and the histograms of the orientated gradients (HOG) in a pyramid shape, A final decision on whether or not the face region is made can be made using a support vector machine. The post-processing filter unit 124 may use a linear kernel to increase the output speed.

The score map unit 126 may perform a post-processing filter on the detected face candidate region and map the result to the score map 340 using the calculated response value. The score map 340 may be a map of the score for the probability that the face candidate region is a face region.

The face area detection unit 128 may determine the face area 350 if the mapped face candidate area is equal to or greater than a preset threshold value. The threshold may be set according to the type of the received image.

4 is a view illustrating an exemplary window for explaining a window search unit according to an embodiment of the present invention.

Referring to FIG. 4, the window searching unit 122 may use a method of rescaling a search window and an Adaptive Searching Stepsize Scheme to efficiently search for a face region.

4A is a view for explaining a method of redesigning a sub-window size in searching a face area.

The window searching unit 122 may use a method of redesigning the size of the search window by a multiple of a certain size. The window searching unit 122 can extract the face region feature only once at the beginning of the search using the redesign method and then use only the feature information of the corresponding coordinate value according to the coordinate value according to the changed size of the redesigned search window have. Accordingly, the window searching unit 122 can re-extract the face region feature without further work.

4 (b) is a view for explaining an adaptive search step size system method for performing face area search.

의 크기만큼 x축, y축으로 이동할 수 있다. 윈도우 탐색부(122)는 해당 탐색 윈도우에서 추출된 얼굴 영역 특징(Gradient magnitude)의 값이

보다 작은 경우, 다음번 탐색할 영역을 스탭 사이즈

대신

크기만큼 이동시키는 적응 탐색 스탭 사이즈 체계를 이용할 수 있다. 윈도우 탐색부(122)는 적응 탐색 스탭 사이즈 체계를 이용하여 얼굴 영역이 아닌 영역을 패싱(passing)하여 얼굴 영역 특징을 추출하는 불필요한 처리 과정을 생략할 수 있다. 여기서,

는 탐색 스탭 사이즈(Searching Stepsize)를 부스팅(boosting)하는 크기이고,

는 탐색하지 않고 패싱한 누적 횟수이다. When the window searching unit 122 detects the next area for an arbitrary search window size,

Axis and y-axis, respectively. The window search unit 122 searches for the value of the face region feature (gradient magnitude)

If the area to be searched next is smaller than the step size

instead

An adaptive search step size system can be used. The window searching unit 122 may skip the unnecessary processing of extracting the face region feature by passing an area other than the face region using the adaptive search step size system. here,

Is a size for boosting the Searching Stepsize,

Is the accumulated number of passes without searching.

을 이용하여 영상에서 배경과 같은 부분에서는 탐색 윈도우의 탐색 스탭 사이즈를 크게 함으로써, 불필한 탐색을 줄일 수 있다. 윈도우 탐색부(122)는 얼굴 영역에서는

값을 초기화하여 기본 탐색 스탭 사이즈

를 이용하여 얼굴 영역을 찾을 수 있다.
The window search unit 122

The unnecessary search can be reduced by enlarging the search step size of the search window in the part such as the background in the image. The window search unit 122 searches the face region

Initialize the value so that the default navigation step size

Can be used to find the face area.

FIG. 5 is a view illustrating an example of a strong classifier according to an embodiment of the present invention. FIG. 6 is an exemplary view for explaining a window search unit using a strong classifier according to an embodiment of the present invention. FIG. 7 is a diagram illustrating an example of a weak classifier according to an embodiment of the present invention. FIG. 8 is a diagram illustrating a region-based feature used in a weak classifier according to an embodiment of the present invention And Fig.

5 to 8, the window search unit 122 may include a plurality of strong classifiers, and the strong classifier may include a plurality of weak classifiers. Further, the weak classifier may be combined with a cascade structure.

As shown in FIG. 5, the window searching unit 122 may exclude windows that are estimated to be non-face regions among the search windows for face detection in the image received through the input unit 110. FIG. The window searching unit 122 extracts the non-face region candidate window 540 estimated as the non-face region among all the candidate windows 510 of the face candidate region 510 using the strong classifier 520, The region candidate window 530 can be extracted.

The window searching unit 122 may include a plurality of strong classifiers 522, 524, and 526. The window searching unit 122 can classify the strong classifiers according to the stage, and according to each stage, the strong classifier 520 can estimate the face area and the non-face area.

The first stage of the window search unit 122 includes a first classifier 522 and the second stage includes a second classifier 524 and the third stage includes a third classifier 526 can do. The number of the stages can be set in advance according to each situation.

The strong classifier 520 may be one in which the weak classifiers shown in FIG. 7 are made up of cascade structures. The features used in the weak classifier may be various shapes as shown in FIG.

The window searching unit 122 may extract a face region candidate window 540 estimated as a face region excluding the non-face region candidate window 540 estimated as a non-face region through the strong classifier 520 according to the stage, 530) can be extracted.

The window image 610 in the first stage shows that not only the face region but also a large portion of the non-face region is estimated through the window. However, through the window image 620 in the second stage and the window image 630 in the third stage, the window search unit 122 excludes the window that is estimated to be a non-face region as the stage increases, It can be confirmed that the estimated window is extracted.

Accordingly, the window search unit 122 can extract only the window estimated as the face region from the image received by the input unit 110 as the stage increases.

FIG. 9 is a graph illustrating an AUC score per stage according to an embodiment of the present invention, and FIG. 10 is a graph illustrating a rejection rate per stage according to an embodiment of the present invention .

Referring to FIGS. 9 and 10, the face detection apparatus 1 can analyze the performance of the face area detection accuracy in accordance with the increase in the number of stages.

FIG. 9 shows the value of the AUC score according to the stage, and the AUC is to evaluate how much accuracy is required to determine this blob variable to be measured. Referring to FIG. 9, the value of the formula (2) converges to a higher value as the number of stages increases and the number of rounds in the stage increases.

10 shows the rejection rate according to the cascade stage according to the stage. The face detection apparatus 1 shows that the rejection rate converges to 1 as the cascade stage increases.

 Therefore, the face detection apparatus 1 can classify the non-face candidate region well as the stage increases, and can detect the face candidate region well.

FIG. 11 is a table for explaining the comparison of face detection performance of Caltech Face DB according to an embodiment of the present invention. FIG. 12 is a diagram illustrating a face detection performance of a CAS-PEAL Face DB according to an embodiment of the present invention. Fig. 6 is a table showing a comparison for explaining the comparison. Fig.

11 and 12, the face detection apparatus 1 can be implemented for performance evaluation. The face detection device (1) can implement the algorithm using hardware of 8GB Ram of Core Quad CPU, Microsoft visual studio 2010, Matlab 2012 and OpenCV library 2.0 on Microsoft windows7 operating system. The face detection apparatus 1 can use Caltech and CAS-PEAL, and the face image can use 15600 images selected from Feret and Multi-PIE data. In addition, the face detection apparatus 1 can collect 20000 non-face images from Google or the like and use them.

The face detecting apparatus 1 can convert a face training image into a 40x40 size image, and a non-face image can be extracted in accordance with the feature window size ratio in the original size. The face detection apparatus 1 can set the target detection rate to 99.5% per stage. By using the value corresponding to the AUC score according to the positive detection in each stage, the erroneous detection rate condition 10 ^-6 .

The face detection apparatus 1 can take about 48 hours to learn cascade. Therefore, the face detecting apparatus 1 can dramatically shorten the time. In addition, the face detection apparatus 1 can take about 0.15 seconds per face for 320 × 240 resolution and 0.4 seconds for 640 × 480.

11 and 12 show an algorithm of the proposed algorithm, VJ (P. Viola and M. Jones, "Rapid object detection using a boosted cascade of simple features", CVPR, 2001) and J.LI (J. Wang, and Y. Zhang, "Face detection using surf cascade." ICCV Workshops, 2011.).

11 is a comparison of face detection performance for Caltech Face DB. Caltech Face DB consists of 450 pictures of 896 × 592 man and woman. Therefore, the performance comparison using Caltech Face DB can be aimed to find each face accurately in complex background.

The face detector (1) uses 308 weak classifiers composed of 6 stages, and it can be confirmed that similar detection performance is obtained at low erroneous detection compared with the VJ method composed of weak classifiers for about 2900 of 24 stages . In the case of J.LI, there is only a demonstration program, so that it is difficult to accurately compute the erroneous detection value, and only the result of the detection is used.

12 is a comparison of face detection performance for the CAS-PEAL Face DB. CAS-PEAL Face DB consists of various men and women. Therefore, the performance comparison using the CAS-PEAL Face DB can be aimed to detect the faces robustly against environmental factors such as image size, pose, wearing ornaments, facial expression, and illumination.

The face detection apparatus 1 has low false detection compared to the conventional methods (VJ, J.LI), and the detection performance is similar or superior to that of the conventional method.

13 is a flowchart illustrating a process of performing a face detection method according to an embodiment of the present invention.

Referring to FIG. 13, the face detection apparatus 1 can detect a face of a received image using an area-based feature. The face detection apparatus 1 can detect a face candidate region existing in an image and detect a face region using region-based features.

The face detection apparatus 1 receives an image (S100). The face detection apparatus 1 can receive an image including at least one face. The format of the image may be at least one of BMP, JPG, JPEG, JPE, JFIF, GIF, TIF, TIFF and PNG.

The face detection apparatus 1 detects a face candidate region using window search (S110). The face detection apparatus 1 can extract region-based features from the received image. The face detection apparatus 1 can detect the face candidate region 320 using the window search technique based on the extracted region-based feature. The face candidate region 320 may be a face region of the received image 310. The face candidate region 320 may also include at least one window.

The face detection apparatus 1 filters the face candidate region using a post-processing filter (S120). The face detection apparatus 1 may perform a post-processing filter 330 to reduce erroneous detection of the face candidate region before obtaining the final face detection result. In order to correctly process the patterns that are erroneously detected by the post-processing filter 330, the face detection apparatus 1 can learn by reusing the erroneous detection areas and the learning face images estimated from the non-face images.

The face detection apparatus 1 generates a score map for the face candidate region (S130). The face detection apparatus 1 can map the detected face candidate region to the score map 340. [ The score map 340 may be a map of the score for the probability that the face candidate region is a face region.

The face detection apparatus 1 detects the face area (S140). The face detection apparatus 1 can determine the face region 350 if the mapped face candidate region is equal to or greater than a preset threshold value. The threshold value can be set according to the type of the received image.

The present invention can also be embodied as computer-readable codes on a computer-readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer apparatus is stored. Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and may be implemented in the form of a carrier wave (for example, transmission via the Internet) . The computer-readable recording medium may also be distributed to networked computer devices so that computer readable code can be stored and executed in a distributed manner.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be taken by way of limitation in the embodiment in which said invention is directed. It will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the appended claims.

1: face detecting device 110: input part
120: control unit 122: window search unit
124: post-filter section 126: score map section
128: face area detection unit 130: output unit
140: storage unit 210: input image
220: output image 310: input image
320: face candidate region 330: post-processing filter
340: Score map 350: Face area
510: All Candidates Window 520: Strong Classifier
522: first class classifier 524: second class classifier
526: Class 3 classifier 530: Face region candidate window
540: Non-face region candidate window

Claims (7)

An input unit receiving an image including a face; And
Including a control section,
Wherein,
A window search unit for detecting face candidate regions based on the received image;
A post-processing filter unit for performing filtering by applying a post-processing filter to the detected face candidate regions;
A score map portion for generating a score map based on the filtered face candidate regions; And
And a face area detecting unit for detecting a face area based on the generated score map,
Wherein the window search unit comprises:
Extracting slope information of pixels existing in a cell region included in the window region based on the image to generate a feature vector, extracting region-based features based on the generated feature vector, And the face candidate regions are detected based on the detected face candidate regions.
The method according to claim 1,
Wherein the window search unit comprises:
Wherein the strong classifier comprises a plurality of strong classifiers, and the strong classifier includes a plurality of weak classifiers.
3. The method of claim 2,
Wherein the window search unit comprises:
Wherein the weak classifier is divided into a face area and a non-face area using the following equation: < EMI ID =
[Mathematical Expression]

here,

Means a face area,

Means a non-face area,

Means the SURF feature,

Means a weighting variable.
3. The method of claim 2,
Wherein the window search unit comprises:
Wherein the strong classifier is generated by combining a classifier of a previous stage and a classifier for a k-th round of a current stage using the following equation: < EMI ID =
[Mathematical Expression]

here,

Means a strong classifier,

Is an objective function for structural learning of cascade,

Means the previous stage,

Means the current stage,

Means a weak classifier in the current stage.
The method according to claim 1,
The post-
The features of the Local Binary Pattern (LBP) and the Histograms of Oriented Gradients (HOG) are built up in a pyramid form, and a support vector machine based on the pyramid- And the face region is determined using the region-based feature.
The method according to claim 1,
Wherein the face area detecting unit detects the face area when the score map of the face candidate area is equal to or larger than a predetermined threshold value.
Receiving an image including a face;
Detecting face candidate regions based on the received image;
Applying a post-processing filter to the detected face candidate regions to perform filtering;
Generating a score map based on the filtered face candidate regions; And
And detecting a face region based on the generated score map,
Wherein the step of detecting the face candidate region comprises:
Extracting slope information of pixels existing in a cell region included in the window region based on the image to generate a feature vector, extracting region-based features based on the generated feature vector, And the face candidate regions are detected based on the face candidate regions.

Images